; NOTE: Assertions have been autogenerated by utils/update_test_checks.py ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii -passes='require,atomic-expand' %s | FileCheck -check-prefixes=ALL,CI %s ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes='require,atomic-expand' %s | FileCheck -check-prefixes=ALL,GFX9 %s ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -passes='require,atomic-expand' %s | FileCheck -check-prefixes=ALL,GFX908 %s ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -passes='require,atomic-expand' %s | FileCheck -check-prefixes=ALL,GFX90A %s ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -passes='require,atomic-expand' %s | FileCheck -check-prefixes=ALL,GFX942 %s ; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -passes='require,atomic-expand' %s | FileCheck -check-prefixes=ALL,GFX11 %s define void @test_atomicrmw_fadd_f32_global_no_use_unsafe(ptr addrspace(1) %ptr, float %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(1) %ptr, float %value syncscope("wavefront") monotonic ret void } define void @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe(ptr addrspace(7) %ptr, float %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_no_use_unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(7) %ptr, float %value syncscope("wavefront") monotonic ret void } define void @test_atomicrmw_fadd_f32_as999_no_use_unsafe(ptr addrspace(999) %ptr, float %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_as999_no_use_unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(999) %ptr, float %value syncscope("wavefront") monotonic ret void } define float @test_atomicrmw_fadd_f32_global_unsafe(ptr addrspace(1) %ptr, float %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f32_global_unsafe( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_global_unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret float [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_global_unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret float [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_global_unsafe( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]] ; GFX90A-NEXT: ret float [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_global_unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]] ; GFX942-NEXT: ret float [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_global_unsafe( ; GFX11-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]] ; GFX11-NEXT: ret float [[RES]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, float %value syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0 ret float %res } define float @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe(ptr addrspace(7) %ptr, float %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret float [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret float [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: ret float [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret float [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr_unsafe( ; GFX11-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(7) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: ret float [[RES]] ; %res = atomicrmw fadd ptr addrspace(7) %ptr, float %value syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0 ret float %res } define float @test_atomicrmw_fadd_f32_as999_unsafe(ptr addrspace(999) %ptr, float %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret float [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret float [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: ret float [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret float [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_as999_unsafe( ; GFX11-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(999) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: ret float [[RES]] ; %res = atomicrmw fadd ptr addrspace(999) %ptr, float %value syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0 ret float %res } define double @test_atomicrmw_fadd_f64_global_unsafe(ptr addrspace(1) %ptr, double %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f64_global_unsafe( ; CI-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_global_unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_global_unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_global_unsafe( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VALUE:%.*]] syncscope("wavefront") monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: ret double [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_global_unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VALUE:%.*]] syncscope("wavefront") monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret double [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_global_unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, double %value syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0 ret double %res } define float @test_atomicrmw_fadd_f32_flat_unsafe(ptr %ptr, float %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f32_flat_unsafe( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !noalias.addrspace [[META1:![0-9]+]], !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_flat_unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !noalias.addrspace [[META1:![0-9]+]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret float [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_flat_unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !noalias.addrspace [[META1:![0-9]+]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret float [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_flat_unsafe( ; GFX90A-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX90A-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX90A: atomicrmw.shared: ; GFX90A-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX90A-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP1]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !noalias.addrspace [[META1:![0-9]+]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX90A: atomicrmw.check.private: ; GFX90A-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX90A-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX90A: atomicrmw.private: ; GFX90A-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX90A-NEXT: [[LOADED_PRIVATE:%.*]] = load float, ptr addrspace(5) [[TMP3]], align 4 ; GFX90A-NEXT: [[NEW:%.*]] = fadd float [[LOADED_PRIVATE]], [[VALUE]] ; GFX90A-NEXT: store float [[NEW]], ptr addrspace(5) [[TMP3]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.global: ; GFX90A-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX90A-NEXT: [[TMP5:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], float [[VALUE]] syncscope("wavefront") monotonic, align 4, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.phi: ; GFX90A-NEXT: [[RES:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ] ; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret float [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_flat_unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !noalias.addrspace [[META1:![0-9]+]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret float [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_flat_unsafe( ; GFX11-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !noalias.addrspace [[META1:![0-9]+]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: ret float [[RES]] ; %res = atomicrmw fadd ptr %ptr, float %value syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0 ret float %res } define double @test_atomicrmw_fadd_f64_flat_unsafe__noprivate(ptr %ptr, double %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe__noprivate( ; CI-NEXT: [[TMP1:%.*]] = load double, ptr [[PTR:%.*]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe__noprivate( ; GFX9-NEXT: [[TMP1:%.*]] = load double, ptr [[PTR:%.*]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe__noprivate( ; GFX908-NEXT: [[TMP1:%.*]] = load double, ptr [[PTR:%.*]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe__noprivate( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], double [[VALUE:%.*]] syncscope("wavefront") monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: ret double [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe__noprivate( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], double [[VALUE:%.*]] syncscope("wavefront") monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret double [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe__noprivate( ; GFX11-NEXT: [[TMP1:%.*]] = load double, ptr [[PTR:%.*]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, double %value syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0 ret double %res } define double @test_atomicrmw_fadd_f64_flat_unsafe(ptr %ptr, double %value) #3 { ; CI-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe( ; CI-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; CI-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; CI: atomicrmw.private: ; CI-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; CI-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; CI-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; CI-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; CI: atomicrmw.global: ; CI-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; CI-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] syncscope("wavefront") monotonic monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; CI-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end1: ; CI-NEXT: br label [[ATOMICRMW_PHI]] ; CI: atomicrmw.phi: ; CI-NEXT: [[RES:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; CI-NEXT: br label [[ATOMICRMW_END:%.*]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[RES]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe( ; GFX9-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX9-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX9: atomicrmw.private: ; GFX9-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX9-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX9-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX9: atomicrmw.global: ; GFX9-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] syncscope("wavefront") monotonic monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX9-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end1: ; GFX9-NEXT: br label [[ATOMICRMW_PHI]] ; GFX9: atomicrmw.phi: ; GFX9-NEXT: [[RES:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX9-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[RES]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe( ; GFX908-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX908-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX908: atomicrmw.private: ; GFX908-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX908-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX908-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX908: atomicrmw.global: ; GFX908-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] syncscope("wavefront") monotonic monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX908-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end1: ; GFX908-NEXT: br label [[ATOMICRMW_PHI]] ; GFX908: atomicrmw.phi: ; GFX908-NEXT: [[RES:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX908-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[RES]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe( ; GFX90A-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX90A-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX90A: atomicrmw.shared: ; GFX90A-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX90A-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP3]], double [[VALUE:%.*]] syncscope("wavefront") monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX90A: atomicrmw.check.private: ; GFX90A-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX90A-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX90A: atomicrmw.private: ; GFX90A-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX90A-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX90A-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE]] ; GFX90A-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.global: ; GFX90A-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX90A-NEXT: [[TMP5:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], double [[VALUE]] syncscope("wavefront") monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.phi: ; GFX90A-NEXT: [[RES:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ] ; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret double [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe( ; GFX942-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX942-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX942: atomicrmw.shared: ; GFX942-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX942-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP3]], double [[VALUE:%.*]] syncscope("wavefront") monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX942: atomicrmw.check.private: ; GFX942-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX942-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX942: atomicrmw.private: ; GFX942-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX942-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX942-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE]] ; GFX942-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX942-NEXT: br label [[ATOMICRMW_PHI]] ; GFX942: atomicrmw.global: ; GFX942-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX942-NEXT: [[TMP5:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], double [[VALUE]] syncscope("wavefront") monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: br label [[ATOMICRMW_PHI]] ; GFX942: atomicrmw.phi: ; GFX942-NEXT: [[RES:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ] ; GFX942-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX942: atomicrmw.end: ; GFX942-NEXT: ret double [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_flat_unsafe( ; GFX11-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX11-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX11: atomicrmw.private: ; GFX11-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX11-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX11-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX11: atomicrmw.global: ; GFX11-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] syncscope("wavefront") monotonic monotonic, align 8, !noalias.addrspace [[META1]], !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX11-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end1: ; GFX11-NEXT: br label [[ATOMICRMW_PHI]] ; GFX11: atomicrmw.phi: ; GFX11-NEXT: [[RES:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX11-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[RES]] ; %res = atomicrmw fadd ptr %ptr, double %value syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0 ret double %res } define float @test_atomicrmw_fadd_f32_flat(ptr %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_flat( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, float %value seq_cst ret float %res } define float @test_atomicrmw_fadd_f32_global(ptr addrspace(1) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_global( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, float %value seq_cst ret float %res } define float @test_atomicrmw_fadd_f32_buffer_fat_ptr(ptr addrspace(7) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_buffer_fat_ptr( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(7) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(7) %ptr, float %value seq_cst ret float %res } define float @test_atomicrmw_fadd_f32_as999(ptr addrspace(999) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_as999( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(999) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(999) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(999) %ptr, float %value seq_cst ret float %res } define void @test_atomicrmw_fadd_f32_global_no_use_ieee(ptr addrspace(1) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_global_no_use_ieee( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(1) %ptr, float %value seq_cst ret void } define void @test_atomicrmw_fadd_f32_global_no_use_denorm_flush(ptr addrspace(1) %ptr, float %value) #4 { ; CI-LABEL: @test_atomicrmw_fadd_f32_global_no_use_denorm_flush( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_global_no_use_denorm_flush( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_global_no_use_denorm_flush( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_global_no_use_denorm_flush( ; GFX90A-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_global_no_use_denorm_flush( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_global_no_use_denorm_flush( ; GFX11-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(1) %ptr, float %value seq_cst, !amdgpu.no.fine.grained.memory !0 ret void } define float @test_atomicrmw_fadd_f32_local(ptr addrspace(3) %ptr, float %value) { ; CI-LABEL: @test_atomicrmw_fadd_f32_local( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_local( ; GFX9-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX9-NEXT: ret float [[RES]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_local( ; GFX908-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX908-NEXT: ret float [[RES]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_local( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX90A-NEXT: ret float [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_local( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX942-NEXT: ret float [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_local( ; GFX11-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX11-NEXT: ret float [[RES]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, float %value seq_cst ret float %res } define half @test_atomicrmw_fadd_f16_flat(ptr %ptr, half %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f16_flat( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4) ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64 ; ALL-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 ; ALL-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to half ; ALL-NEXT: [[NEW:%.*]] = fadd half [[TMP4]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast half [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to half ; ALL-NEXT: ret half [[TMP7]] ; %res = atomicrmw fadd ptr %ptr, half %value seq_cst ret half %res } define half @test_atomicrmw_fadd_f16_global(ptr addrspace(1) %ptr, half %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f16_global( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4) ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64 ; ALL-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 ; ALL-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to half ; ALL-NEXT: [[NEW:%.*]] = fadd half [[TMP4]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast half [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to half ; ALL-NEXT: ret half [[TMP7]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, half %value seq_cst ret half %res } define half @test_atomicrmw_fadd_f16_global_align4(ptr addrspace(1) %ptr, half %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f16_global_align4( ; ALL-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i16 ; ALL-NEXT: [[TMP2:%.*]] = bitcast i16 [[EXTRACTED]] to half ; ALL-NEXT: [[NEW:%.*]] = fadd half [[TMP2]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP3:%.*]] = bitcast half [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP3]] to i32 ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536 ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]] ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16 ; ALL-NEXT: [[TMP5:%.*]] = bitcast i16 [[EXTRACTED1]] to half ; ALL-NEXT: ret half [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, half %value seq_cst, align 4 ret half %res } define half @test_atomicrmw_fadd_f16_local(ptr addrspace(3) %ptr, half %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f16_local( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr addrspace(3) @llvm.ptrmask.p3.i32(ptr addrspace(3) [[PTR:%.*]], i32 -4) ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr addrspace(3) [[PTR]] to i32 ; ALL-NEXT: [[PTRLSB:%.*]] = and i32 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[TMP2]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(3) [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to half ; ALL-NEXT: [[NEW:%.*]] = fadd half [[TMP4]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast half [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to half ; ALL-NEXT: ret half [[TMP7]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, half %value seq_cst ret half %res } define double @test_atomicrmw_fadd_f64_flat(ptr %ptr, double %value) { ; CI-LABEL: @test_atomicrmw_fadd_f64_flat( ; CI-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; CI-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; CI: atomicrmw.private: ; CI-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; CI-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; CI-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; CI-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; CI: atomicrmw.global: ; CI-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; CI-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8, !noalias.addrspace [[META1]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; CI-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end1: ; CI-NEXT: br label [[ATOMICRMW_PHI]] ; CI: atomicrmw.phi: ; CI-NEXT: [[RES:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; CI-NEXT: br label [[ATOMICRMW_END:%.*]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[RES]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_flat( ; GFX9-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX9-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX9: atomicrmw.private: ; GFX9-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX9-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX9-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX9: atomicrmw.global: ; GFX9-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8, !noalias.addrspace [[META1]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX9-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end1: ; GFX9-NEXT: br label [[ATOMICRMW_PHI]] ; GFX9: atomicrmw.phi: ; GFX9-NEXT: [[RES:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX9-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[RES]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_flat( ; GFX908-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX908-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX908: atomicrmw.private: ; GFX908-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX908-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX908-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX908: atomicrmw.global: ; GFX908-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8, !noalias.addrspace [[META1]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX908-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end1: ; GFX908-NEXT: br label [[ATOMICRMW_PHI]] ; GFX908: atomicrmw.phi: ; GFX908-NEXT: [[RES:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX908-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[RES]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_flat( ; GFX90A-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX90A-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX90A: atomicrmw.shared: ; GFX90A-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX90A-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP1]], double [[VALUE:%.*]] seq_cst, align 8 ; GFX90A-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX90A: atomicrmw.check.private: ; GFX90A-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX90A-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX90A: atomicrmw.private: ; GFX90A-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX90A-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP3]], align 8 ; GFX90A-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE]] ; GFX90A-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP3]], align 8 ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.global: ; GFX90A-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX90A-NEXT: [[TMP5:%.*]] = load double, ptr addrspace(1) [[TMP4]], align 8 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi double [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP9:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX90A-NEXT: [[TMP6:%.*]] = bitcast double [[NEW2]] to i64 ; GFX90A-NEXT: [[TMP7:%.*]] = bitcast double [[LOADED]] to i64 ; GFX90A-NEXT: [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[TMP4]], i64 [[TMP7]], i64 [[TMP6]] seq_cst seq_cst, align 8 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP8]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP8]], 0 ; GFX90A-NEXT: [[TMP9]] = bitcast i64 [[NEWLOADED]] to double ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end1: ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.phi: ; GFX90A-NEXT: [[RES:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP9]], [[ATOMICRMW_END1]] ] ; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret double [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_flat( ; GFX942-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX942-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX942: atomicrmw.shared: ; GFX942-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX942-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP1]], double [[VALUE:%.*]] seq_cst, align 8 ; GFX942-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX942: atomicrmw.check.private: ; GFX942-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX942-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX942: atomicrmw.private: ; GFX942-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX942-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP3]], align 8 ; GFX942-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE]] ; GFX942-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP3]], align 8 ; GFX942-NEXT: br label [[ATOMICRMW_PHI]] ; GFX942: atomicrmw.global: ; GFX942-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX942-NEXT: [[TMP5:%.*]] = load double, ptr addrspace(1) [[TMP4]], align 8 ; GFX942-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX942: atomicrmw.start: ; GFX942-NEXT: [[LOADED:%.*]] = phi double [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP9:%.*]], [[ATOMICRMW_START]] ] ; GFX942-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX942-NEXT: [[TMP6:%.*]] = bitcast double [[NEW2]] to i64 ; GFX942-NEXT: [[TMP7:%.*]] = bitcast double [[LOADED]] to i64 ; GFX942-NEXT: [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[TMP4]], i64 [[TMP7]], i64 [[TMP6]] seq_cst seq_cst, align 8 ; GFX942-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP8]], 1 ; GFX942-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP8]], 0 ; GFX942-NEXT: [[TMP9]] = bitcast i64 [[NEWLOADED]] to double ; GFX942-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX942: atomicrmw.end1: ; GFX942-NEXT: br label [[ATOMICRMW_PHI]] ; GFX942: atomicrmw.phi: ; GFX942-NEXT: [[RES:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP9]], [[ATOMICRMW_END1]] ] ; GFX942-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX942: atomicrmw.end: ; GFX942-NEXT: ret double [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_flat( ; GFX11-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX11-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX11: atomicrmw.private: ; GFX11-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX11-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX11-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX11: atomicrmw.global: ; GFX11-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8, !noalias.addrspace [[META1]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX11-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end1: ; GFX11-NEXT: br label [[ATOMICRMW_PHI]] ; GFX11: atomicrmw.phi: ; GFX11-NEXT: [[RES:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX11-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[RES]] ; %res = atomicrmw fadd ptr %ptr, double %value seq_cst ret double %res } define double @test_atomicrmw_fadd_f64_flat__noprivate(ptr %ptr, double %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f64_flat__noprivate( ; ALL-NEXT: [[TMP1:%.*]] = load double, ptr [[PTR:%.*]], align 8 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; ALL-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP3]], i64 [[TMP2]] seq_cst seq_cst, align 8, !noalias.addrspace [[META1:![0-9]+]] ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret double [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, double %value seq_cst, !noalias.addrspace !1 ret double %res } define double @test_atomicrmw_fadd_f64_global(ptr addrspace(1) %ptr, double %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f64_global( ; ALL-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; ALL-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] seq_cst seq_cst, align 8 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret double [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, double %value seq_cst ret double %res } define double @test_atomicrmw_fadd_f64_local(ptr addrspace(3) %ptr, double %value) { ; CI-LABEL: @test_atomicrmw_fadd_f64_local( ; CI-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] seq_cst seq_cst, align 8 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_local( ; GFX9-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] seq_cst seq_cst, align 8 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_local( ; GFX908-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] seq_cst seq_cst, align 8 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_local( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] seq_cst, align 8 ; GFX90A-NEXT: ret double [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_local( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] seq_cst, align 8 ; GFX942-NEXT: ret double [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_local( ; GFX11-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] seq_cst seq_cst, align 8 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, double %value seq_cst ret double %res } define float @test_atomicrmw_fadd_f32_global_agent(ptr addrspace(1) %ptr, float %value) { ; CI-LABEL: @test_atomicrmw_fadd_f32_global_agent( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") monotonic monotonic, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_global_agent( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") monotonic monotonic, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret float [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_global_agent( ; GFX908-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") monotonic monotonic, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret float [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_global_agent( ; GFX90A-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") monotonic monotonic, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret float [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_global_agent( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("agent") monotonic, align 4 ; GFX942-NEXT: ret float [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_global_agent( ; GFX11-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") monotonic monotonic, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret float [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, float %value syncscope("agent") monotonic ret float %res } define float @test_atomicrmw_fadd_f32_global_one_as(ptr addrspace(1) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_global_one_as( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("one-as") monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, float %value syncscope("one-as") monotonic ret float %res } define void @test_atomicrmw_fadd_f32_global_no_use_unsafe_strictfp(ptr addrspace(1) %ptr, float %value) #1 { ; CI-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe_strictfp( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[LOADED]], float [[VALUE:%.*]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8:[0-9]+]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe_strictfp( ; GFX9-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[LOADED]], float [[VALUE:%.*]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8:[0-9]+]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe_strictfp( ; GFX908-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe_strictfp( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe_strictfp( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_global_no_use_unsafe_strictfp( ; GFX11-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(1) %ptr, float %value syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0 ret void } define double @test_atomicrmw_fadd_f64_global_unsafe_strictfp(ptr addrspace(1) %ptr, double %value) #1 { ; CI-LABEL: @test_atomicrmw_fadd_f64_global_unsafe_strictfp( ; CI-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[LOADED]], double [[VALUE:%.*]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]] ; CI-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_global_unsafe_strictfp( ; GFX9-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[LOADED]], double [[VALUE:%.*]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_global_unsafe_strictfp( ; GFX908-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[LOADED]], double [[VALUE:%.*]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8:[0-9]+]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_global_unsafe_strictfp( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VALUE:%.*]] syncscope("wavefront") monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: ret double [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_global_unsafe_strictfp( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VALUE:%.*]] syncscope("wavefront") monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret double [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_global_unsafe_strictfp( ; GFX11-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = call double @llvm.experimental.constrained.fadd.f64(double [[LOADED]], double [[VALUE:%.*]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8:[0-9]+]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] syncscope("wavefront") monotonic monotonic, align 8, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, double %value syncscope("wavefront") monotonic, !amdgpu.no.fine.grained.memory !0 ret double %res } define float @test_atomicrmw_fadd_f32_local_strictfp(ptr addrspace(3) %ptr, float %value) #2 { ; CI-LABEL: @test_atomicrmw_fadd_f32_local_strictfp( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[LOADED]], float [[VALUE:%.*]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_local_strictfp( ; GFX9-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX9-NEXT: ret float [[RES]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_local_strictfp( ; GFX908-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX908-NEXT: ret float [[RES]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_local_strictfp( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX90A-NEXT: ret float [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_local_strictfp( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX942-NEXT: ret float [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_local_strictfp( ; GFX11-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] seq_cst, align 4 ; GFX11-NEXT: ret float [[RES]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, float %value seq_cst ret float %res } define bfloat @test_atomicrmw_fadd_bf16_local(ptr addrspace(3) %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_local( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr addrspace(3) @llvm.ptrmask.p3.i32(ptr addrspace(3) [[PTR:%.*]], i32 -4) ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr addrspace(3) [[PTR]] to i32 ; ALL-NEXT: [[PTRLSB:%.*]] = and i32 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[TMP2]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(3) [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP4]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to bfloat ; ALL-NEXT: ret bfloat [[TMP7]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, bfloat %value monotonic ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_local_align4(ptr addrspace(3) %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_local_align4( ; ALL-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(3) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i16 ; ALL-NEXT: [[TMP2:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP2]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP3:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP3]] to i32 ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536 ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]] ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16 ; ALL-NEXT: [[TMP5:%.*]] = bitcast i16 [[EXTRACTED1]] to bfloat ; ALL-NEXT: ret bfloat [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, bfloat %value monotonic, align 4 ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_global_agent(ptr addrspace(1) %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_global_agent( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4) ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64 ; ALL-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 ; ALL-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP4]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to bfloat ; ALL-NEXT: ret bfloat [[TMP7]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, bfloat %value syncscope("agent") monotonic ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_global_agent_align4(ptr addrspace(1) %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_global_agent_align4( ; ALL-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i16 ; ALL-NEXT: [[TMP2:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP2]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP3:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP3]] to i32 ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536 ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]] ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16 ; ALL-NEXT: [[TMP5:%.*]] = bitcast i16 [[EXTRACTED1]] to bfloat ; ALL-NEXT: ret bfloat [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, bfloat %value syncscope("agent") monotonic, align 4 ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_global_system(ptr addrspace(1) %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_global_system( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4) ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64 ; ALL-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 ; ALL-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP4]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to bfloat ; ALL-NEXT: ret bfloat [[TMP7]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, bfloat %value monotonic ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_global_system_align4(ptr addrspace(1) %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_global_system_align4( ; ALL-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i16 ; ALL-NEXT: [[TMP2:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP2]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP3:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP3]] to i32 ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536 ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]] ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16 ; ALL-NEXT: [[TMP5:%.*]] = bitcast i16 [[EXTRACTED1]] to bfloat ; ALL-NEXT: ret bfloat [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, bfloat %value monotonic, align 4 ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_local_strictfp(ptr addrspace(3) %ptr, bfloat %value) #2 { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_local_strictfp( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr addrspace(3) @llvm.ptrmask.p3.i32(ptr addrspace(3) [[PTR:%.*]], i32 -4) #[[ATTR8:[0-9]+]] ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr addrspace(3) [[PTR]] to i32 ; ALL-NEXT: [[PTRLSB:%.*]] = and i32 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[TMP2]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(3) [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = call bfloat @llvm.experimental.constrained.fadd.bf16(bfloat [[TMP4]], bfloat [[VALUE:%.*]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to bfloat ; ALL-NEXT: ret bfloat [[TMP7]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, bfloat %value monotonic ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_flat_agent(ptr %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_flat_agent( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4) ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64 ; ALL-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 ; ALL-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP4]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to bfloat ; ALL-NEXT: ret bfloat [[TMP7]] ; %res = atomicrmw fadd ptr %ptr, bfloat %value syncscope("agent") monotonic ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_flat_agent_align4(ptr %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_flat_agent_align4( ; ALL-NEXT: [[TMP1:%.*]] = load i32, ptr [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i16 ; ALL-NEXT: [[TMP2:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP2]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP3:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP3]] to i32 ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536 ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]] ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16 ; ALL-NEXT: [[TMP5:%.*]] = bitcast i16 [[EXTRACTED1]] to bfloat ; ALL-NEXT: ret bfloat [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, bfloat %value syncscope("agent") monotonic, align 4 ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_flat_system(ptr %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_flat_system( ; ALL-NEXT: [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4) ; ALL-NEXT: [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64 ; ALL-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 ; ALL-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 ; ALL-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 ; ALL-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]] ; ALL-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 ; ALL-NEXT: [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 ; ALL-NEXT: [[TMP4:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP4]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP5:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP5]] to i32 ; ALL-NEXT: [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]] ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]] ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]] ; ALL-NEXT: [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]] ; ALL-NEXT: [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16 ; ALL-NEXT: [[TMP7:%.*]] = bitcast i16 [[EXTRACTED3]] to bfloat ; ALL-NEXT: ret bfloat [[TMP7]] ; %res = atomicrmw fadd ptr %ptr, bfloat %value monotonic ret bfloat %res } define bfloat @test_atomicrmw_fadd_bf16_flat_system_align4(ptr %ptr, bfloat %value) { ; ALL-LABEL: @test_atomicrmw_fadd_bf16_flat_system_align4( ; ALL-NEXT: [[TMP1:%.*]] = load i32, ptr [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i16 ; ALL-NEXT: [[TMP2:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat ; ALL-NEXT: [[NEW:%.*]] = fadd bfloat [[TMP2]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP3:%.*]] = bitcast bfloat [[NEW]] to i16 ; ALL-NEXT: [[EXTENDED:%.*]] = zext i16 [[TMP3]] to i32 ; ALL-NEXT: [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536 ; ALL-NEXT: [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]] ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16 ; ALL-NEXT: [[TMP5:%.*]] = bitcast i16 [[EXTRACTED1]] to bfloat ; ALL-NEXT: ret bfloat [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, bfloat %value monotonic, align 4 ret bfloat %res } define void @test_atomicrmw_fadd_f32_global_system_noret(ptr addrspace(1) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_global_system_noret( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic ret void } define float @test_atomicrmw_fadd_f32_global_system_ret(ptr addrspace(1) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_global_system_ret( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic ret float %ret } define void @test_atomicrmw_fadd_f32_global_system_noret__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_global_system_noret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret void } define float @test_atomicrmw_fadd_f32_global_system_ret__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_global_system_ret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret float %ret } define void @test_atomicrmw_fadd_f32_daz_global_system_noret(ptr addrspace(1) %ptr, float %value) #3 { ; ALL-LABEL: @test_atomicrmw_fadd_f32_daz_global_system_noret( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic ret void } define float @test_atomicrmw_fadd_f32_daz_global_system_ret(ptr addrspace(1) %ptr, float %value) #3 { ; ALL-LABEL: @test_atomicrmw_fadd_f32_daz_global_system_ret( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic ret float %ret } define void @test_atomicrmw_fadd_f32_daz_global_system_noret__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %value) #3 { ; ALL-LABEL: @test_atomicrmw_fadd_f32_daz_global_system_noret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret void } define float @test_atomicrmw_fadd_f32_daz_global_system_ret__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %value) #3 { ; ALL-LABEL: @test_atomicrmw_fadd_f32_daz_global_system_ret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret float %ret } define void @test_atomicrmw_fadd_f32_dyndenorm_global_system_noret__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %value) #4 { ; ALL-LABEL: @test_atomicrmw_fadd_f32_dyndenorm_global_system_noret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret void } define float @test_atomicrmw_fadd_f32_dyndenorm_global_system_ret__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, float %value) #4 { ; ALL-LABEL: @test_atomicrmw_fadd_f32_dyndenorm_global_system_ret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(1) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(1) %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret float %ret } define void @test_atomicrmw_fadd_f32_local_noret(ptr addrspace(3) %ptr, float %value) { ; CI-LABEL: @test_atomicrmw_fadd_f32_local_noret( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_local_noret( ; GFX9-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_local_noret( ; GFX908-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_local_noret( ; GFX90A-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_local_noret( ; GFX942-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_local_noret( ; GFX11-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX11-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(3) %ptr, float %value monotonic ret void } define float @test_atomicrmw_fadd_f32_local_ret(ptr addrspace(3) %ptr, float %value) { ; CI-LABEL: @test_atomicrmw_fadd_f32_local_ret( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_local_ret( ; GFX9-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX9-NEXT: ret float [[RET]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_local_ret( ; GFX908-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX908-NEXT: ret float [[RET]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_local_ret( ; GFX90A-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX90A-NEXT: ret float [[RET]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_local_ret( ; GFX942-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX942-NEXT: ret float [[RET]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_local_ret( ; GFX11-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4 ; GFX11-NEXT: ret float [[RET]] ; %ret = atomicrmw fadd ptr addrspace(3) %ptr, float %value monotonic ret float %ret } define void @test_atomicrmw_fadd_f32_local_noret__amdgpu_ignore_denormal_mode(ptr addrspace(3) %ptr, float %value) { ; CI-LABEL: @test_atomicrmw_fadd_f32_local_noret__amdgpu_ignore_denormal_mode( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_local_noret__amdgpu_ignore_denormal_mode( ; GFX9-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_local_noret__amdgpu_ignore_denormal_mode( ; GFX908-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_local_noret__amdgpu_ignore_denormal_mode( ; GFX90A-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_local_noret__amdgpu_ignore_denormal_mode( ; GFX942-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_local_noret__amdgpu_ignore_denormal_mode( ; GFX11-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX11-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(3) %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret void } define float @test_atomicrmw_fadd_f32_local_ret__amdgpu_ignore_denormal_mode(ptr addrspace(3) %ptr, float %value) { ; CI-LABEL: @test_atomicrmw_fadd_f32_local_ret__amdgpu_ignore_denormal_mode( ; CI-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret float [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f32_local_ret__amdgpu_ignore_denormal_mode( ; GFX9-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX9-NEXT: ret float [[RET]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f32_local_ret__amdgpu_ignore_denormal_mode( ; GFX908-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX908-NEXT: ret float [[RET]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f32_local_ret__amdgpu_ignore_denormal_mode( ; GFX90A-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX90A-NEXT: ret float [[RET]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f32_local_ret__amdgpu_ignore_denormal_mode( ; GFX942-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX942-NEXT: ret float [[RET]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f32_local_ret__amdgpu_ignore_denormal_mode( ; GFX11-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], float [[VALUE:%.*]] monotonic, align 4, !amdgpu.ignore.denormal.mode [[META0]] ; GFX11-NEXT: ret float [[RET]] ; %ret = atomicrmw fadd ptr addrspace(3) %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret float %ret } define void @test_atomicrmw_fadd_f64_dyndenorm_global_system_noret(ptr addrspace(1) %ptr, double %value) #5 { ; ALL-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_global_system_noret( ; ALL-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; ALL-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(1) %ptr, double %value monotonic ret void } define double @test_atomicrmw_fadd_f64_dyndenorm_global_system_ret(ptr addrspace(1) %ptr, double %value) #5 { ; ALL-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_global_system_ret( ; ALL-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; ALL-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret double [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(1) %ptr, double %value monotonic ret double %ret } define void @test_atomicrmw_fadd_f64_dyndenorm_global_system_noret__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, double %value) #5 { ; ALL-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_global_system_noret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; ALL-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(1) %ptr, double %value monotonic, !amdgpu.ignore.denormal.mode !0 ret void } define double @test_atomicrmw_fadd_f64_dyndenorm_global_system_ret__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, double %value) #5 { ; ALL-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_global_system_ret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(1) [[PTR:%.*]], align 8 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; ALL-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret double [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(1) %ptr, double %value monotonic, !amdgpu.ignore.denormal.mode !0 ret double %ret } define void @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret(ptr addrspace(3) %ptr, double %value) #5 { ; CI-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret( ; CI-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret( ; GFX9-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret( ; GFX908-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret( ; GFX90A-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] monotonic, align 8 ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret( ; GFX942-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] monotonic, align 8 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret( ; GFX11-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(3) %ptr, double %value monotonic ret void } define double @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret(ptr addrspace(3) %ptr, double %value) #5 { ; CI-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret( ; CI-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret( ; GFX9-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret( ; GFX908-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret( ; GFX90A-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] monotonic, align 8 ; GFX90A-NEXT: ret double [[RET]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret( ; GFX942-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] monotonic, align 8 ; GFX942-NEXT: ret double [[RET]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret( ; GFX11-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(3) %ptr, double %value monotonic ret double %ret } define void @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret__amdgpu_ignore_denormal_mode(ptr addrspace(3) %ptr, double %value) #5 { ; CI-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret__amdgpu_ignore_denormal_mode( ; CI-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret__amdgpu_ignore_denormal_mode( ; GFX9-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret__amdgpu_ignore_denormal_mode( ; GFX908-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret__amdgpu_ignore_denormal_mode( ; GFX90A-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] monotonic, align 8, !amdgpu.ignore.denormal.mode [[META0]] ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret__amdgpu_ignore_denormal_mode( ; GFX942-NEXT: [[UNUSED:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] monotonic, align 8, !amdgpu.ignore.denormal.mode [[META0]] ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_noret__amdgpu_ignore_denormal_mode( ; GFX11-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(3) %ptr, double %value monotonic, !amdgpu.ignore.denormal.mode !0 ret void } define double @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret__amdgpu_ignore_denormal_mode(ptr addrspace(3) %ptr, double %value) #5 { ; CI-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret__amdgpu_ignore_denormal_mode( ; CI-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret__amdgpu_ignore_denormal_mode( ; GFX9-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret__amdgpu_ignore_denormal_mode( ; GFX908-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret__amdgpu_ignore_denormal_mode( ; GFX90A-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] monotonic, align 8, !amdgpu.ignore.denormal.mode [[META0]] ; GFX90A-NEXT: ret double [[RET]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret__amdgpu_ignore_denormal_mode( ; GFX942-NEXT: [[RET:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], double [[VALUE:%.*]] monotonic, align 8, !amdgpu.ignore.denormal.mode [[META0]] ; GFX942-NEXT: ret double [[RET]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_local_system_ret__amdgpu_ignore_denormal_mode( ; GFX11-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(3) [[PTR:%.*]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(3) %ptr, double %value monotonic, !amdgpu.ignore.denormal.mode !0 ret double %ret } define void @test_atomicrmw_fadd_f32_flat_system_noret__amdgpu_ignore_denormal_mode(ptr %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_flat_system_noret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret void } define float @test_atomicrmw_fadd_f32_flat_system_ret__amdgpu_ignore_denormal_mode(ptr %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_flat_system_ret__amdgpu_ignore_denormal_mode( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %ret = atomicrmw fadd ptr %ptr, float %value monotonic, !amdgpu.ignore.denormal.mode !0 ret float %ret } define void @test_atomicrmw_fadd_f64_dyndenorm_flat_system_noret__amdgpu_ignore_denormal_mode(ptr %ptr, double %value) #5 { ; CI-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_noret__amdgpu_ignore_denormal_mode( ; CI-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; CI-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; CI: atomicrmw.private: ; CI-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; CI-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; CI-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; CI-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; CI: atomicrmw.global: ; CI-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; CI-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] monotonic monotonic, align 8, !noalias.addrspace [[META1]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; CI-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end1: ; CI-NEXT: br label [[ATOMICRMW_PHI]] ; CI: atomicrmw.phi: ; CI-NEXT: br label [[ATOMICRMW_END:%.*]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_noret__amdgpu_ignore_denormal_mode( ; GFX9-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX9-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX9: atomicrmw.private: ; GFX9-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX9-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX9-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX9: atomicrmw.global: ; GFX9-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] monotonic monotonic, align 8, !noalias.addrspace [[META1]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX9-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end1: ; GFX9-NEXT: br label [[ATOMICRMW_PHI]] ; GFX9: atomicrmw.phi: ; GFX9-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_noret__amdgpu_ignore_denormal_mode( ; GFX908-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX908-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX908: atomicrmw.private: ; GFX908-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX908-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX908-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX908: atomicrmw.global: ; GFX908-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] monotonic monotonic, align 8, !noalias.addrspace [[META1]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX908-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end1: ; GFX908-NEXT: br label [[ATOMICRMW_PHI]] ; GFX908: atomicrmw.phi: ; GFX908-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_noret__amdgpu_ignore_denormal_mode( ; GFX90A-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX90A-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX90A: atomicrmw.shared: ; GFX90A-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX90A-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP1]], double [[VALUE:%.*]] monotonic, align 8, !amdgpu.ignore.denormal.mode [[META0]] ; GFX90A-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX90A: atomicrmw.check.private: ; GFX90A-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX90A-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX90A: atomicrmw.private: ; GFX90A-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX90A-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP3]], align 8 ; GFX90A-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE]] ; GFX90A-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP3]], align 8 ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.global: ; GFX90A-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX90A-NEXT: [[TMP5:%.*]] = load double, ptr addrspace(1) [[TMP4]], align 8 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi double [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP9:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX90A-NEXT: [[TMP6:%.*]] = bitcast double [[NEW2]] to i64 ; GFX90A-NEXT: [[TMP7:%.*]] = bitcast double [[LOADED]] to i64 ; GFX90A-NEXT: [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[TMP4]], i64 [[TMP7]], i64 [[TMP6]] monotonic monotonic, align 8 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP8]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP8]], 0 ; GFX90A-NEXT: [[TMP9]] = bitcast i64 [[NEWLOADED]] to double ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end1: ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.phi: ; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_noret__amdgpu_ignore_denormal_mode( ; GFX942-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX942-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX942: atomicrmw.shared: ; GFX942-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX942-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP1]], double [[VALUE:%.*]] monotonic, align 8, !amdgpu.ignore.denormal.mode [[META0]] ; GFX942-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX942: atomicrmw.check.private: ; GFX942-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX942-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX942: atomicrmw.private: ; GFX942-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX942-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP3]], align 8 ; GFX942-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE]] ; GFX942-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP3]], align 8 ; GFX942-NEXT: br label [[ATOMICRMW_PHI]] ; GFX942: atomicrmw.global: ; GFX942-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX942-NEXT: [[TMP5:%.*]] = load double, ptr addrspace(1) [[TMP4]], align 8 ; GFX942-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX942: atomicrmw.start: ; GFX942-NEXT: [[LOADED:%.*]] = phi double [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP9:%.*]], [[ATOMICRMW_START]] ] ; GFX942-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX942-NEXT: [[TMP6:%.*]] = bitcast double [[NEW2]] to i64 ; GFX942-NEXT: [[TMP7:%.*]] = bitcast double [[LOADED]] to i64 ; GFX942-NEXT: [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[TMP4]], i64 [[TMP7]], i64 [[TMP6]] monotonic monotonic, align 8 ; GFX942-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP8]], 1 ; GFX942-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP8]], 0 ; GFX942-NEXT: [[TMP9]] = bitcast i64 [[NEWLOADED]] to double ; GFX942-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX942: atomicrmw.end1: ; GFX942-NEXT: br label [[ATOMICRMW_PHI]] ; GFX942: atomicrmw.phi: ; GFX942-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX942: atomicrmw.end: ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_noret__amdgpu_ignore_denormal_mode( ; GFX11-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX11-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX11: atomicrmw.private: ; GFX11-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX11-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX11-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX11: atomicrmw.global: ; GFX11-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] monotonic monotonic, align 8, !noalias.addrspace [[META1]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX11-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end1: ; GFX11-NEXT: br label [[ATOMICRMW_PHI]] ; GFX11: atomicrmw.phi: ; GFX11-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %unused = atomicrmw fadd ptr %ptr, double %value monotonic, !amdgpu.ignore.denormal.mode !0 ret void } define double @test_atomicrmw_fadd_f64_dyndenorm_flat_system_ret__amdgpu_ignore_denormal_mode(ptr %ptr, double %value) #5 { ; CI-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_ret__amdgpu_ignore_denormal_mode( ; CI-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; CI-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; CI: atomicrmw.private: ; CI-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; CI-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; CI-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; CI-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; CI-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; CI: atomicrmw.global: ; CI-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; CI-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; CI-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; CI-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] monotonic monotonic, align 8, !noalias.addrspace [[META1]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; CI-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end1: ; CI-NEXT: br label [[ATOMICRMW_PHI]] ; CI: atomicrmw.phi: ; CI-NEXT: [[RET:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; CI-NEXT: br label [[ATOMICRMW_END:%.*]] ; CI: atomicrmw.end: ; CI-NEXT: ret double [[RET]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_ret__amdgpu_ignore_denormal_mode( ; GFX9-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX9-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX9: atomicrmw.private: ; GFX9-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX9-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX9-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX9-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX9: atomicrmw.global: ; GFX9-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX9-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX9-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX9-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] monotonic monotonic, align 8, !noalias.addrspace [[META1]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX9-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end1: ; GFX9-NEXT: br label [[ATOMICRMW_PHI]] ; GFX9: atomicrmw.phi: ; GFX9-NEXT: [[RET:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX9-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret double [[RET]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_ret__amdgpu_ignore_denormal_mode( ; GFX908-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX908-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX908: atomicrmw.private: ; GFX908-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX908-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX908-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX908-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX908: atomicrmw.global: ; GFX908-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX908-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX908-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX908-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] monotonic monotonic, align 8, !noalias.addrspace [[META1]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX908-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end1: ; GFX908-NEXT: br label [[ATOMICRMW_PHI]] ; GFX908: atomicrmw.phi: ; GFX908-NEXT: [[RET:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX908-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret double [[RET]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_ret__amdgpu_ignore_denormal_mode( ; GFX90A-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX90A-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX90A: atomicrmw.shared: ; GFX90A-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX90A-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP1]], double [[VALUE:%.*]] monotonic, align 8, !amdgpu.ignore.denormal.mode [[META0]] ; GFX90A-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX90A: atomicrmw.check.private: ; GFX90A-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX90A-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX90A: atomicrmw.private: ; GFX90A-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX90A-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP3]], align 8 ; GFX90A-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE]] ; GFX90A-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP3]], align 8 ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.global: ; GFX90A-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX90A-NEXT: [[TMP5:%.*]] = load double, ptr addrspace(1) [[TMP4]], align 8 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi double [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP9:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX90A-NEXT: [[TMP6:%.*]] = bitcast double [[NEW2]] to i64 ; GFX90A-NEXT: [[TMP7:%.*]] = bitcast double [[LOADED]] to i64 ; GFX90A-NEXT: [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[TMP4]], i64 [[TMP7]], i64 [[TMP6]] monotonic monotonic, align 8 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP8]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP8]], 0 ; GFX90A-NEXT: [[TMP9]] = bitcast i64 [[NEWLOADED]] to double ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end1: ; GFX90A-NEXT: br label [[ATOMICRMW_PHI]] ; GFX90A: atomicrmw.phi: ; GFX90A-NEXT: [[RET:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP9]], [[ATOMICRMW_END1]] ] ; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret double [[RET]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_ret__amdgpu_ignore_denormal_mode( ; GFX942-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[PTR:%.*]]) ; GFX942-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]] ; GFX942: atomicrmw.shared: ; GFX942-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(3) ; GFX942-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP1]], double [[VALUE:%.*]] monotonic, align 8, !amdgpu.ignore.denormal.mode [[META0]] ; GFX942-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX942: atomicrmw.check.private: ; GFX942-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR]]) ; GFX942-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX942: atomicrmw.private: ; GFX942-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX942-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP3]], align 8 ; GFX942-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE]] ; GFX942-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP3]], align 8 ; GFX942-NEXT: br label [[ATOMICRMW_PHI]] ; GFX942: atomicrmw.global: ; GFX942-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(1) ; GFX942-NEXT: [[TMP5:%.*]] = load double, ptr addrspace(1) [[TMP4]], align 8 ; GFX942-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX942: atomicrmw.start: ; GFX942-NEXT: [[LOADED:%.*]] = phi double [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP9:%.*]], [[ATOMICRMW_START]] ] ; GFX942-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX942-NEXT: [[TMP6:%.*]] = bitcast double [[NEW2]] to i64 ; GFX942-NEXT: [[TMP7:%.*]] = bitcast double [[LOADED]] to i64 ; GFX942-NEXT: [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[TMP4]], i64 [[TMP7]], i64 [[TMP6]] monotonic monotonic, align 8 ; GFX942-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP8]], 1 ; GFX942-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP8]], 0 ; GFX942-NEXT: [[TMP9]] = bitcast i64 [[NEWLOADED]] to double ; GFX942-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX942: atomicrmw.end1: ; GFX942-NEXT: br label [[ATOMICRMW_PHI]] ; GFX942: atomicrmw.phi: ; GFX942-NEXT: [[RET:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP9]], [[ATOMICRMW_END1]] ] ; GFX942-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX942: atomicrmw.end: ; GFX942-NEXT: ret double [[RET]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_f64_dyndenorm_flat_system_ret__amdgpu_ignore_denormal_mode( ; GFX11-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[PTR:%.*]]) ; GFX11-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]] ; GFX11: atomicrmw.private: ; GFX11-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[PTR]] to ptr addrspace(5) ; GFX11-NEXT: [[LOADED_PRIVATE:%.*]] = load double, ptr addrspace(5) [[TMP1]], align 8 ; GFX11-NEXT: [[NEW:%.*]] = fadd double [[LOADED_PRIVATE]], [[VALUE:%.*]] ; GFX11-NEXT: store double [[NEW]], ptr addrspace(5) [[TMP1]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_PHI:%.*]] ; GFX11: atomicrmw.global: ; GFX11-NEXT: [[TMP2:%.*]] = load double, ptr [[PTR]], align 8 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi double [ [[TMP2]], [[ATOMICRMW_GLOBAL]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW2:%.*]] = fadd double [[LOADED]], [[VALUE]] ; GFX11-NEXT: [[TMP3:%.*]] = bitcast double [[NEW2]] to i64 ; GFX11-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 ; GFX11-NEXT: [[TMP5:%.*]] = cmpxchg ptr [[PTR]], i64 [[TMP4]], i64 [[TMP3]] monotonic monotonic, align 8, !noalias.addrspace [[META1]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GFX11-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END1:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end1: ; GFX11-NEXT: br label [[ATOMICRMW_PHI]] ; GFX11: atomicrmw.phi: ; GFX11-NEXT: [[RET:%.*]] = phi double [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP6]], [[ATOMICRMW_END1]] ] ; GFX11-NEXT: br label [[ATOMICRMW_END:%.*]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret double [[RET]] ; %ret = atomicrmw fadd ptr %ptr, double %value monotonic, !amdgpu.ignore.denormal.mode !0 ret double %ret } define void @test_atomicrmw_fadd_f32_region_noret(ptr addrspace(2) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_region_noret( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(2) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(2) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(2) %ptr, float %value monotonic ret void } define float @test_atomicrmw_fadd_f32_region_ret(ptr addrspace(2) %ptr, float %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f32_region_ret( ; ALL-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(2) [[PTR:%.*]], align 4 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi float [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd float [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast float [[NEW]] to i32 ; ALL-NEXT: [[TMP3:%.*]] = bitcast float [[LOADED]] to i32 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(2) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] monotonic monotonic, align 4 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to float ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret float [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(2) %ptr, float %value monotonic ret float %ret } define void @test_atomicrmw_fadd_f64_region_noret(ptr addrspace(2) %ptr, double %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f64_region_noret( ; ALL-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(2) [[PTR:%.*]], align 8 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; ALL-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(2) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret void ; %unused = atomicrmw fadd ptr addrspace(2) %ptr, double %value monotonic ret void } define double @test_atomicrmw_fadd_f64_region_ret(ptr addrspace(2) %ptr, double %value) { ; ALL-LABEL: @test_atomicrmw_fadd_f64_region_ret( ; ALL-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(2) [[PTR:%.*]], align 8 ; ALL-NEXT: br label [[ATOMICRMW_START:%.*]] ; ALL: atomicrmw.start: ; ALL-NEXT: [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; ALL-NEXT: [[NEW:%.*]] = fadd double [[LOADED]], [[VALUE:%.*]] ; ALL-NEXT: [[TMP2:%.*]] = bitcast double [[NEW]] to i64 ; ALL-NEXT: [[TMP3:%.*]] = bitcast double [[LOADED]] to i64 ; ALL-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(2) [[PTR]], i64 [[TMP3]], i64 [[TMP2]] monotonic monotonic, align 8 ; ALL-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP4]], 1 ; ALL-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP4]], 0 ; ALL-NEXT: [[TMP5]] = bitcast i64 [[NEWLOADED]] to double ; ALL-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; ALL: atomicrmw.end: ; ALL-NEXT: ret double [[TMP5]] ; %ret = atomicrmw fadd ptr addrspace(2) %ptr, double %value monotonic ret double %ret } define <2 x half> @test_atomicrmw_fadd_v2f16_flat_agent(ptr %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x half> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x half> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x half> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x half> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret <2 x half> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x half> [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, <2 x half> %value syncscope("agent") seq_cst ret <2 x half> %res } define void @test_atomicrmw_fadd_v2f16_flat_agent_noret(ptr %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr %ptr, <2 x half> %value syncscope("agent") seq_cst ret void } define <2 x half> @test_atomicrmw_fadd_v2f16_global_agent(ptr addrspace(1) %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_global_agent( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x half> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_global_agent( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x half> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_global_agent( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x half> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_global_agent( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x half> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_global_agent( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret <2 x half> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_global_agent( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x half> [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value syncscope("agent") seq_cst ret <2 x half> %res } define void @test_atomicrmw_fadd_v2f16_flat_global_noret(ptr addrspace(1) %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_flat_global_noret( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_flat_global_noret( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_flat_global_noret( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_flat_global_noret( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_flat_global_noret( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_flat_global_noret( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value syncscope("agent") seq_cst ret void } define <2 x half> @test_atomicrmw_fadd_v2f16_local_agent(ptr addrspace(3) %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_local_agent( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x half> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_local_agent( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x half> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_local_agent( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x half> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_local_agent( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x half> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_local_agent( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret <2 x half> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_local_agent( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x half> [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, <2 x half> %value syncscope("agent") seq_cst ret <2 x half> %res } define void @test_atomicrmw_fadd_v2f16_flat_local_noret(ptr addrspace(3) %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_flat_local_noret( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_flat_local_noret( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_flat_local_noret( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_flat_local_noret( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_flat_local_noret( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_flat_local_noret( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(3) %ptr, <2 x half> %value syncscope("agent") seq_cst ret void } define <2 x bfloat> @test_atomicrmw_fadd_v2bf16_flat_agent(ptr %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret <2 x bfloat> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x bfloat> [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, <2 x bfloat> %value syncscope("agent") seq_cst ret <2 x bfloat> %res } define void @test_atomicrmw_fadd_v2bf16_flat_agent_noret(ptr %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr %ptr, <2 x bfloat> %value syncscope("agent") seq_cst ret void } define <2 x bfloat> @test_atomicrmw_fadd_v2bf16_global_agent(ptr addrspace(1) %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret <2 x bfloat> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x bfloat> [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x bfloat> %value syncscope("agent") seq_cst ret <2 x bfloat> %res } define void @test_atomicrmw_fadd_v2bf16_flat_global_noret(ptr addrspace(1) %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_flat_global_noret( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_flat_global_noret( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_flat_global_noret( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_flat_global_noret( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_flat_global_noret( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_flat_global_noret( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x bfloat> %value syncscope("agent") seq_cst ret void } define <2 x bfloat> @test_atomicrmw_fadd_v2bf16_local_agent(ptr addrspace(3) %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_local_agent( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_local_agent( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_local_agent( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_local_agent( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_local_agent( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret <2 x bfloat> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_local_agent( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x bfloat> [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(3) %ptr, <2 x bfloat> %value syncscope("agent") seq_cst ret <2 x bfloat> %res } define void @test_atomicrmw_fadd_v2bf16_flat_local_noret(ptr addrspace(3) %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_flat_local_noret( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_flat_local_noret( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_flat_local_noret( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_flat_local_noret( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_flat_local_noret( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(3) [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_flat_local_noret( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(3) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(3) %ptr, <2 x bfloat> %value syncscope("agent") seq_cst ret void } define <2 x half> @test_atomicrmw_fadd_v2f16_flat_agent__unsafe(ptr %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent__unsafe( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x half> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent__unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x half> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent__unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x half> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent__unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x half> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent__unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret <2 x half> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent__unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x half> [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, <2 x half> %value syncscope("agent") seq_cst ret <2 x half> %res } define void @test_atomicrmw_fadd_v2f16_flat_agent_noret__unsafe(ptr %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret__unsafe( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret__unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret__unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret__unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret__unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4 ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_flat_agent_noret__unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4 ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr %ptr, <2 x half> %value syncscope("agent") seq_cst ret void } define <2 x half> @test_atomicrmw_fadd_v2f16_global_agent__unsafe(ptr addrspace(1) %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_global_agent__unsafe( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x half> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_global_agent__unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x half> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_global_agent__unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x half> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_global_agent__unsafe( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: ret <2 x half> [[RES]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_global_agent__unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret <2 x half> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_global_agent__unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x half> [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0 ret <2 x half> %res } define void @test_atomicrmw_fadd_v2f16_global_agent_noret__unsafe(ptr addrspace(1) %ptr, <2 x half> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2f16_global_agent_noret__unsafe( ; CI-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2f16_global_agent_noret__unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2f16_global_agent_noret__unsafe( ; GFX908-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2f16_global_agent_noret__unsafe( ; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2f16_global_agent_noret__unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2f16_global_agent_noret__unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0 ret void } define <2 x bfloat> @test_atomicrmw_fadd_v2bf16_flat_agent__unsafe(ptr %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent__unsafe( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent__unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent__unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent__unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent__unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret <2 x bfloat> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent__unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x bfloat> [[TMP5]] ; %res = atomicrmw fadd ptr %ptr, <2 x bfloat> %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0 ret <2 x bfloat> %res } define void @test_atomicrmw_fadd_v2bf16_flat_agent_noret__unsafe(ptr %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret__unsafe( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret__unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret__unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret__unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret__unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_flat_agent_noret__unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr %ptr, <2 x bfloat> %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0 ret void } define <2 x bfloat> @test_atomicrmw_fadd_v2bf16_global_agent__unsafe(ptr addrspace(1) %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent__unsafe( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent__unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent__unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent__unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret <2 x bfloat> [[TMP5]] ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent__unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret <2 x bfloat> [[RES]] ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent__unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret <2 x bfloat> [[TMP5]] ; %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x bfloat> %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0 ret <2 x bfloat> %res } define void @test_atomicrmw_fadd_v2bf16_global_agent_noret__unsafe(ptr addrspace(1) %ptr, <2 x bfloat> %value) { ; CI-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent_noret__unsafe( ; CI-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; CI-NEXT: br label [[ATOMICRMW_START:%.*]] ; CI: atomicrmw.start: ; CI-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; CI-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; CI-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; CI-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; CI-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; CI-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; CI-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; CI-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; CI-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; CI: atomicrmw.end: ; CI-NEXT: ret void ; ; GFX9-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent_noret__unsafe( ; GFX9-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX9-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX9: atomicrmw.start: ; GFX9-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX9-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX9-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX9-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX9-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX9-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX9-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX9-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX9-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX9: atomicrmw.end: ; GFX9-NEXT: ret void ; ; GFX908-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent_noret__unsafe( ; GFX908-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX908-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX908: atomicrmw.start: ; GFX908-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX908-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX908-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX908-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX908-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX908-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX908-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX908-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX908-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX908: atomicrmw.end: ; GFX908-NEXT: ret void ; ; GFX90A-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent_noret__unsafe( ; GFX90A-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX90A: atomicrmw.start: ; GFX90A-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX90A-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX90A-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX90A-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX90A-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX90A-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX90A-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX90A: atomicrmw.end: ; GFX90A-NEXT: ret void ; ; GFX942-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent_noret__unsafe( ; GFX942-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], <2 x bfloat> [[VALUE:%.*]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX942-NEXT: ret void ; ; GFX11-LABEL: @test_atomicrmw_fadd_v2bf16_global_agent_noret__unsafe( ; GFX11-NEXT: [[TMP1:%.*]] = load <2 x bfloat>, ptr addrspace(1) [[PTR:%.*]], align 4 ; GFX11-NEXT: br label [[ATOMICRMW_START:%.*]] ; GFX11: atomicrmw.start: ; GFX11-NEXT: [[LOADED:%.*]] = phi <2 x bfloat> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ] ; GFX11-NEXT: [[NEW:%.*]] = fadd <2 x bfloat> [[LOADED]], [[VALUE:%.*]] ; GFX11-NEXT: [[TMP2:%.*]] = bitcast <2 x bfloat> [[NEW]] to i32 ; GFX11-NEXT: [[TMP3:%.*]] = bitcast <2 x bfloat> [[LOADED]] to i32 ; GFX11-NEXT: [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]] ; GFX11-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1 ; GFX11-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0 ; GFX11-NEXT: [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x bfloat> ; GFX11-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] ; GFX11: atomicrmw.end: ; GFX11-NEXT: ret void ; %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x bfloat> %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0 ret void } attributes #0 = { denormal_fpenv(float: preservesign) } attributes #1 = { strictfp denormal_fpenv(float: preservesign) } attributes #2 = { strictfp } attributes #3 = { denormal_fpenv(float: preservesign) } attributes #4 = { denormal_fpenv(float: dynamic) } attributes #5 = { denormal_fpenv(dynamic) } !0 = !{} !1 = !{i32 5, i32 6}