; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck -check-prefixes=COMMON,GFX803 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck -check-prefixes=COMMON,GFX906 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck -check-prefixes=COMMON,GFX908 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck -check-prefixes=COMMON,GFX90A %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck -check-prefixes=COMMON,GFX942 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1030 -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck -check-prefixes=COMMON,GFX10 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck -check-prefixes=COMMON,GFX11 %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck -check-prefixes=COMMON,GFX12 %s

;---------------------------------------------------------------------
; TODO: atomicrmw xchg
;---------------------------------------------------------------------

; ; xchg is supported over PCIe, so no expansion is necessary
; define <2 x half> @test_atomicrmw_xchg_v2f16_global_system(ptr addrspace(1) %ptr, <2 x half> %value) {
;   %res = atomicrmw xchg ptr addrspace(1) %ptr, <2 x half> %value seq_cst
;   ret <2 x half> %res
; }

; ; xchg is supported over PCIe, so no expansion is necessary. Metadata should be ignored.
; define <2 x half> @test_atomicrmw_xchg_v2f16_global_system__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
;   %res = atomicrmw xchg ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0
;   ret <2 x half> %res
; }

; ; xchg is supported over PCIe, so no expansion is necessary. Metadata should be ignored.
; define <2 x half> @test_atomicrmw_xchg_v2f16_global_system__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
;   %res = atomicrmw xchg ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.remote.memory !0
;   ret <2 x half> %res
; }

; ; xchg is supported over PCIe, so no expansion is necessary. Metadata should be ignored.
; define <2 x half> @test_atomicrmw_xchg_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
;   %res = atomicrmw xchg ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
;   ret <2 x half> %res
; }

;---------------------------------------------------------------------
; atomicrmw fadd
;---------------------------------------------------------------------

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0:[0-9]+]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP5]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, 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 <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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]]
; GFX90A-NEXT:    ret <2 x half> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]]
; 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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.remote.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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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:%.*]] ], [ [[RES:%.*]], [[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]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX90A-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX90A-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX90A-NEXT:    [[RES]] = 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> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.remote.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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_remote_memory(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX90A-NEXT:    ret <2 x half> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(ptr addrspace(1) %ptr, <2 x half> %value) #0 {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1:[0-9]+]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1:[0-9]+]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1:[0-9]+]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1:[0-9]+]] {
; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX90A-NEXT:    ret <2 x half> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1:[0-9]+]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1:[0-9]+]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1:[0-9]+]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_daz(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1:[0-9]+]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(ptr addrspace(1) %ptr, <2 x half> %value) #1 {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2:[0-9]+]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2:[0-9]+]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2:[0-9]+]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2:[0-9]+]] {
; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX90A-NEXT:    ret <2 x half> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2:[0-9]+]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2:[0-9]+]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2:[0-9]+]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory___denormal_fp_mode_v2f16_dynamic(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2:[0-9]+]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP5]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] 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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX90A-NEXT:    ret <2 x half> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] 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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.remote.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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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:%.*]] ], [ [[RES:%.*]], [[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]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX90A-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX90A-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX90A-NEXT:    [[RES]] = 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> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.remote.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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX90A-NEXT:    ret <2 x half> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(ptr addrspace(1) %ptr, <2 x half> %value) #0 {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1]] {
; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX90A-NEXT:    ret <2 x half> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_daz(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR1]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(ptr addrspace(1) %ptr, <2 x half> %value) #1 {
; GFX803-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(
; GFX803-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2]] {
; GFX803-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX803-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX803:       atomicrmw.start:
; GFX803-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX803-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX803-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX803-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX803-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX803-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX803-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX803-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX803-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX803:       atomicrmw.end:
; GFX803-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX906-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(
; GFX906-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2]] {
; GFX906-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX906-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX906:       atomicrmw.start:
; GFX906-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX906-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX906-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX906-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX906-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX906-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX906-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX906-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX906-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX906:       atomicrmw.end:
; GFX906-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX908-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(
; GFX908-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(
; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2]] {
; GFX90A-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX90A-NEXT:    ret <2 x half> [[RES]]
;
; GFX942-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(
; GFX942-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2]] {
; GFX942-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX942-NEXT:    ret <2 x half> [[RES]]
;
; GFX10-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(
; GFX10-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2]] {
; GFX10-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; GFX10-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GFX10:       atomicrmw.start:
; GFX10-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; GFX10-NEXT:    [[NEW:%.*]] = fadd <2 x half> [[LOADED]], [[VALUE]]
; GFX10-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; GFX10-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; GFX10-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; GFX10-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; GFX10-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; GFX10-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; GFX10-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GFX10:       atomicrmw.end:
; GFX10-NEXT:    ret <2 x half> [[TMP5]]
;
; GFX11-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(
; GFX11-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2]] {
; 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]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.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]]
;
; GFX12-LABEL: define <2 x half> @test_atomicrmw_fadd_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory__denormal_mode_dynamic(
; GFX12-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR2]] {
; GFX12-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR]], <2 x half> [[VALUE]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX12-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

;---------------------------------------------------------------------
; atomicrmw fsub
;---------------------------------------------------------------------

define <2 x half> @test_atomicrmw_fsub_v2f16_global_system(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fsub_v2f16_global_system(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fsub <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fsub ptr addrspace(1) %ptr, <2 x half> %value seq_cst
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_no_fine_grained_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fsub <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fsub ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fsub <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fsub ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fsub <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fsub ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_ignore_denormal_mode(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fsub <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP5]]
;
  %res = atomicrmw fsub ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fsub <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP5]]
;
  %res = atomicrmw fsub ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fsub <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP5]]
;
  %res = atomicrmw fsub ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fsub_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP5:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[NEW:%.*]] = fsub <2 x half> [[LOADED]], [[VALUE]]
; COMMON-NEXT:    [[TMP2:%.*]] = bitcast <2 x half> [[NEW]] to i32
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP3]], i32 [[TMP2]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP4]], 0
; COMMON-NEXT:    [[TMP5]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP5]]
;
  %res = atomicrmw fsub ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

;---------------------------------------------------------------------
; atomicrmw fmax
;---------------------------------------------------------------------

define <2 x half> @test_atomicrmw_fmax_v2f16_global_system(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmax_v2f16_global_system(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fmax ptr addrspace(1) %ptr, <2 x half> %value seq_cst
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_no_fine_grained_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fmax ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fmax ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fmax ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_ignore_denormal_mode(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[TMP6]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP6]]
;
  %res = atomicrmw fmax ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[TMP6]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP6]]
;
  %res = atomicrmw fmax ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[TMP6]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP6]]
;
  %res = atomicrmw fmax ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmax_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.maxnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[TMP6]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP6]]
;
  %res = atomicrmw fmax ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

;---------------------------------------------------------------------
; atomicrmw fmin
;---------------------------------------------------------------------

define <2 x half> @test_atomicrmw_fmin_v2f16_global_system(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmin_v2f16_global_system(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fmin ptr addrspace(1) %ptr, <2 x half> %value seq_cst
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_no_fine_grained_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fmin ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fmin ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[RES:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[RES]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[RES]]
;
  %res = atomicrmw fmin ptr addrspace(1) %ptr, <2 x half> %value seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_ignore_denormal_mode(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_ignore_denormal_mode(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[TMP6]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP6]]
;
  %res = atomicrmw fmin ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[TMP6]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP6]]
;
  %res = atomicrmw fmin ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[TMP6]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP6]]
;
  %res = atomicrmw fmin ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, <2 x half> %value) {
; COMMON-LABEL: define <2 x half> @test_atomicrmw_fmin_v2f16_global_system__amdgpu_ignore_denormal_mode__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; COMMON-SAME: ptr addrspace(1) [[PTR:%.*]], <2 x half> [[VALUE:%.*]]) #[[ATTR0]] {
; COMMON-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr addrspace(1) [[PTR]], align 4
; COMMON-NEXT:    br label [[ATOMICRMW_START:%.*]]
; COMMON:       atomicrmw.start:
; COMMON-NEXT:    [[LOADED:%.*]] = phi <2 x half> [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
; COMMON-NEXT:    [[TMP2:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[LOADED]], <2 x half> [[VALUE]])
; COMMON-NEXT:    [[TMP3:%.*]] = bitcast <2 x half> [[TMP2]] to i32
; COMMON-NEXT:    [[TMP4:%.*]] = bitcast <2 x half> [[LOADED]] to i32
; COMMON-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.no.remote.memory [[META0]]
; COMMON-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; COMMON-NEXT:    [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0
; COMMON-NEXT:    [[TMP6]] = bitcast i32 [[NEWLOADED]] to <2 x half>
; COMMON-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; COMMON:       atomicrmw.end:
; COMMON-NEXT:    ret <2 x half> [[TMP6]]
;
  %res = atomicrmw fmin ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0, !amdgpu.ignore.denormal.mode !0
  ret <2 x half> %res
}

attributes #0 = { "denormal-fp-mode"="preserve-sign,preserve-sign" }
attributes #1 = { "denormal-fp-mode"="dynamic,dynamic" }

!0 = !{}
;.
; GFX803: [[META0]] = !{}
;.
; GFX906: [[META0]] = !{}
;.
; GFX908: [[META0]] = !{}
;.
; GFX90A: [[META0]] = !{}
;.
; GFX942: [[META0]] = !{}
;.
; GFX10: [[META0]] = !{}
;.
; GFX11: [[META0]] = !{}
;.
; GFX12: [[META0]] = !{}
;.