derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
245 lines
8.4 KiB
LLVM
245 lines
8.4 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
|
|
; RUN: opt < %s -mtriple=amdgcn -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,DEFAULT
|
|
; RUN: opt < %s -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,GFX9
|
|
; RUN: opt < %s -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,GFX10
|
|
; RUN: opt < %s -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,GFX11
|
|
|
|
; Test that the m0 operand is folded to poison for message types that don't use it.
|
|
; For the default target (no mcpu), conflicting encodings (IDs 2, 3) should NOT be folded.
|
|
|
|
; MSG_INTERRUPT (1) DOES use m0 - should NOT be folded
|
|
define void @test_sendmsg_interrupt(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_interrupt(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 1, i32 [[VAL:%.*]])
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 1, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_GS (2) + GS_OP_EMIT (2 << 4) DOES use m0 (pre-GFX11) - should NOT be folded
|
|
; On GFX11+ this is MSG_HS_TESSFACTOR which also uses m0 - should NOT be folded
|
|
; On default target, this should NOT be folded (conflicting encoding)
|
|
define void @test_sendmsg_gs_emit(i32 %val) {
|
|
; DEFAULT-LABEL: @test_sendmsg_gs_emit(
|
|
; DEFAULT-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 34, i32 [[VAL:%.*]])
|
|
; DEFAULT-NEXT: ret void
|
|
;
|
|
; GFX9-LABEL: @test_sendmsg_gs_emit(
|
|
; GFX9-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 34, i32 [[VAL:%.*]])
|
|
; GFX9-NEXT: ret void
|
|
;
|
|
; GFX10-LABEL: @test_sendmsg_gs_emit(
|
|
; GFX10-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 34, i32 [[VAL:%.*]])
|
|
; GFX10-NEXT: ret void
|
|
;
|
|
; GFX11-LABEL: @test_sendmsg_gs_emit(
|
|
; GFX11-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 34, i32 [[VAL:%.*]])
|
|
; GFX11-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 34, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_GS_DONE (3) DOES use m0 (pre-GFX11) - should NOT be folded
|
|
; On GFX11+ this is ID_DEALLOC_VGPRS which doesn't use m0
|
|
; On default target, this should NOT be folded (conflicting encoding)
|
|
define void @test_sendmsg_gs_done(i32 %val) {
|
|
; DEFAULT-LABEL: @test_sendmsg_gs_done(
|
|
; DEFAULT-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 3, i32 [[VAL:%.*]])
|
|
; DEFAULT-NEXT: ret void
|
|
;
|
|
; GFX9-LABEL: @test_sendmsg_gs_done(
|
|
; GFX9-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 3, i32 [[VAL:%.*]])
|
|
; GFX9-NEXT: ret void
|
|
;
|
|
; GFX10-LABEL: @test_sendmsg_gs_done(
|
|
; GFX10-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 3, i32 [[VAL:%.*]])
|
|
; GFX10-NEXT: ret void
|
|
;
|
|
; GFX11-LABEL: @test_sendmsg_gs_done(
|
|
; GFX11-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 3, i32 poison)
|
|
; GFX11-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 3, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_SAVEWAVE (4) doesn't use m0 (GFX8-GFX10)
|
|
define void @test_sendmsg_savewave(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_savewave(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 4, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 4, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_STALL_WAVE_GEN (5) doesn't use m0 (GFX9-GFX11)
|
|
define void @test_sendmsg_stall_wave_gen(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_stall_wave_gen(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 5, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 5, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_HALT_WAVES (6) doesn't use m0 (GFX9-GFX11)
|
|
define void @test_sendmsg_halt_waves(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_halt_waves(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 6, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 6, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_ORDERED_PS_DONE (7) doesn't use m0 (GFX9-GFX10)
|
|
define void @test_sendmsg_ordered_ps_done(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_ordered_ps_done(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 7, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 7, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_EARLY_PRIM_DEALLOC (8) doesn't use m0 (GFX9 only)
|
|
define void @test_sendmsg_early_prim_dealloc(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_early_prim_dealloc(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 8, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 8, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_GS_ALLOC_REQ (9) DOES use m0 - should NOT be folded
|
|
define void @test_sendmsg_gs_alloc_req(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_gs_alloc_req(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 9, i32 [[VAL:%.*]])
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 9, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_GET_DOORBELL (10) doesn't use m0 (GFX9-GFX10)
|
|
define void @test_sendmsg_get_doorbell(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_get_doorbell(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 10, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 10, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_GET_DDID (11) doesn't use m0 (GFX10 only)
|
|
define void @test_sendmsg_get_ddid(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_get_ddid(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 11, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 11, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; MSG_SYSMSG (15) doesn't use m0
|
|
define void @test_sendmsg_sysmsg(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_sysmsg(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 15, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 15, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; Test sendmsghalt as well - MSG_INTERRUPT (1) DOES use m0 - should NOT be folded
|
|
define void @test_sendmsghalt_interrupt(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsghalt_interrupt(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsghalt(i32 1, i32 [[VAL:%.*]])
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsghalt(i32 1, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; Test sendmsghalt - MSG_GS_ALLOC_REQ (9) DOES use m0 - should NOT be folded
|
|
define void @test_sendmsghalt_gs_alloc_req(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsghalt_gs_alloc_req(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsghalt(i32 9, i32 [[VAL:%.*]])
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsghalt(i32 9, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; m0 already poison - should be a no-op
|
|
define void @test_sendmsg_already_poison() {
|
|
; CHECK-LABEL: @test_sendmsg_already_poison(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 1, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 1, i32 poison)
|
|
ret void
|
|
}
|
|
|
|
; Test that noundef attribute is dropped when folding to poison
|
|
; Using MSG_SAVEWAVE (4) which doesn't use m0
|
|
define void @test_sendmsg_noundef(i32 noundef %val) {
|
|
; CHECK-LABEL: @test_sendmsg_noundef(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 4, i32 poison)
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 4, i32 noundef %val)
|
|
ret void
|
|
}
|
|
|
|
; Test unknown message ID - should NOT be folded (future-proofing)
|
|
define void @test_sendmsg_unknown_id(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_unknown_id(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 14, i32 [[VAL:%.*]])
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 14, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; Test MSG_HS_TESSFACTOR (2) on GFX11+ - DOES use m0, should NOT be folded
|
|
; On pre-GFX11, this is MSG_GS which also uses m0
|
|
define void @test_sendmsg_hs_tessfactor(i32 %val) {
|
|
; CHECK-LABEL: @test_sendmsg_hs_tessfactor(
|
|
; CHECK-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 2, i32 [[VAL:%.*]])
|
|
; CHECK-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 2, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
; Test MSG_DEALLOC_VGPRS (3) on GFX11+ - doesn't use m0
|
|
; On default target, this should NOT be folded (conflicting encoding with MSG_GS_DONE)
|
|
define void @test_sendmsg_dealloc_vgprs(i32 %val) {
|
|
; DEFAULT-LABEL: @test_sendmsg_dealloc_vgprs(
|
|
; DEFAULT-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 3, i32 [[VAL:%.*]])
|
|
; DEFAULT-NEXT: ret void
|
|
;
|
|
; GFX9-LABEL: @test_sendmsg_dealloc_vgprs(
|
|
; GFX9-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 3, i32 [[VAL:%.*]])
|
|
; GFX9-NEXT: ret void
|
|
;
|
|
; GFX10-LABEL: @test_sendmsg_dealloc_vgprs(
|
|
; GFX10-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 3, i32 [[VAL:%.*]])
|
|
; GFX10-NEXT: ret void
|
|
;
|
|
; GFX11-LABEL: @test_sendmsg_dealloc_vgprs(
|
|
; GFX11-NEXT: call void @llvm.amdgcn.s.sendmsg(i32 3, i32 poison)
|
|
; GFX11-NEXT: ret void
|
|
;
|
|
call void @llvm.amdgcn.s.sendmsg(i32 3, i32 %val)
|
|
ret void
|
|
}
|
|
|
|
declare void @llvm.amdgcn.s.sendmsg(i32 immarg, i32)
|
|
declare void @llvm.amdgcn.s.sendmsghalt(i32 immarg, i32)
|