[AMDGPU] Set TGID_EN_X/Y/Z when cluster ID intrinsics are used #159120
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Hardware initializes a single value in ttmp9 which is either the workgroup ID X or cluster ID X. Most of this patch is a refactoring to use a single `PreloadedValue` enumerator for this value, instead of two enumerators `WORKGROUP_ID_X` and `CLUSTER_ID_X` referring to the same value. This makes it simpler to have a single attribute `amdgpu-no-workgroup-id-x` indicating that this value is not used, which in turns sets the TGID_EN_X bit appropriately to tell the hardware whether to initialize it. All of the above applies to Y and Z similarly. Fixes: LWPSCGFX13-568 Co-authored-by: Jay Foad <jay.foad@amd.com>
|
@llvm/pr-subscribers-llvm-globalisel @llvm/pr-subscribers-backend-amdgpu Author: Shilei Tian (shiltian) ChangesHardware initializes a single value in ttmp9 which is either the workgroup ID X or cluster ID X. Most of this patch is a refactoring to use a single This makes it simpler to have a single attribute All of the above applies to Y and Z similarly. Fixes: LWPSCGFX13-568 Co-authored-by: Jay Foad <jay.foad@amd.com> Patch is 76.69 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/159120.diff 31 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
index 21255f691e4ad..7afaddea164f8 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUCallLowering.cpp
@@ -809,15 +809,15 @@ bool AMDGPUCallLowering::passSpecialInputs(MachineIRBuilder &MIRBuilder,
AMDGPUFunctionArgInfo::LDS_KERNEL_ID,
};
- static constexpr StringLiteral ImplicitAttrNames[] = {
- "amdgpu-no-dispatch-ptr",
- "amdgpu-no-queue-ptr",
- "amdgpu-no-implicitarg-ptr",
- "amdgpu-no-dispatch-id",
- "amdgpu-no-workgroup-id-x",
- "amdgpu-no-workgroup-id-y",
- "amdgpu-no-workgroup-id-z",
- "amdgpu-no-lds-kernel-id",
+ static constexpr StringLiteral ImplicitAttrNames[][2] = {
+ {"amdgpu-no-dispatch-ptr", ""},
+ {"amdgpu-no-queue-ptr", ""},
+ {"amdgpu-no-implicitarg-ptr", ""},
+ {"amdgpu-no-dispatch-id", ""},
+ {"amdgpu-no-workgroup-id-x", "amdgpu-no-cluster-id-x"},
+ {"amdgpu-no-workgroup-id-y", "amdgpu-no-cluster-id-y"},
+ {"amdgpu-no-workgroup-id-z", "amdgpu-no-cluster-id-z"},
+ {"amdgpu-no-lds-kernel-id", ""},
};
MachineRegisterInfo &MRI = MF.getRegInfo();
@@ -833,7 +833,9 @@ bool AMDGPUCallLowering::passSpecialInputs(MachineIRBuilder &MIRBuilder,
LLT ArgTy;
// If the callee does not use the attribute value, skip copying the value.
- if (Info.CB->hasFnAttr(ImplicitAttrNames[I++]))
+ if (all_of(ImplicitAttrNames[I++], [&](StringRef AttrName) {
+ return AttrName.empty() || Info.CB->hasFnAttr(AttrName);
+ }))
continue;
std::tie(OutgoingArg, ArgRC, ArgTy) =
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 015f8fe49ebcf..363717b017ef0 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -3793,21 +3793,23 @@ void SITargetLowering::passSpecialInputs(
// in the same location as the input.
// clang-format off
static constexpr std::pair<AMDGPUFunctionArgInfo::PreloadedValue,
- StringLiteral> ImplicitAttrs[] = {
- {AMDGPUFunctionArgInfo::DISPATCH_PTR, "amdgpu-no-dispatch-ptr"},
- {AMDGPUFunctionArgInfo::QUEUE_PTR, "amdgpu-no-queue-ptr" },
- {AMDGPUFunctionArgInfo::IMPLICIT_ARG_PTR, "amdgpu-no-implicitarg-ptr"},
- {AMDGPUFunctionArgInfo::DISPATCH_ID, "amdgpu-no-dispatch-id"},
- {AMDGPUFunctionArgInfo::WORKGROUP_ID_X, "amdgpu-no-workgroup-id-x"},
- {AMDGPUFunctionArgInfo::WORKGROUP_ID_Y,"amdgpu-no-workgroup-id-y"},
- {AMDGPUFunctionArgInfo::WORKGROUP_ID_Z,"amdgpu-no-workgroup-id-z"},
- {AMDGPUFunctionArgInfo::LDS_KERNEL_ID,"amdgpu-no-lds-kernel-id"},
- };
+ std::array<StringLiteral, 2>> ImplicitAttrs[] = {
+ {AMDGPUFunctionArgInfo::DISPATCH_PTR, {"amdgpu-no-dispatch-ptr", ""}},
+ {AMDGPUFunctionArgInfo::QUEUE_PTR, {"amdgpu-no-queue-ptr", ""}},
+ {AMDGPUFunctionArgInfo::IMPLICIT_ARG_PTR, {"amdgpu-no-implicitarg-ptr", ""}},
+ {AMDGPUFunctionArgInfo::DISPATCH_ID, {"amdgpu-no-dispatch-id", ""}},
+ {AMDGPUFunctionArgInfo::WORKGROUP_ID_X, {"amdgpu-no-workgroup-id-x", "amdgpu-no-cluster-id-x"}},
+ {AMDGPUFunctionArgInfo::WORKGROUP_ID_Y, {"amdgpu-no-workgroup-id-y", "amdgpu-no-cluster-id-y"}},
+ {AMDGPUFunctionArgInfo::WORKGROUP_ID_Z, {"amdgpu-no-workgroup-id-z", "amdgpu-no-cluster-id-z"}},
+ {AMDGPUFunctionArgInfo::LDS_KERNEL_ID, {"amdgpu-no-lds-kernel-id", ""}},
+ };
// clang-format on
- for (auto [InputID, Attr] : ImplicitAttrs) {
+ for (auto [InputID, Attrs] : ImplicitAttrs) {
// If the callee does not use the attribute value, skip copying the value.
- if (CLI.CB->hasFnAttr(Attr))
+ if (all_of(Attrs, [&](StringRef Attr) {
+ return Attr.empty() || CLI.CB->hasFnAttr(Attr);
+ }))
continue;
const auto [OutgoingArg, ArgRC, ArgTy] =
diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
index 1f11be475e9f8..908d856d386f5 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
@@ -132,13 +132,16 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const Function &F,
if (!AMDGPU::isGraphics(CC) ||
((CC == CallingConv::AMDGPU_CS || CC == CallingConv::AMDGPU_Gfx) &&
ST.hasArchitectedSGPRs())) {
- if (IsKernel || !F.hasFnAttribute("amdgpu-no-workgroup-id-x"))
+ if (IsKernel || !F.hasFnAttribute("amdgpu-no-workgroup-id-x") ||
+ !F.hasFnAttribute("amdgpu-no-cluster-id-x"))
WorkGroupIDX = true;
- if (!F.hasFnAttribute("amdgpu-no-workgroup-id-y"))
+ if (!F.hasFnAttribute("amdgpu-no-workgroup-id-y") ||
+ !F.hasFnAttribute("amdgpu-no-cluster-id-y"))
WorkGroupIDY = true;
- if (!F.hasFnAttribute("amdgpu-no-workgroup-id-z"))
+ if (!F.hasFnAttribute("amdgpu-no-workgroup-id-z") ||
+ !F.hasFnAttribute("amdgpu-no-cluster-id-z"))
WorkGroupIDZ = true;
}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll
index 679d4a26d26b2..c16c8e2128c72 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/call-outgoing-stack-args.ll
@@ -430,5 +430,5 @@ define void @func_caller_byval(ptr addrspace(5) %argptr) {
declare void @llvm.memset.p5.i32(ptr addrspace(5) nocapture writeonly, i8, i32, i1 immarg) #1
-attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
+attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
attributes #1 = { argmemonly nofree nounwind willreturn writeonly }
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/dereferenceable-declaration.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/dereferenceable-declaration.ll
index 13828c2d8a6a0..c92e5c5f8bfff 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/dereferenceable-declaration.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/dereferenceable-declaration.ll
@@ -232,4 +232,4 @@ define i64 @load_deref_or_null_maxmimum_callsite_declaration_only() {
ret i64 %add
}
-attributes #0 = { "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
+attributes #0 = { "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-abi-attribute-hints.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-abi-attribute-hints.ll
index bbbce9a0719ab..1bf2a589cb597 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-abi-attribute-hints.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-call-abi-attribute-hints.ll
@@ -88,7 +88,7 @@ define amdgpu_kernel void @kernel_call_no_workgroup_ids() {
; CHECK-NEXT: $sgpr30_sgpr31 = noconvergent G_SI_CALL [[GV]](p0), @extern, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr15, implicit $vgpr31
; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc
; CHECK-NEXT: S_ENDPGM 0
- call void @extern() "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z"
+ call void @extern() "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z"
ret void
}
@@ -124,7 +124,7 @@ define amdgpu_kernel void @kernel_call_no_other_sgprs() {
; CHECK-NEXT: $sgpr30_sgpr31 = noconvergent G_SI_CALL [[GV]](p0), @extern, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr8_sgpr9, implicit $sgpr15, implicit $vgpr31
; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc
; CHECK-NEXT: S_ENDPGM 0
- call void @extern() "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z"
+ call void @extern() "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z"
ret void
}
@@ -198,7 +198,7 @@ define void @func_call_no_workgroup_ids() {
; CHECK-NEXT: $sgpr30_sgpr31 = noconvergent G_SI_CALL [[GV]](p0), @extern, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr15, implicit $vgpr31
; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc
; CHECK-NEXT: SI_RETURN
- call void @extern() "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z"
+ call void @extern() "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z"
ret void
}
@@ -223,7 +223,7 @@ define void @func_call_no_other_sgprs() {
; CHECK-NEXT: $sgpr30_sgpr31 = noconvergent G_SI_CALL [[GV]](p0), @extern, csr_amdgpu, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr8_sgpr9, implicit $sgpr15, implicit $vgpr31
; CHECK-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def $scc
; CHECK-NEXT: SI_RETURN
- call void @extern() "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z"
+ call void @extern() "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z"
ret void
}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sibling-call.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sibling-call.ll
index 97c3e903c9aec..7b2e3bf13c368 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sibling-call.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/irtranslator-sibling-call.ll
@@ -1486,5 +1486,5 @@ entry:
}
attributes #0 = { nounwind }
-attributes #1 = { nounwind noinline "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
+attributes #1 = { nounwind noinline "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
diff --git a/llvm/test/CodeGen/AMDGPU/amdhsa-kernarg-preload-num-sgprs.ll b/llvm/test/CodeGen/AMDGPU/amdhsa-kernarg-preload-num-sgprs.ll
index dd760c2a215ca..a160cdc950eb5 100644
--- a/llvm/test/CodeGen/AMDGPU/amdhsa-kernarg-preload-num-sgprs.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdhsa-kernarg-preload-num-sgprs.ll
@@ -70,4 +70,4 @@ define amdgpu_kernel void @amdhsa_kernarg_preload_1_implicit_2(i32 inreg) #0 { r
define amdgpu_kernel void @amdhsa_kernarg_preload_0_implicit_2(i32) #0 { ret void }
-attributes #0 = { "amdgpu-agpr-alloc"="0" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
+attributes #0 = { "amdgpu-agpr-alloc"="0" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
diff --git a/llvm/test/CodeGen/AMDGPU/call-argument-types.ll b/llvm/test/CodeGen/AMDGPU/call-argument-types.ll
index b8dd377377dab..306fe33bfb7ac 100644
--- a/llvm/test/CodeGen/AMDGPU/call-argument-types.ll
+++ b/llvm/test/CodeGen/AMDGPU/call-argument-types.ll
@@ -7283,7 +7283,7 @@ declare hidden void @external_void_func_12xv3f32(<3 x float>, <3 x float>, <3 x
declare hidden void @external_void_func_8xv5f32(<5 x float>, <5 x float>, <5 x float>, <5 x float>,
<5 x float>, <5 x float>, <5 x float>, <5 x float>) #0
-attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
+attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
attributes #1 = { nounwind readnone }
attributes #2 = { nounwind noinline }
diff --git a/llvm/test/CodeGen/AMDGPU/call-defs-mode-register.ll b/llvm/test/CodeGen/AMDGPU/call-defs-mode-register.ll
index 0c4974f347a8f..ffe536d347c53 100644
--- a/llvm/test/CodeGen/AMDGPU/call-defs-mode-register.ll
+++ b/llvm/test/CodeGen/AMDGPU/call-defs-mode-register.ll
@@ -54,4 +54,4 @@ define float @asm_changes_mode(float %x, float %y) #0 {
declare float @llvm.experimental.constrained.fadd.f32(float, float, metadata, metadata)
-attributes #0 = { strictfp "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
+attributes #0 = { strictfp "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
diff --git a/llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll b/llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll
index 69ad8e96c7c5d..61a195f9c314f 100644
--- a/llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll
+++ b/llvm/test/CodeGen/AMDGPU/call-preserved-registers.ll
@@ -747,4 +747,4 @@ define amdgpu_kernel void @callee_saved_sgpr_vgpr_kernel() #2 {
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }
attributes #2 = { nounwind noinline }
-attributes #3 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
+attributes #3 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
diff --git a/llvm/test/CodeGen/AMDGPU/call-reqd-group-size.ll b/llvm/test/CodeGen/AMDGPU/call-reqd-group-size.ll
index 093ca55698fe3..33eb8c1e8f4f5 100644
--- a/llvm/test/CodeGen/AMDGPU/call-reqd-group-size.ll
+++ b/llvm/test/CodeGen/AMDGPU/call-reqd-group-size.ll
@@ -208,7 +208,7 @@ define amdgpu_kernel void @known_xyz_0(ptr addrspace(1) %out) !reqd_work_group_s
}
; CHECK: .amdhsa_system_vgpr_workitem_id 0
-attributes #0 = { "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" }
+attributes #0 = { "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" }
!0 = !{i32 1, i32 64, i32 64}
!1 = !{i32 64, i32 1, i32 64}
diff --git a/llvm/test/CodeGen/AMDGPU/call-waitcnt.ll b/llvm/test/CodeGen/AMDGPU/call-waitcnt.ll
index 9abb50651146a..675acd0eedfc5 100644
--- a/llvm/test/CodeGen/AMDGPU/call-waitcnt.ll
+++ b/llvm/test/CodeGen/AMDGPU/call-waitcnt.ll
@@ -154,5 +154,5 @@ declare hidden void @func(i32) #0
declare hidden i32 @func.return(i32) #0
declare void @got.func(i32) #0
-attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
+attributes #0 = { nounwind "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-cluster-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-cluster-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
diff --git a/llvm/test/CodeGen/AMDGPU/elf-notes.ll b/llvm/test/CodeGen/AMDGPU/elf-notes.ll
index 22d699a8f4809..d0dec1f1fe7e4 100644
--- a/llvm/test/CodeGen/AMDGPU/elf-notes.ll
+++ b/llvm/test/CodeGen/AMDGPU/elf-notes.ll
@@ -85,7 +85,7 @@ define amdgpu_kernel void @elf_notes() #0 {
ret void
}
-attributes #0 = { "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }
+attributes #0 = { "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-cluster-id-x" "amdgpu-...
[truncated]
|
rampitec
approved these changes
Sep 16, 2025
arsenm
reviewed
Sep 17, 2025
Comment on lines
+139
to
+140
| if (!F.hasFnAttribute("amdgpu-no-workgroup-id-y") || | ||
| !F.hasFnAttribute("amdgpu-no-cluster-id-y")) |
There was a problem hiding this comment.
Why isn't this &&? I'd also expect the original condition to be strong enough if the attributor understood the implied attribute?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Hardware initializes a single value in ttmp9 which is either the workgroup ID X or cluster ID X. Most of this patch is a refactoring to use a single
PreloadedValueenumerator for this value, instead of two enumeratorsWORKGROUP_ID_XandCLUSTER_ID_Xreferring to the same value.This makes it simpler to have a single attribute
amdgpu-no-workgroup-id-xindicating that this value is not used, which in turns sets the TGID_EN_X bit appropriately to tell the hardware whether to initialize it.All of the above applies to Y and Z similarly.
Fixes: LWPSCGFX13-568
Co-authored-by: Jay Foad jay.foad@amd.com