[llvm] [AMDGPU][LDS] Fix dynamic LDS interaction with "amdgpu-no-lds-kernel-id" (PR #107092)

[Juan Manuel Martinez Caamaño via llvm-commits]

https://github.com/jmmartinez updated https://github.com/llvm/llvm-project/pull/107092

>From 511ee2aa9e5f7633595676a69cfa5469a93d512f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Juan=20Manuel=20Martinez=20Caama=C3=B1o?= <juamarti at amd.com>
Date: Tue, 3 Sep 2024 10:41:08 +0200
Subject: [PATCH] [AMDGPU][LDS] Fix dynamic LDS interaction with
 "amdgpu-no-lds-kernel-id"

Dynamic lds and Table lds both use the amdgpu_lds_kernel_id intrinsic.
Kernels and functons that make an indirect use of this should not have the
"amdgpu-no-lds-kernel-id" attribute.

For the later, this was done. For the dynamic lds case, this was
missing. This patch fixes it.
---
 .../Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp    | 15 ++++++++-------
 .../AMDGPU/lower-module-lds-zero-size-arr.ll      |  9 ++++-----
 2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp b/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp
index 38f0b6dda1997b..a166087a5cdc78 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp
@@ -1010,13 +1010,6 @@ class AMDGPULowerModuleLDS {
           M, TableLookupVariablesOrdered, OrderedKernels, KernelToReplacement);
       replaceUsesInInstructionsWithTableLookup(M, TableLookupVariablesOrdered,
                                                LookupTable);
-
-      // Strip amdgpu-no-lds-kernel-id from all functions reachable from the
-      // kernel. We may have inferred this wasn't used prior to the pass.
-      //
-      // TODO: We could filter out subgraphs that do not access LDS globals.
-      for (Function *F : KernelsThatAllocateTableLDS)
-        removeFnAttrFromReachable(CG, F, {"amdgpu-no-lds-kernel-id"});
     }
 
     DenseMap<Function *, GlobalVariable *> KernelToCreatedDynamicLDS =
@@ -1024,6 +1017,14 @@ class AMDGPULowerModuleLDS {
                                  KernelsThatIndirectlyAllocateDynamicLDS,
                                  DynamicVariables, OrderedKernels);
 
+    // Strip amdgpu-no-lds-kernel-id from all functions reachable from the
+    // kernel. We may have inferred this wasn't used prior to the pass.
+    // TODO: We could filter out subgraphs that do not access LDS globals.
+    for (auto *KernelSet : {&KernelsThatIndirectlyAllocateDynamicLDS,
+                            &KernelsThatAllocateTableLDS})
+      for (Function *F : *KernelSet)
+        removeFnAttrFromReachable(CG, F, {"amdgpu-no-lds-kernel-id"});
+
     // All kernel frames have been allocated. Calculate and record the
     // addresses.
     {
diff --git a/llvm/test/CodeGen/AMDGPU/lower-module-lds-zero-size-arr.ll b/llvm/test/CodeGen/AMDGPU/lower-module-lds-zero-size-arr.ll
index c7829be5653738..59dfe3300c2937 100644
--- a/llvm/test/CodeGen/AMDGPU/lower-module-lds-zero-size-arr.ll
+++ b/llvm/test/CodeGen/AMDGPU/lower-module-lds-zero-size-arr.ll
@@ -12,7 +12,7 @@
 ;.
 define void @fn(float %val, i32 %idx) #0 {
 ; CHECK-LABEL: define void @fn(
-; CHECK-SAME: float [[VAL:%.*]], i32 [[IDX:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-SAME: float [[VAL:%.*]], i32 [[IDX:%.*]]) {
 ; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @llvm.amdgcn.lds.kernel.id()
 ; CHECK-NEXT:    [[VAR0:%.*]] = getelementptr inbounds [1 x i32], ptr addrspace(4) @llvm.amdgcn.dynlds.offset.table, i32 0, i32 [[TMP1]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(4) [[VAR0]], align 4
@@ -28,7 +28,7 @@ define void @fn(float %val, i32 %idx) #0 {
 
 define amdgpu_kernel void @kernelA(float %val, i32 %idx) #0 {
 ; CHECK-LABEL: define amdgpu_kernel void @kernelA(
-; CHECK-SAME: float [[VAL:%.*]], i32 [[IDX:%.*]]) #[[ATTR0]] !llvm.amdgcn.lds.kernel.id [[META1:![0-9]+]] {
+; CHECK-SAME: float [[VAL:%.*]], i32 [[IDX:%.*]]) !llvm.amdgcn.lds.kernel.id [[META1:![0-9]+]] {
 ; CHECK-NEXT:    call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.kernelA.dynlds) ]
 ; CHECK-NEXT:    tail call void @fn(float [[VAL]], i32 [[IDX]])
 ; CHECK-NEXT:    ret void
@@ -40,9 +40,8 @@ define amdgpu_kernel void @kernelA(float %val, i32 %idx) #0 {
 attributes #0 = { "amdgpu-no-lds-kernel-id" }
 
 ;.
-; CHECK: attributes #[[ATTR0]] = { "amdgpu-no-lds-kernel-id" }
-; CHECK: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(none) }
-; CHECK: attributes #[[ATTR2:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
+; CHECK: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(none) }
+; CHECK: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
 ;.
 ; CHECK: [[META0]] = !{i32 0, i32 1}
 ; CHECK: [[META1]] = !{i32 0}