[llvm] [AMDGPU][Attributor] Stop inferring amdgpu-no-flat-scratch-init in sanitized functions. (PR #161319)

Mon Sep 29 22:54:30 PDT 2025

llvmbot wrote:




@llvm/pr-subscribers-backend-amdgpu

Author: Chaitanya (skc7)

<details>
<summary>Changes</summary>

This PR stops the attributor pass to infer `amdgpu-no-flat-scratch-init` for functions marked with `sanitize_address` attribute.

See #160541 for discussion regarding same.

---
Full diff: https://github.com/llvm/llvm-project/pull/161319.diff


3 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp (+10-10) 
- (added) llvm/test/CodeGen/AMDGPU/amdgpu-attributor-flat-scratch-init-asan.ll (+17) 
- (modified) llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll (+2-2) 


``````````diff

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
index 2ba31562c4784..25f74b578e428 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
@@ -131,10 +131,8 @@ static bool isDSAddress(const Constant *C) {
   return AS == AMDGPUAS::LOCAL_ADDRESS || AS == AMDGPUAS::REGION_ADDRESS;
 }
 
-/// Returns true if the function requires the implicit argument be passed
-/// regardless of the function contents.
-static bool funcRequiresHostcallPtr(const Function &F) {
-  // Sanitizers require the hostcall buffer passed in the implicit arguments.
+/// Returns true if sanitizer attributes are present on a function.
+static bool hasSanitizerAttributes(const Function &F) {
   return F.hasFnAttribute(Attribute::SanitizeAddress) ||
          F.hasFnAttribute(Attribute::SanitizeThread) ||
          F.hasFnAttribute(Attribute::SanitizeMemory) ||
@@ -467,17 +465,19 @@ struct AAAMDAttributesFunction : public AAAMDAttributes {
   void initialize(Attributor &A) override {
     Function *F = getAssociatedFunction();
 
-    // If the function requires the implicit arg pointer due to sanitizers,
-    // assume it's needed even if explicitly marked as not requiring it.
-    const bool NeedsHostcall = funcRequiresHostcallPtr(*F);
-    if (NeedsHostcall) {
+    // If the function has sanitizer attributes, we cannot remove
+    // implicitarg_ptr, hostcall_ptr, or flat_scratch_init.
+    const bool HasSanitizerAttrs = hasSanitizerAttributes(*F);
+    if (HasSanitizerAttrs) {
       removeAssumedBits(IMPLICIT_ARG_PTR);
       removeAssumedBits(HOSTCALL_PTR);
+      removeAssumedBits(FLAT_SCRATCH_INIT);
     }
 
     for (auto Attr : ImplicitAttrs) {
-      if (NeedsHostcall &&
-          (Attr.first == IMPLICIT_ARG_PTR || Attr.first == HOSTCALL_PTR))
+      if (HasSanitizerAttrs &&
+          (Attr.first == IMPLICIT_ARG_PTR || Attr.first == HOSTCALL_PTR ||
+           Attr.first == FLAT_SCRATCH_INIT))
         continue;
 
       if (F->hasFnAttribute(Attr.second))
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-flat-scratch-init-asan.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-flat-scratch-init-asan.ll
new file mode 100644
index 0000000000000..8cb5d49b09cb5
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-attributor-flat-scratch-init-asan.ll
@@ -0,0 +1,17 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --check-globals all --version 4
+; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx90a -passes=amdgpu-attributor %s | FileCheck %s
+
+ at lds_1 = internal addrspace(3) global [1 x i8] poison, align 4
+
+; CHECK-LABEL: @k0
+define amdgpu_kernel void @k0() #0 {
+  store i8 7, ptr addrspace(3) @lds_1, align 4
+  ret void
+}
+
+attributes #0 = { sanitize_address}
+
+!llvm.module.flags = !{!0}
+!0 = !{i32 4, !"nosanitize_address", i32 1}
+
+; CHECK-NOT: attributes {{.*}} = {{.*}} amdgpu-no-flat-scratch-init
diff --git a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll
index a688b6fc6399f..fb566e5bb75f4 100644
--- a/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll
+++ b/llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll
@@ -707,8 +707,8 @@ attributes #6 = { "enqueued-block" }
 ; ATTRIBUTOR_HSA: attributes #[[ATTR14]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "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" "target-cpu"="gfx900" "uniform-work-group-size"="false" }
 ; ATTRIBUTOR_HSA: attributes #[[ATTR15:[0-9]+]] = { nounwind "uniform-work-group-size"="false" }
 ; ATTRIBUTOR_HSA: attributes #[[ATTR16]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "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" "uniform-work-group-size"="false" }
-; ATTRIBUTOR_HSA: attributes #[[ATTR17]] = { nounwind sanitize_address "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-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" }
-; ATTRIBUTOR_HSA: attributes #[[ATTR18]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-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" }
+; ATTRIBUTOR_HSA: attributes #[[ATTR17]] = { nounwind sanitize_address "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-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" }
+; ATTRIBUTOR_HSA: attributes #[[ATTR18]] = { nounwind "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-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" }
 ; ATTRIBUTOR_HSA: attributes #[[ATTR19:[0-9]+]] = { nounwind sanitize_address "amdgpu-no-implicitarg-ptr" "uniform-work-group-size"="false" }
 ; ATTRIBUTOR_HSA: attributes #[[ATTR20:[0-9]+]] = { "enqueued-block" "uniform-work-group-size"="false" }
 ; ATTRIBUTOR_HSA: attributes #[[ATTR21]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "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" "enqueued-block" "uniform-work-group-size"="false" }

``````````

</details>


https://github.com/llvm/llvm-project/pull/161319
