[llvm-branch-commits] [llvm] AMDGPU: Preserve atomicrmw name when specializing address space (PR #102470)
Matt Arsenault via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Thu Aug 8 07:11:51 PDT 2024
https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/102470
>From dd1cd7103b426389c879c0a54d23c61910f6553f Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Thu, 8 Aug 2024 18:02:11 +0400
Subject: [PATCH] AMDGPU: Preserve atomicrmw name when specializing address
space
---
llvm/lib/Target/AMDGPU/SIISelLowering.cpp | 3 ++-
...mic-rmw-fadd-flat-specialization-preserve-name.ll | 10 ++++++++++
.../expand-atomic-rmw-fadd-flat-specialization.ll | 12 ++++++------
.../AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll | 4 ++--
4 files changed, 20 insertions(+), 9 deletions(-)
create mode 100644 llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd-flat-specialization-preserve-name.ll
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index e982f1e643289..7735833b1ad2c 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -16654,12 +16654,13 @@ void SITargetLowering::emitExpandAtomicRMW(AtomicRMWInst *AI) const {
Builder.CreateBr(PhiBB);
Builder.SetInsertPoint(PhiBB);
- PHINode *Loaded = Builder.CreatePHI(ValTy, 3, "loaded.phi");
+ PHINode *Loaded = Builder.CreatePHI(ValTy, 3);
Loaded->addIncoming(LoadedShared, SharedBB);
Loaded->addIncoming(LoadedPrivate, PrivateBB);
Loaded->addIncoming(LoadedGlobal, GlobalBB);
Builder.CreateBr(ExitBB);
+ Loaded->takeName(AI);
AI->replaceAllUsesWith(Loaded);
AI->eraseFromParent();
}
diff --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd-flat-specialization-preserve-name.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd-flat-specialization-preserve-name.ll
new file mode 100644
index 0000000000000..44cd7097059cd
--- /dev/null
+++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd-flat-specialization-preserve-name.ll
@@ -0,0 +1,10 @@
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -passes=atomic-expand %s | FileCheck %s
+
+; CHECK: %preserve_me = phi float [ %{{[0-9]+}}, %atomicrmw.shared ], [ %loaded.private, %atomicrmw.private ], [ %{{[0-9]+}}, %atomicrmw.global ]
+; CHECK: ret float %preserve_me
+define float @expand_preserve_name(ptr %addr, float %val) {
+ %preserve_me = atomicrmw fadd ptr %addr, float %val seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.ignore.denormal.mode !0
+ ret float %preserve_me
+}
+
+!0 = !{}
diff --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd-flat-specialization.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd-flat-specialization.ll
index 96cbf057b490a..fc586a01e3bcf 100644
--- a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd-flat-specialization.ll
+++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd-flat-specialization.ll
@@ -42,10 +42,10 @@ define float @syncscope_system(ptr %addr, float %val) {
; GFX90A-NEXT: [[TMP5:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], float [[VAL]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX90A-NEXT: br label [[ATOMICRMW_PHI]]
; GFX90A: atomicrmw.phi:
-; GFX90A-NEXT: [[LOADED_PHI:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
+; GFX90A-NEXT: [[RES:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]]
; GFX90A: atomicrmw.end:
-; GFX90A-NEXT: ret float [[LOADED_PHI]]
+; GFX90A-NEXT: ret float [[RES]]
;
; GFX940-LABEL: @syncscope_system(
; GFX940-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[ADDR:%.*]], float [[VAL:%.*]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]], !amdgpu.ignore.denormal.mode [[META0]]
@@ -112,10 +112,10 @@ define float @syncscope_workgroup_rtn(ptr %addr, float %val) {
; GFX90A-NEXT: [[TMP5:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], float [[VAL]] syncscope("workgroup") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX90A-NEXT: br label [[ATOMICRMW_PHI]]
; GFX90A: atomicrmw.phi:
-; GFX90A-NEXT: [[LOADED_PHI:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
+; GFX90A-NEXT: [[RES:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]]
; GFX90A: atomicrmw.end:
-; GFX90A-NEXT: ret float [[LOADED_PHI]]
+; GFX90A-NEXT: ret float [[RES]]
;
; GFX940-LABEL: @syncscope_workgroup_rtn(
; GFX940-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[ADDR:%.*]], float [[VAL:%.*]] syncscope("workgroup") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
@@ -166,7 +166,7 @@ define void @syncscope_workgroup_nortn(ptr %addr, float %val) {
; GFX908-NEXT: [[TMP5:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], float [[VAL]] syncscope("workgroup") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX908-NEXT: br label [[ATOMICRMW_PHI]]
; GFX908: atomicrmw.phi:
-; GFX908-NEXT: [[LOADED_PHI:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
+; GFX908-NEXT: [[RES:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
; GFX908-NEXT: br label [[ATOMICRMW_END:%.*]]
; GFX908: atomicrmw.end:
; GFX908-NEXT: ret void
@@ -192,7 +192,7 @@ define void @syncscope_workgroup_nortn(ptr %addr, float %val) {
; GFX90A-NEXT: [[TMP5:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], float [[VAL]] syncscope("workgroup") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0]], !amdgpu.ignore.denormal.mode [[META0]]
; GFX90A-NEXT: br label [[ATOMICRMW_PHI]]
; GFX90A: atomicrmw.phi:
-; GFX90A-NEXT: [[LOADED_PHI:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
+; GFX90A-NEXT: [[RES:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]]
; GFX90A: atomicrmw.end:
; GFX90A-NEXT: ret void
diff --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll
index 7eaaf2ae1ec99..0a091bd0fc9ad 100644
--- a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll
+++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll
@@ -615,10 +615,10 @@ define float @test_atomicrmw_fadd_f32_flat_unsafe(ptr %ptr, float %value) #0 {
; GFX90A-NEXT: [[TMP5:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], float [[VALUE]] syncscope("wavefront") monotonic, align 4
; GFX90A-NEXT: br label [[ATOMICRMW_PHI]]
; GFX90A: atomicrmw.phi:
-; GFX90A-NEXT: [[LOADED_PHI:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
+; GFX90A-NEXT: [[RES:%.*]] = phi float [ [[TMP2]], [[ATOMICRMW_SHARED]] ], [ [[LOADED_PRIVATE]], [[ATOMICRMW_PRIVATE]] ], [ [[TMP5]], [[ATOMICRMW_GLOBAL]] ]
; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]]
; GFX90A: atomicrmw.end:
-; GFX90A-NEXT: ret float [[LOADED_PHI]]
+; GFX90A-NEXT: ret float [[RES]]
;
; GFX940-LABEL: @test_atomicrmw_fadd_f32_flat_unsafe(
; GFX940-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[PTR:%.*]], float [[VALUE:%.*]] syncscope("wavefront") monotonic, align 4
More information about the llvm-branch-commits
mailing list