[llvm-branch-commits] [llvm] AMDGPU: Add noalias.addrspace metadata when autoupgrading atomic intrinsics (PR #102599)

Matt Arsenault via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Mon Sep 30 00:09:26 PDT 2024


https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/102599

>From a2719d4938a1eaf135c275257b1b6c0318ccc801 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Fri, 9 Aug 2024 14:51:41 +0400
Subject: [PATCH] AMDGPU: Add noalias.addrspace metadata when autoupgrading
 atomic intrinsics

This will be needed to continue generating the raw instruction in the flat case.
---
 llvm/lib/IR/AutoUpgrade.cpp        | 13 ++++++++-
 llvm/test/Bitcode/amdgcn-atomic.ll | 45 ++++++++++++++++--------------
 2 files changed, 36 insertions(+), 22 deletions(-)

diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 6f833acd6dbc0d..ca2602e56136a7 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -34,9 +34,11 @@
 #include "llvm/IR/IntrinsicsWebAssembly.h"
 #include "llvm/IR/IntrinsicsX86.h"
 #include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/MDBuilder.h"
 #include "llvm/IR/Metadata.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Verifier.h"
+#include "llvm/Support/AMDGPUAddrSpace.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Regex.h"
@@ -4235,13 +4237,22 @@ static Value *upgradeAMDGCNIntrinsicCall(StringRef Name, CallBase *CI,
   AtomicRMWInst *RMW =
       Builder.CreateAtomicRMW(RMWOp, Ptr, Val, std::nullopt, Order, SSID);
 
-  if (PtrTy->getAddressSpace() != 3) {
+  unsigned AddrSpace = PtrTy->getAddressSpace();
+  if (AddrSpace != AMDGPUAS::LOCAL_ADDRESS) {
     MDNode *EmptyMD = MDNode::get(F->getContext(), {});
     RMW->setMetadata("amdgpu.no.fine.grained.memory", EmptyMD);
     if (RMWOp == AtomicRMWInst::FAdd && RetTy->isFloatTy())
       RMW->setMetadata("amdgpu.ignore.denormal.mode", EmptyMD);
   }
 
+  if (AddrSpace == AMDGPUAS::FLAT_ADDRESS) {
+    MDBuilder MDB(F->getContext());
+    MDNode *RangeNotPrivate =
+        MDB.createRange(APInt(32, AMDGPUAS::PRIVATE_ADDRESS),
+                        APInt(32, AMDGPUAS::PRIVATE_ADDRESS + 1));
+    RMW->setMetadata(LLVMContext::MD_noalias_addrspace, RangeNotPrivate);
+  }
+
   if (IsVolatile)
     RMW->setVolatile(true);
 
diff --git a/llvm/test/Bitcode/amdgcn-atomic.ll b/llvm/test/Bitcode/amdgcn-atomic.ll
index d642372799f56b..87ca1e3a617ed9 100644
--- a/llvm/test/Bitcode/amdgcn-atomic.ll
+++ b/llvm/test/Bitcode/amdgcn-atomic.ll
@@ -2,10 +2,10 @@
 
 
 define void @atomic_inc(ptr %ptr0, ptr addrspace(1) %ptr1, ptr addrspace(3) %ptr3) {
-  ; CHECK: atomicrmw uinc_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw uinc_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result0 = call i32 @llvm.amdgcn.atomic.inc.i32.p0(ptr %ptr0, i32 42, i32 0, i32 0, i1 false)
 
-  ; CHECK: atomicrmw uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !1
   %result1 = call i32 @llvm.amdgcn.atomic.inc.i32.p1(ptr addrspace(1) %ptr1, i32 43, i32 0, i32 0, i1 false)
 
   ; CHECK: atomicrmw uinc_wrap ptr addrspace(3) %ptr3, i32 46 syncscope("agent") seq_cst, align 4{{$}}
@@ -26,10 +26,10 @@ define void @atomic_inc(ptr %ptr0, ptr addrspace(1) %ptr1, ptr addrspace(3) %ptr
 }
 
 define void @atomic_dec(ptr %ptr0, ptr addrspace(1) %ptr1, ptr addrspace(3) %ptr3) {
-  ; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result0 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 0, i32 0, i1 false)
 
-  ; CHECK: atomicrmw udec_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw udec_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !1
   %result1 = call i32 @llvm.amdgcn.atomic.dec.i32.p1(ptr addrspace(1) %ptr1, i32 43, i32 0, i32 0, i1 false)
 
   ; CHECK: atomicrmw udec_wrap ptr addrspace(3) %ptr3, i32 46 syncscope("agent") seq_cst, align 4{{$}}
@@ -51,49 +51,49 @@ define void @atomic_dec(ptr %ptr0, ptr addrspace(1) %ptr1, ptr addrspace(3) %ptr
 
 ; Test some invalid ordering handling
 define void @ordering(ptr %ptr0, ptr addrspace(1) %ptr1, ptr addrspace(3) %ptr3) {
-  ; CHECK: atomicrmw volatile uinc_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw volatile uinc_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result0 = call i32 @llvm.amdgcn.atomic.inc.i32.p0(ptr %ptr0, i32 42, i32 -1, i32 0, i1 true)
 
-  ; CHECK: atomicrmw volatile uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw volatile uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !1
   %result1 = call i32 @llvm.amdgcn.atomic.inc.i32.p1(ptr addrspace(1) %ptr1, i32 43, i32 0, i32 0, i1 true)
 
-  ; CHECK: atomicrmw uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !1
   %result2 = call i32 @llvm.amdgcn.atomic.inc.i32.p1(ptr addrspace(1) %ptr1, i32 43, i32 1, i32 0, i1 false)
 
-  ; CHECK: atomicrmw volatile uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") monotonic, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw volatile uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") monotonic, align 4, !amdgpu.no.fine.grained.memory !1
   %result3 = call i32 @llvm.amdgcn.atomic.inc.i32.p1(ptr addrspace(1) %ptr1, i32 43, i32 2, i32 0, i1 true)
 
-  ; CHECK: atomicrmw uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw uinc_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !1
   %result4 = call i32 @llvm.amdgcn.atomic.inc.i32.p1(ptr addrspace(1) %ptr1, i32 43, i32 3, i32 0, i1 false)
 
-  ; CHECK: atomicrmw volatile udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw volatile udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result5 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 0, i32 4, i1 true)
 
-  ; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result6 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 0, i32 5, i1 false)
 
-  ; CHECK: atomicrmw volatile udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw volatile udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result7 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 0, i32 6, i1 true)
 
-  ; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result8 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 0, i32 7, i1 false)
 
-  ; CHECK:= atomicrmw volatile udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK:= atomicrmw volatile udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result9 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 0, i32 8, i1 true)
 
-  ; CHECK:= atomicrmw volatile udec_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK:= atomicrmw volatile udec_wrap ptr addrspace(1) %ptr1, i32 43 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !1
   %result10 = call i32 @llvm.amdgcn.atomic.dec.i32.p1(ptr addrspace(1) %ptr1, i32 43, i32 3, i32 0, i1 true)
   ret void
 }
 
 define void @immarg_violations(ptr %ptr0, i32 %val32, i1 %val1) {
-  ; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result0 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 %val32, i32 0, i1 false)
 
-; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") monotonic, align 4, !amdgpu.no.fine.grained.memory !0
+; CHECK: atomicrmw udec_wrap ptr %ptr0, i32 42 syncscope("agent") monotonic, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result1 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 2, i32 %val32, i1 false)
 
-  ; CHECK: atomicrmw volatile udec_wrap ptr %ptr0, i32 42 syncscope("agent") monotonic, align 4, !amdgpu.no.fine.grained.memory !0
+  ; CHECK: atomicrmw volatile udec_wrap ptr %ptr0, i32 42 syncscope("agent") monotonic, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !1{{$}}
   %result2 = call i32 @llvm.amdgcn.atomic.dec.i32.p0(ptr %ptr0, i32 42, i32 2, i32 0, i1 %val1)
   ret void
 }
@@ -304,7 +304,7 @@ declare <2 x i16> @llvm.amdgcn.flat.atomic.fadd.v2bf16.p0(ptr, <2 x i16>)
 
 define <2 x i16> @upgrade_amdgcn_flat_atomic_fadd_v2bf16_p0(ptr %ptr, <2 x i16> %data) {
   ; CHECK: [[BC0:%.+]] = bitcast <2 x i16> %data to <2 x bfloat>
-  ; CHECK-NEXT: [[ATOMIC:%.+]] = atomicrmw fadd ptr %ptr, <2 x bfloat> [[BC0]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
+  ; CHECK-NEXT: [[ATOMIC:%.+]] = atomicrmw fadd ptr %ptr, <2 x bfloat> [[BC0]] syncscope("agent") seq_cst, align 4, !noalias.addrspace !0, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
   ; CHECK-NEXT: [[BC1:%.+]] = bitcast <2 x bfloat> [[ATOMIC]] to <2 x i16>
   ; CHECK-NEXT: ret <2 x i16> [[BC1]]
   %result = call <2 x i16> @llvm.amdgcn.flat.atomic.fadd.v2bf16.p0(ptr %ptr, <2 x i16> %data)
@@ -325,7 +325,7 @@ define <2 x i16> @upgrade_amdgcn_global_atomic_fadd_v2bf16_p1(ptr addrspace(1) %
 declare <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p0.v2f16(ptr nocapture, <2 x half>) #0
 
 define <2 x half> @upgrade_amdgcn_flat_atomic_fadd_v2f16_p0_v2f16(ptr %ptr, <2 x half> %data) {
-  ; CHECK: %{{.+}} = atomicrmw fadd ptr %ptr, <2 x half> %data syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
+  ; CHECK: %{{.+}} = atomicrmw fadd ptr %ptr, <2 x half> %data syncscope("agent") seq_cst, align 4, !noalias.addrspace !{{[0-9]+}}, !amdgpu.no.fine.grained.memory !{{[0-9]+$}}
   %result = call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p0.v2f16(ptr %ptr, <2 x half> %data)
   ret <2 x half> %result
 }
@@ -341,7 +341,7 @@ define <2 x half> @upgrade_amdgcn_global_atomic_fadd_v2f16_p1_v2f16(ptr addrspac
 declare float @llvm.amdgcn.flat.atomic.fadd.f32.p0.f32(ptr nocapture, float) #0
 
 define float @upgrade_amdgcn_flat_atomic_fadd_f32_p0_f32(ptr %ptr, float %data) {
-  ; CHECK: %{{.+}} = atomicrmw fadd ptr %ptr, float %data syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !{{[0-9]+}}, !amdgpu.ignore.denormal.mode !{{[0-9]+$}}
+  ; CHECK: %{{.+}} = atomicrmw fadd ptr %ptr, float %data syncscope("agent") seq_cst, align 4, !noalias.addrspace !{{[0-9]+}}, !amdgpu.no.fine.grained.memory !{{[0-9]+}}, !amdgpu.ignore.denormal.mode !{{[0-9]+$}}
   %result = call float @llvm.amdgcn.flat.atomic.fadd.f32.p0.f32(ptr %ptr, float %data)
   ret float %result
 }
@@ -355,3 +355,6 @@ define float @upgrade_amdgcn_global_atomic_fadd_f32_p1_f32(ptr addrspace(1) %ptr
 }
 
 attributes #0 = { argmemonly nounwind willreturn }
+
+; CHECK: !0 = !{i32 5, i32 6}
+; CHECK: !1 = !{}



More information about the llvm-branch-commits mailing list