From: Jay Foad <jay.foad@amd.com>
Date: Mon, 14 Feb 2022 17:50:21 +0000 (+0000)
Subject: [AMDGPU] Honor !invariant.load metadata on load-like intrinsics
X-Git-Tag: upstream/15.0.7~16456
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=f72d8897acfcc3c8476d70353e2c4f91615ff4f0;p=platform%2Fupstream%2Fllvm.git

[AMDGPU] Honor !invariant.load metadata on load-like intrinsics

Differential Revision: https://reviews.llvm.org/D119739
---

diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index ea73fea..0a02f64 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1120,6 +1120,10 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
                                           const CallInst &CI,
                                           MachineFunction &MF,
                                           unsigned IntrID) const {
+  Info.flags = MachineMemOperand::MONone;
+  if (CI.hasMetadata(LLVMContext::MD_invariant_load))
+    Info.flags |= MachineMemOperand::MOInvariant;
+
   if (const AMDGPU::RsrcIntrinsic *RsrcIntr =
           AMDGPU::lookupRsrcIntrinsic(IntrID)) {
     AttributeList Attr = Intrinsic::getAttributes(CI.getContext(),
@@ -1138,7 +1142,7 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
           MFI->getBufferPSV(*MF.getSubtarget<GCNSubtarget>().getInstrInfo());
     }
 
-    Info.flags = MachineMemOperand::MODereferenceable;
+    Info.flags |= MachineMemOperand::MODereferenceable;
     if (Attr.hasFnAttr(Attribute::ReadOnly)) {
       unsigned DMaskLanes = 4;
 
@@ -1180,9 +1184,9 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
       Info.opc = CI.getType()->isVoidTy() ? ISD::INTRINSIC_VOID :
                                             ISD::INTRINSIC_W_CHAIN;
       Info.memVT = MVT::getVT(CI.getArgOperand(0)->getType());
-      Info.flags = MachineMemOperand::MOLoad |
-                   MachineMemOperand::MOStore |
-                   MachineMemOperand::MODereferenceable;
+      Info.flags |= MachineMemOperand::MOLoad |
+                    MachineMemOperand::MOStore |
+                    MachineMemOperand::MODereferenceable;
 
       // XXX - Should this be volatile without known ordering?
       Info.flags |= MachineMemOperand::MOVolatile;
@@ -1202,7 +1206,7 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
     Info.memVT = MVT::getVT(CI.getType());
     Info.ptrVal = CI.getOperand(0);
     Info.align.reset();
-    Info.flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
+    Info.flags |= MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
 
     const ConstantInt *Vol = cast<ConstantInt>(CI.getOperand(4));
     if (!Vol->isZero())
@@ -1218,7 +1222,7 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
     Info.ptrVal =
         MFI->getBufferPSV(*MF.getSubtarget<GCNSubtarget>().getInstrInfo());
     Info.align.reset();
-    Info.flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
+    Info.flags |= MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
 
     const ConstantInt *Vol = dyn_cast<ConstantInt>(CI.getOperand(4));
     if (!Vol || !Vol->isZero())
@@ -1232,7 +1236,7 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
     Info.memVT = MVT::getVT(CI.getType());
     Info.ptrVal = CI.getOperand(0);
     Info.align.reset();
-    Info.flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
+    Info.flags |= MachineMemOperand::MOLoad | MachineMemOperand::MOStore;
 
     const ConstantInt *Vol = cast<ConstantInt>(CI.getOperand(1));
     if (!Vol->isZero())
@@ -1245,9 +1249,9 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
     Info.memVT = MVT::getVT(CI.getType());
     Info.ptrVal = CI.getOperand(0);
     Info.align.reset();
-    Info.flags = MachineMemOperand::MOLoad |
-                 MachineMemOperand::MOStore |
-                 MachineMemOperand::MOVolatile;
+    Info.flags |= MachineMemOperand::MOLoad |
+                  MachineMemOperand::MOStore |
+                  MachineMemOperand::MOVolatile;
     return true;
   }
   case Intrinsic::amdgcn_image_bvh_intersect_ray: {
@@ -1257,8 +1261,8 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
     Info.ptrVal =
         MFI->getImagePSV(*MF.getSubtarget<GCNSubtarget>().getInstrInfo());
     Info.align.reset();
-    Info.flags = MachineMemOperand::MOLoad |
-                 MachineMemOperand::MODereferenceable;
+    Info.flags |= MachineMemOperand::MOLoad |
+                  MachineMemOperand::MODereferenceable;
     return true;
   }
   case Intrinsic::amdgcn_global_atomic_fadd:
@@ -1271,10 +1275,10 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
     Info.memVT = MVT::getVT(CI.getType());
     Info.ptrVal = CI.getOperand(0);
     Info.align.reset();
-    Info.flags = MachineMemOperand::MOLoad |
-                 MachineMemOperand::MOStore |
-                 MachineMemOperand::MODereferenceable |
-                 MachineMemOperand::MOVolatile;
+    Info.flags |= MachineMemOperand::MOLoad |
+                  MachineMemOperand::MOStore |
+                  MachineMemOperand::MODereferenceable |
+                  MachineMemOperand::MOVolatile;
     return true;
   }
   case Intrinsic::amdgcn_ds_gws_init:
@@ -1294,9 +1298,10 @@ bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
     Info.size = 4;
     Info.align = Align(4);
 
-    Info.flags = MachineMemOperand::MOStore;
     if (IntrID == Intrinsic::amdgcn_ds_gws_barrier)
-      Info.flags = MachineMemOperand::MOLoad;
+      Info.flags |= MachineMemOperand::MOLoad;
+    else
+      Info.flags |= MachineMemOperand::MOStore;
     return true;
   }
   default:
diff --git a/llvm/test/CodeGen/AMDGPU/invariant-image-load.ll b/llvm/test/CodeGen/AMDGPU/invariant-image-load.ll
index a315735..3b2a4c5 100644
--- a/llvm/test/CodeGen/AMDGPU/invariant-image-load.ll
+++ b/llvm/test/CodeGen/AMDGPU/invariant-image-load.ll
@@ -8,13 +8,13 @@ define amdgpu_ps void @test(<8 x i32> inreg %load, <8 x i32> inreg %store) {
 ; GFX9-LABEL: test:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-NEXT:    image_load v1, v0, s[0:7] dmask:0x1 unorm
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    image_store v1, v0, s[8:15] dmask:0x1 unorm
-; GFX9-NEXT:    v_mov_b32_e32 v0, 1
-; GFX9-NEXT:    image_load v1, v0, s[0:7] dmask:0x1 unorm
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    image_store v1, v0, s[8:15] dmask:0x1 unorm
+; GFX9-NEXT:    v_mov_b32_e32 v1, 1
+; GFX9-NEXT:    image_load v2, v0, s[0:7] dmask:0x1 unorm
+; GFX9-NEXT:    image_load v3, v1, s[0:7] dmask:0x1 unorm
+; GFX9-NEXT:    s_waitcnt vmcnt(1)
+; GFX9-NEXT:    image_store v2, v0, s[8:15] dmask:0x1 unorm
+; GFX9-NEXT:    s_waitcnt vmcnt(1)
+; GFX9-NEXT:    image_store v3, v1, s[8:15] dmask:0x1 unorm
 ; GFX9-NEXT:    s_endpgm
   %data0 = call float @llvm.amdgcn.image.load.1d.f32.i32(i32 1, i32 0, <8 x i32> %load, i32 0, i32 0), !invariant.load !0
   call void @llvm.amdgcn.image.store.1d.f32.i32(float %data0, i32 1, i32 0, <8 x i32> %store, i32 0, i32 0)