[InstCombine] Cleanup the TFE/LWE check in AMDGPU SimplifyDemanded

Summary:
The fix added in r352904 is not quite correct, or rather misleading:

1. When the texfailctrl (TFC) argument was non-constant, the fix assumed
   non-TFE/LWE, which is incorrect.

2. Regardless, this code path cannot even be hit for correct
   TFE/LWE-enabled calls, because those return a struct. Added
   a test case for those for completeness.

Change-Id: I92d314dbc67a2670f6d7adaab765ef45f56a49cf

Reviewers: hliao, dstuttard, arsenm

Subscribers: kzhuravl, jvesely, wdng, yaxunl, tpr, t-tye, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D57681

llvm-svn: 353097

llvm/lib/Transforms/InstCombine/InstCombineInternal.h

@@ -800,8 +800,7 @@ class LLVM_LIBRARY_VISIBILITY InstCombiner
 
   Value *simplifyAMDGCNMemoryIntrinsicDemanded(IntrinsicInst *II,
                                                APInt DemandedElts,
-                                               int DmaskIdx = -1,
-                                               int TFCIdx = -1);
+                                               int DmaskIdx = -1);
 
   Value *SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
                                     APInt &UndefElts, unsigned Depth = 0);

llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp

@@ -966,25 +966,16 @@ InstCombiner::simplifyShrShlDemandedBits(Instruction *Shr, const APInt &ShrOp1,
 }
 
 /// Implement SimplifyDemandedVectorElts for amdgcn buffer and image intrinsics.
+///
+/// Note: This only supports non-TFE/LWE image intrinsic calls; those have
+///       struct returns.
 Value *InstCombiner::simplifyAMDGCNMemoryIntrinsicDemanded(IntrinsicInst *II,
                                                            APInt DemandedElts,
-                                                           int DMaskIdx,
-                                                           int TFCIdx) {
+                                                           int DMaskIdx) {
   unsigned VWidth = II->getType()->getVectorNumElements();
   if (VWidth == 1)
     return nullptr;
 
-  // Need to change to new instruction format
-  bool TFELWEEnabled = false;
-  if (TFCIdx > 0) {
-    if (ConstantInt *TFC = dyn_cast<ConstantInt>(II->getArgOperand(TFCIdx)))
-      TFELWEEnabled =    TFC->getZExtValue() & 0x1  // TFE
-                      || TFC->getZExtValue() & 0x2; // LWE
-  }
-
-  if (TFELWEEnabled)
-    return nullptr; // TFE not yet supported
-
   ConstantInt *NewDMask = nullptr;
 
   if (DMaskIdx < 0) {
@@ -1648,9 +1639,15 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
     case Intrinsic::amdgcn_struct_buffer_load_format:
       return simplifyAMDGCNMemoryIntrinsicDemanded(II, DemandedElts);
     default: {
-      if (getAMDGPUImageDMaskIntrinsic(II->getIntrinsicID()))
-        return simplifyAMDGCNMemoryIntrinsicDemanded(
-            II, DemandedElts, 0, II->getNumArgOperands() - 2);
+      if (getAMDGPUImageDMaskIntrinsic(II->getIntrinsicID())) {
+        LLVM_DEBUG(
+          Value *TFC = II->getArgOperand(II->getNumOperands() - 2);
+          assert(!isa<ConstantInt>(TFC) ||
+                 dyn_cast<ConstantInt>(TFC)->getZExtValue() == 0);
+        );
+
+        return simplifyAMDGCNMemoryIntrinsicDemanded(II, DemandedElts, 0);
+      }
 
       break;
     }

llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts.ll

@@ -2404,6 +2404,21 @@ define protected <4 x half> @__llvm_amdgcn_image_sample_d_1darray_v4f16_f32_f32(
 declare <4 x float> @llvm.amdgcn.image.getresinfo.1d.v4f32.i32(i32, i32, <8 x i32>, i32, i32) #1
 declare <4 x half> @llvm.amdgcn.image.sample.d.1darray.v4f16.f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)
 
+; --------------------------------------------------------------------
+; TFE / LWE
+; --------------------------------------------------------------------
+
+; CHECK-LABEL: @extract_elt0_tfe_image_load_1d_v4f32i32_i32(
+; CHECK-NEXT: %data = call { <4 x float>, i32 } @llvm.amdgcn.image.load.1d.sl_v4f32i32s.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 1)
+define amdgpu_ps float @extract_elt0_tfe_image_load_1d_v4f32i32_i32(i32 %s, <8 x i32> inreg %rsrc) #0 {
+  %data = call { <4 x float>, i32 } @llvm.amdgcn.image.load.1d.sl_v4f32i32s.i32(i32 15, i32 %s, <8 x i32> %rsrc, i32 0, i32 1)
+  %rgba = extractvalue { <4 x float>, i32 } %data, 0
+  %elt0 = extractelement <4 x float> %rgba, i32 0
+  ret float %elt0
+}
+
+declare {<4 x float>, i32} @llvm.amdgcn.image.load.1d.sl_v4f32i32s.i32(i32, i32, <8 x i32>, i32, i32) #1
+
 attributes #0 = { nounwind }
 attributes #1 = { nounwind readonly }