AMDGPU: Support for SimplifyDemandedVectorElts for load intrinsics

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297408 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
index d8824b473a6a27eae2442f5bf0a25ec58f003f18..78ee9d5de3fd1cc7110ce5d9a168baa336a5f05d 100644 (file)
--- a/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
+++ b/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
@@ -1561,6 +1561,47 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
     case Intrinsic::x86_sse4a_insertqi:
       UndefElts |= APInt::getHighBitsSet(VWidth, VWidth / 2);
       break;
+    case Intrinsic::amdgcn_buffer_load:
+    case Intrinsic::amdgcn_buffer_load_format: {
+      if (VWidth == 1 || !APIntOps::isMask(DemandedElts))
+        return nullptr;
+
+      // TODO: Handle 3 vectors when supported in code gen.
+      unsigned NewNumElts = PowerOf2Ceil(DemandedElts.countTrailingOnes());
+      if (NewNumElts == VWidth)
+        return nullptr;
+
+      Module *M = II->getParent()->getParent()->getParent();
+      Type *EltTy = V->getType()->getVectorElementType();
+
+      Type *NewTy = (NewNumElts == 1) ? EltTy :
+        VectorType::get(EltTy, NewNumElts);
+
+      Function *NewIntrin = Intrinsic::getDeclaration(M, II->getIntrinsicID(),
+                                                      NewTy);
+
+      SmallVector<Value *, 5> Args;
+      for (unsigned I = 0, E = II->getNumArgOperands(); I != E; ++I)
+        Args.push_back(II->getArgOperand(I));
+
+      CallInst *NewCall = Builder->CreateCall(NewIntrin, Args);
+      NewCall->takeName(II);
+      NewCall->copyMetadata(*II);
+      if (NewNumElts == 1) {
+        return Builder->CreateInsertElement(UndefValue::get(V->getType()),
+                                            NewCall, static_cast<uint64_t>(0));
+      }
+
+      SmallVector<uint32_t, 8> EltMask;
+      for (unsigned I = 0; I < VWidth; ++I)
+        EltMask.push_back(I);
+
+      Value *Shuffle = Builder->CreateShuffleVector(
+        NewCall, UndefValue::get(NewTy), EltMask);
+
+      MadeChange = true;
+      return Shuffle;
+    }
     }
     break;
   }

diff --git a/test/Transforms/InstCombine/amdgcn-demanded-vector-elts.ll b/test/Transforms/InstCombine/amdgcn-demanded-vector-elts.ll
new file mode 100644 (file)
index 0000000..642f537
--- /dev/null
+++ b/test/Transforms/InstCombine/amdgcn-demanded-vector-elts.ll
@@ -0,0 +1,284 @@
+; RUN: opt -S -instcombine %s | FileCheck %s
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.buffer.load
+; --------------------------------------------------------------------
+
+; CHECK-LABEL: @buffer_load_f32(
+; CHECK-NEXT: %data = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret float %data
+define amdgpu_ps float @buffer_load_f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  ret float %data
+}
+
+; CHECK-LABEL: @buffer_load_v1f32(
+; CHECK-NEXT: %data = call <1 x float> @llvm.amdgcn.buffer.load.v1f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret <1 x float> %data
+define amdgpu_ps <1 x float> @buffer_load_v1f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <1 x float> @llvm.amdgcn.buffer.load.v1f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  ret <1 x float> %data
+}
+
+; CHECK-LABEL: @buffer_load_v2f32(
+; CHECK-NEXT: %data = call <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret <2 x float> %data
+define amdgpu_ps <2 x float> @buffer_load_v2f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  ret <2 x float> %data
+}
+
+; CHECK-LABEL: @buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret <4 x float> %data
+define amdgpu_ps <4 x float> @buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  ret <4 x float> %data
+}
+
+; CHECK-LABEL: @extract_elt0_buffer_load_v2f32(
+; CHECK: %data = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret float %data
+define amdgpu_ps float @extract_elt0_buffer_load_v2f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt0 = extractelement <2 x float> %data, i32 0
+  ret float %elt0
+}
+
+; CHECK-LABEL: @extract_elt1_buffer_load_v2f32(
+; CHECK-NEXT: %data = call <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %elt1 = extractelement <2 x float> %data, i32 1
+; CHECK-NEXT: ret float %elt1
+define amdgpu_ps float @extract_elt1_buffer_load_v2f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt1 = extractelement <2 x float> %data, i32 1
+  ret float %elt1
+}
+
+; CHECK-LABEL: @extract_elt0_buffer_load_v4f32(
+; CHECK-NEXT: %data = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret float %data
+define amdgpu_ps float @extract_elt0_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt0 = extractelement <4 x float> %data, i32 0
+  ret float %elt0
+}
+
+; CHECK-LABEL: @extract_elt1_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %elt1 = extractelement <4 x float> %data, i32 1
+; CHECK-NEXT: ret float %elt1
+define amdgpu_ps float @extract_elt1_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt1 = extractelement <4 x float> %data, i32 1
+  ret float %elt1
+}
+
+; CHECK-LABEL: @extract_elt2_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %elt1 = extractelement <4 x float> %data, i32 2
+; CHECK-NEXT: ret float %elt1
+define amdgpu_ps float @extract_elt2_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt1 = extractelement <4 x float> %data, i32 2
+  ret float %elt1
+}
+
+; CHECK-LABEL: @extract_elt3_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %elt1 = extractelement <4 x float> %data, i32 3
+; CHECK-NEXT: ret float %elt1
+define amdgpu_ps float @extract_elt3_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt1 = extractelement <4 x float> %data, i32 3
+  ret float %elt1
+}
+
+; CHECK-LABEL: @extract_elt0_elt1_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret <2 x float>
+define amdgpu_ps <2 x float> @extract_elt0_elt1_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <4 x float> %data, <4 x float> undef, <2 x i32> <i32 0, i32 1>
+  ret <2 x float> %shuf
+}
+
+; CHECK-LABEL: @extract_elt1_elt2_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %shuf = shufflevector <4 x float> %data, <4 x float> undef, <2 x i32> <i32 1, i32 2>
+; CHECK-NEXT: ret <2 x float> %shuf
+define amdgpu_ps <2 x float> @extract_elt1_elt2_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <4 x float> %data, <4 x float> undef, <2 x i32> <i32 1, i32 2>
+  ret <2 x float> %shuf
+}
+
+; CHECK-LABEL: @extract_elt2_elt3_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %shuf = shufflevector <4 x float> %data, <4 x float> undef, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT: ret <2 x float> %shuf
+define amdgpu_ps <2 x float> @extract_elt2_elt3_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <4 x float> %data, <4 x float> undef, <2 x i32> <i32 2, i32 3>
+  ret <2 x float> %shuf
+}
+
+; CHECK-LABEL: @extract_elt0_elt1_elt2_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %shuf = shufflevector <4 x float> %data, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+; CHECK-NEXT: ret <3 x float> %shuf
+define amdgpu_ps <3 x float> @extract_elt0_elt1_elt2_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <4 x float> %data, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
+  ret <3 x float> %shuf
+}
+
+; CHECK-LABEL: @extract_elt1_elt2_elt3_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %shuf = shufflevector <4 x float> %data, <4 x float> undef, <3 x i32> <i32 1, i32 2, i32 3>
+; CHECK-NEXT: ret <3 x float> %shuf
+define amdgpu_ps <3 x float> @extract_elt1_elt2_elt3_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <4 x float> %data, <4 x float> undef, <3 x i32> <i32 1, i32 2, i32 3>
+  ret <3 x float> %shuf
+}
+
+; CHECK-LABEL: @extract_elt0_elt2_elt3_buffer_load_v4f32(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %shuf = shufflevector <4 x float> %data, <4 x float> undef, <3 x i32> <i32 0, i32 2, i32 3>
+; CHECK-NEXT: ret <3 x float> %shuf
+define amdgpu_ps <3 x float> @extract_elt0_elt2_elt3_buffer_load_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <4 x float> %data, <4 x float> undef, <3 x i32> <i32 0, i32 2, i32 3>
+  ret <3 x float> %shuf
+}
+
+; FIXME: Not handled even though only 2 elts used
+; CHECK-LABEL: @extract_elt0_elt1_buffer_load_v4f32_2(
+; CHECK-NEXT: %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %elt0 = extractelement <4 x float> %data, i32 0
+; CHECK-NEXT: %elt1 = extractelement <4 x float> %data, i32 1
+; CHECK-NEXT: %ins0 = insertvalue { float, float } undef, float %elt0, 0
+; CHECK-NEXT: %ins1 = insertvalue { float, float } %ins0, float %elt1, 1
+; CHECK-NEXT: ret { float, float } %ins1
+define amdgpu_ps { float, float } @extract_elt0_elt1_buffer_load_v4f32_2(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt0 = extractelement <4 x float> %data, i32 0
+  %elt1 = extractelement <4 x float> %data, i32 1
+  %ins0 = insertvalue { float, float } undef, float %elt0, 0
+  %ins1 = insertvalue { float, float } %ins0, float %elt1, 1
+  ret { float, float } %ins1
+}
+
+; CHECK-LABEL: @extract_elt0_buffer_load_v3f32(
+; CHECK-NEXT: %data = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret float %data
+define amdgpu_ps float @extract_elt0_buffer_load_v3f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt0 = extractelement <3 x float> %data, i32 0
+  ret float %elt0
+}
+
+; CHECK-LABEL: @extract_elt1_buffer_load_v3f32(
+; CHECK-NEXT: %data = call <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %elt1 = extractelement <3 x float> %data, i32 1
+; CHECK-NEXT: ret float %elt1
+define amdgpu_ps float @extract_elt1_buffer_load_v3f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt1 = extractelement <3 x float> %data, i32 1
+  ret float %elt1
+}
+
+; CHECK-LABEL: @extract_elt2_buffer_load_v3f32(
+; CHECK-NEXT: %data = call <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %elt1 = extractelement <3 x float> %data, i32 2
+; CHECK-NEXT: ret float %elt1
+define amdgpu_ps float @extract_elt2_buffer_load_v3f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %elt1 = extractelement <3 x float> %data, i32 2
+  ret float %elt1
+}
+
+; CHECK-LABEL: @extract_elt0_elt1_buffer_load_v3f32(
+; CHECK-NEXT: %data = call <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret <2 x float>
+define amdgpu_ps <2 x float> @extract_elt0_elt1_buffer_load_v3f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <3 x float> %data, <3 x float> undef, <2 x i32> <i32 0, i32 1>
+  ret <2 x float> %shuf
+}
+
+; CHECK-LABEL: @extract_elt1_elt2_buffer_load_v3f32(
+; CHECK-NEXT: %data = call <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: %shuf = shufflevector <3 x float> %data, <3 x float> undef, <2 x i32> <i32 1, i32 2>
+; CHECK-NEXT: ret <2 x float> %shuf
+define amdgpu_ps <2 x float> @extract_elt1_elt2_buffer_load_v3f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <3 x float> %data, <3 x float> undef, <2 x i32> <i32 1, i32 2>
+  ret <2 x float> %shuf
+}
+
+; CHECK-LABEL: @preserve_metadata_extract_elt0_buffer_load_v2f32(
+; CHECK-NEXT: %data = call float @llvm.amdgcn.buffer.load.f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false), !fpmath !0
+; CHECK-NEXT: ret float %data
+define amdgpu_ps float @preserve_metadata_extract_elt0_buffer_load_v2f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false), !fpmath !0
+  %elt0 = extractelement <2 x float> %data, i32 0
+  ret float %elt0
+}
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.buffer.load.format
+; --------------------------------------------------------------------
+
+; CHECK-LABEL: @buffer_load_format_v1f32(
+; CHECK-NEXT: %data = call <1 x float> @llvm.amdgcn.buffer.load.format.v1f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 true)
+; CHECK-NEXT: ret <1 x float> %data
+define amdgpu_ps <1 x float> @buffer_load_format_v1f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <1 x float> @llvm.amdgcn.buffer.load.format.v1f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 true)
+  ret <1 x float> %data
+}
+
+; CHECK-LABEL: @extract_elt0_buffer_load_format_v2f32(
+; CHECK-NEXT: %data = call float @llvm.amdgcn.buffer.load.format.f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 true, i1 false)
+; CHECK-NEXT: ret float %data
+define amdgpu_ps float @extract_elt0_buffer_load_format_v2f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <2 x float> @llvm.amdgcn.buffer.load.format.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 true, i1 false)
+  %elt0 = extractelement <2 x float> %data, i32 0
+  ret float %elt0
+}
+
+; CHECK-LABEL: @extract_elt0_elt1_buffer_load_format_v3f32(
+; CHECK-NEXT: %data = call <2 x float> @llvm.amdgcn.buffer.load.format.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret <2 x float> %data
+define amdgpu_ps <2 x float> @extract_elt0_elt1_buffer_load_format_v3f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <3 x float> @llvm.amdgcn.buffer.load.format.v3f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <3 x float> %data, <3 x float> undef, <2 x i32> <i32 0, i32 1>
+  ret <2 x float> %shuf
+}
+
+; CHECK-LABEL: @extract_elt0_elt1_buffer_load_format_v4f32(
+; CHECK-NEXT: %data = call <2 x float> @llvm.amdgcn.buffer.load.format.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+; CHECK-NEXT: ret <2 x float> %data
+define amdgpu_ps <2 x float> @extract_elt0_elt1_buffer_load_format_v4f32(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) #0 {
+  %data = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i1 false, i1 false)
+  %shuf = shufflevector <4 x float> %data, <4 x float> undef, <2 x i32> <i32 0, i32 1>
+  ret <2 x float> %shuf
+}
+
+declare float @llvm.amdgcn.buffer.load.f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <1 x float> @llvm.amdgcn.buffer.load.v1f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <2 x float> @llvm.amdgcn.buffer.load.v2f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <3 x float> @llvm.amdgcn.buffer.load.v3f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <4 x float> @llvm.amdgcn.buffer.load.v4f32(<4 x i32>, i32, i32, i1, i1) #1
+
+declare float @llvm.amdgcn.buffer.load.format.f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <1 x float> @llvm.amdgcn.buffer.load.format.v1f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <2 x float> @llvm.amdgcn.buffer.load.format.v2f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <3 x float> @llvm.amdgcn.buffer.load.format.v3f32(<4 x i32>, i32, i32, i1, i1) #1
+declare <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32>, i32, i32, i1, i1) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readonly }
+
+!0 = !{float 2.500000e+00}