R600: PV stores Reg id, not index

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184117 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/R600InstrInfo.cpp b/lib/Target/R600/R600InstrInfo.cpp
index 4f5cfcd732ccc25253b9bbcd2d9287d319026dee..018583dd52d7b52b4a3543b7a50ee4f84a964640 100644 (file)
--- a/lib/Target/R600/R600InstrInfo.cpp
+++ b/lib/Target/R600/R600InstrInfo.cpp
@@ -239,7 +239,7 @@ R600InstrInfo::ExtractSrcs(MachineInstr *MI,
       Result.push_back(DummyPair);
       continue;
     }
-    if (PV.find(Index) != PV.end()) {
+    if (PV.find(Reg) != PV.end()) {
       Result.push_back(DummyPair);
       continue;
     }

diff --git a/test/CodeGen/R600/pv-packing.ll b/test/CodeGen/R600/pv-packing.ll
new file mode 100644 (file)
index 0000000..03fc204
--- /dev/null
+++ b/test/CodeGen/R600/pv-packing.ll
@@ -0,0 +1,50 @@
+; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s
+
+;CHECK: DOT4  T{{[0-9]\.X}}
+;CHECK: MULADD_IEEE * T{{[0-9]\.W}}
+
+define void @main() #0 {
+main_body:
+  %0 = call float @llvm.R600.load.input(i32 4)
+  %1 = call float @llvm.R600.load.input(i32 5)
+  %2 = call float @llvm.R600.load.input(i32 6)
+  %3 = call float @llvm.R600.load.input(i32 8)
+  %4 = call float @llvm.R600.load.input(i32 9)
+  %5 = call float @llvm.R600.load.input(i32 10)
+  %6 = call float @llvm.R600.load.input(i32 12)
+  %7 = call float @llvm.R600.load.input(i32 13)
+  %8 = call float @llvm.R600.load.input(i32 14)
+  %9 = load <4 x float> addrspace(8)* null
+  %10 = load <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>] addrspace(8)* null, i64 0, i32 1)
+  %11 = call float @llvm.AMDGPU.dp4(<4 x float> %9, <4 x float> %9)
+  %12 = fmul float %0, %3
+  %13 = fadd float %12, %6
+  %14 = fmul float %1, %4
+  %15 = fadd float %14, %7
+  %16 = fmul float %2, %5
+  %17 = fadd float %16, %8
+  %18 = fmul float %11, %11
+  %19 = fadd float %18, %0
+  %20 = insertelement <4 x float> undef, float %13, i32 0
+  %21 = insertelement <4 x float> %20, float %15, i32 1
+  %22 = insertelement <4 x float> %21, float %17, i32 2
+  %23 = insertelement <4 x float> %22, float %19, i32 3
+  %24 = call float @llvm.AMDGPU.dp4(<4 x float> %23, <4 x float> %10)
+  %25 = insertelement <4 x float> undef, float %24, i32 0
+  call void @llvm.R600.store.swizzle(<4 x float> %25, i32 0, i32 2)
+  ret void
+}
+
+; Function Attrs: readnone
+declare float @llvm.R600.load.input(i32) #1
+
+; Function Attrs: readnone
+declare float @llvm.AMDGPU.dp4(<4 x float>, <4 x float>) #1
+
+
+declare void @llvm.R600.store.swizzle(<4 x float>, i32, i32)
+
+attributes #0 = { "ShaderType"="1" }
+attributes #1 = { readnone }
+attributes #2 = { readonly }
+attributes #3 = { nounwind readonly }