Fixed some bugs in register stack pass.
[llvm/zpu.git] / test / CodeGen / X86 / 2009-08-14-Win64MemoryIndirectArg.ll
blob6b0d6d9790de4ee3ca0b292555bd5151784b98de
1 ; RUN: llc < %s
2 target triple = "x86_64-mingw"
4 ; ModuleID = 'mm.bc'
5         type opaque             ; type %0
6         type opaque             ; type %1
8 define internal fastcc float @computeMipmappingRho(%0* %shaderExecutionStatePtr, i32 %index, <4 x float> %texCoord, <4 x float> %texCoordDX, <4 x float> %texCoordDY) readonly {
9 indexCheckBlock:
10         %indexCmp = icmp ugt i32 %index, 16             ; <i1> [#uses=1]
11         br i1 %indexCmp, label %zeroReturnBlock, label %primitiveTextureFetchBlock
13 primitiveTextureFetchBlock:             ; preds = %indexCheckBlock
14         %pointerArithmeticTmp = bitcast %0* %shaderExecutionStatePtr to i8*             ; <i8*> [#uses=1]
15         %pointerArithmeticTmp1 = getelementptr i8* %pointerArithmeticTmp, i64 1808              ; <i8*> [#uses=1]
16         %pointerArithmeticTmp2 = bitcast i8* %pointerArithmeticTmp1 to %1**             ; <%1**> [#uses=1]
17         %primitivePtr = load %1** %pointerArithmeticTmp2                ; <%1*> [#uses=1]
18         %pointerArithmeticTmp3 = bitcast %1* %primitivePtr to i8*               ; <i8*> [#uses=1]
19         %pointerArithmeticTmp4 = getelementptr i8* %pointerArithmeticTmp3, i64 19408            ; <i8*> [#uses=1]
20         %pointerArithmeticTmp5 = bitcast i8* %pointerArithmeticTmp4 to %1**             ; <%1**> [#uses=1]
21         %primitiveTexturePtr = getelementptr %1** %pointerArithmeticTmp5, i32 %index            ; <%1**> [#uses=1]
22         %primitiveTexturePtr6 = load %1** %primitiveTexturePtr          ; <%1*> [#uses=2]
23         br label %textureCheckBlock
25 textureCheckBlock:              ; preds = %primitiveTextureFetchBlock
26         %texturePtrInt = ptrtoint %1* %primitiveTexturePtr6 to i64              ; <i64> [#uses=1]
27         %testTextureNULL = icmp eq i64 %texturePtrInt, 0                ; <i1> [#uses=1]
28         br i1 %testTextureNULL, label %zeroReturnBlock, label %rhoCalculateBlock
30 rhoCalculateBlock:              ; preds = %textureCheckBlock
31         %pointerArithmeticTmp7 = bitcast %1* %primitiveTexturePtr6 to i8*               ; <i8*> [#uses=1]
32         %pointerArithmeticTmp8 = getelementptr i8* %pointerArithmeticTmp7, i64 640              ; <i8*> [#uses=1]
33         %pointerArithmeticTmp9 = bitcast i8* %pointerArithmeticTmp8 to <4 x float>*             ; <<4 x float>*> [#uses=1]
34         %dimensionsPtr = load <4 x float>* %pointerArithmeticTmp9, align 1              ; <<4 x float>> [#uses=2]
35         %texDiffDX = fsub <4 x float> %texCoordDX, %texCoord            ; <<4 x float>> [#uses=1]
36         %texDiffDY = fsub <4 x float> %texCoordDY, %texCoord            ; <<4 x float>> [#uses=1]
37         %ddx = fmul <4 x float> %texDiffDX, %dimensionsPtr              ; <<4 x float>> [#uses=2]
38         %ddx10 = fmul <4 x float> %texDiffDY, %dimensionsPtr            ; <<4 x float>> [#uses=2]
39         %ddxSquared = fmul <4 x float> %ddx, %ddx               ; <<4 x float>> [#uses=3]
40         %0 = shufflevector <4 x float> %ddxSquared, <4 x float> %ddxSquared, <4 x i32> <i32 1, i32 0, i32 0, i32 0>             ; <<4 x float>> [#uses=1]
41         %dxSquared = fadd <4 x float> %ddxSquared, %0           ; <<4 x float>> [#uses=1]
42         %1 = call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> %dxSquared)             ; <<4 x float>> [#uses=1]
43         %ddySquared = fmul <4 x float> %ddx10, %ddx10           ; <<4 x float>> [#uses=3]
44         %2 = shufflevector <4 x float> %ddySquared, <4 x float> %ddySquared, <4 x i32> <i32 1, i32 0, i32 0, i32 0>             ; <<4 x float>> [#uses=1]
45         %dySquared = fadd <4 x float> %ddySquared, %2           ; <<4 x float>> [#uses=1]
46         %3 = call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> %dySquared)             ; <<4 x float>> [#uses=1]
47         %4 = call <4 x float> @llvm.x86.sse.max.ss(<4 x float> %1, <4 x float> %3)              ; <<4 x float>> [#uses=1]
48         %rho = extractelement <4 x float> %4, i32 0             ; <float> [#uses=1]
49         ret float %rho
51 zeroReturnBlock:                ; preds = %textureCheckBlock, %indexCheckBlock
52         ret float 0.000000e+00
55 declare <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float>) nounwind readnone
57 declare <4 x float> @llvm.x86.sse.max.ss(<4 x float>, <4 x float>) nounwind readnone