[llvm] r175663 - Additional fixes for bug 15155.
Anna Zaks
ganna at apple.com
Wed Feb 20 15:16:21 PST 2013
On Feb 20, 2013, at 3:05 PM, Hal Finkel <hfinkel at anl.gov> wrote:
> ----- Original Message -----
>> From: "Anna Zaks" <ganna at apple.com>
>> To: "Bill Schmidt" <wschmidt at linux.vnet.ibm.com>
>> Cc: llvm-commits at cs.uiuc.edu, "benny kra" <benny.kra at gmail.com>
>> Sent: Wednesday, February 20, 2013 5:01:14 PM
>> Subject: Re: [llvm] r175663 - Additional fixes for bug 15155.
>>
>>
>> Bill,
>>
>>
>> Please, revert the change until a solution is found as this issue is
>> blocking other testers.
>
> Did r175683 fix this?
>
Thanks!
This did suppress the failure, at least on my local build.
Anna.
> -Hal
>
>>
>>
>> Thanks,
>> Anna.
>>
>>
>> On Feb 20, 2013, at 2:52 PM, Anna Zaks < ganna at apple.com > wrote:
>>
>>
>>
>> This is now failing on a public builder:
>> http://lab.llvm.org:8013/builders/clang-x86_64-darwin11-nobootstrap-RA/builds/1755
>>
>>
>> Have you tried to repo this with Release+Asserts?
>>
>>
>> Please, let me know what exactly you'd like to see...
>>
>>
>>
>> ../build.clang-x86_64-darwin10-nobootstrap-RA/clang-build/Release+Asserts/bin/llc
>> <
>> ../build.clang-x86_64-darwin10-nobootstrap-RA/llvm/test/CodeGen/PowerPC/vec_constants.ll
>> -march=ppc32 -mcpu=g5
>> .machine ppc970
>> .section __TEXT,__textcoal_nt,coalesced,pure_instructions
>> .section __TEXT,__symbol_stub1,symbol_stubs,pure_instructions,16
>> .section __TEXT,__text,regular,pure_instructions
>> .globl _test1
>> .align 4
>> _test1: ; @test1
>> ; BB#0:
>> vspltisb v2, -1
>> mfspr r6, 256
>> vslw v2, v2, v2
>> oris r7, r6, 12288
>> mtspr 256, r7
>> lvx v3, 0, r3
>> vand v3, v3, v2
>> stvx v3, 0, r3
>> lvx v3, 0, r4
>> vandc v3, v3, v2
>> stvx v3, 0, r4
>> lvx v3, 0, r5
>> vandc v2, v3, v2
>> stvx v2, 0, r5
>> mtspr 256, r6
>> blr
>>
>>
>> .globl _test_30
>> .align 4
>> _test_30: ; @test_30
>> ; BB#0:
>> vspltisw v2, 15
>> vadduwm v2, v2, v2
>> blr
>>
>>
>> .globl _test_29
>> .align 4
>> _test_29: ; @test_29
>> ; BB#0:
>> vspltisw v2, -16
>> vspltisw v3, 13
>> mfspr r3, 256
>> vsubuwm v2, v3, v2
>> oris r4, r3, 4096
>> mtspr 256, r4
>> mtspr 256, r3
>> blr
>>
>>
>> .globl _test_n30
>> .align 4
>> _test_n30: ; @test_n30
>> ; BB#0:
>> vspltish v2, -15
>> vadduhm v2, v2, v2
>> blr
>>
>>
>> .globl _test_n104
>> .align 4
>> _test_n104: ; @test_n104
>> ; BB#0:
>> vspltisb v2, -13
>> vslb v2, v2, v2
>> blr
>>
>>
>> .globl _test_vsldoi
>> .align 4
>> _test_vsldoi: ; @test_vsldoi
>> ; BB#0:
>> vspltisw v2, 2
>> vsldoi v2, v2, v2, 1
>> blr
>>
>>
>> .globl _test_vsldoi_65023
>> .align 4
>> _test_vsldoi_65023: ; @test_vsldoi_65023
>> ; BB#0:
>> vspltish v2, -3
>> vsldoi v2, v2, v2, 1
>> blr
>>
>>
>> .globl _test_rol
>> .align 4
>> _test_rol: ; @test_rol
>> ; BB#0:
>> vspltisw v2, -12
>> vrlw v2, v2, v2
>> blr
>>
>>
>> .subsections_via_symbols
>>
>>
>> smoosh-12:test buildslave$
>>
>>
>> Anna.
>>
>>
>> On Feb 20, 2013, at 1:39 PM, Bill Schmidt <
>> wschmidt at linux.vnet.ibm.com > wrote:
>>
>>
>> On Wed, 2013-02-20 at 13:21 -0800, Anna Zaks wrote:
>>
>>
>> This commit introduced a failure in one of our internal bots.
>> (Possibly, the external ones haven't picked up this change yet.
>> Please, let me know if you need help reproducing the failure.)
>>
>> Hm. It would be helpful to see the assembly output produced on your
>> machine. This test is passing for me and I see that it passed for
>> http://bb.pgr.jp/builders/cmake-llvm-x86_64-linux/builds/2992 . This
>> seems to indicate that you are seeing an extra instruction
>> interposed,
>> and I can't guess what that would be.
>>
>> I just took the XFAIL off this test because it started working for me
>> when I restored these optimizations. From what I can tell, Ben
>> originally XFAIL'd it when the constant folding broke them. Ben, is
>> there anything else I need to do besides removing the XFAIL?
>>
>>
>>
>> Thanks,
>> Bill
>>
>>
>>
>> Cheers,
>> Anna.
>> ******************** TEST 'LLVM :: CodeGen/PowerPC/vec_constants.ll'
>> FAILED ********************
>> Script:
>> --
>> ../build.clang-x86_64-darwin10-nobootstrap-RA/clang-build/Release+Asserts/bin/llc
>> < ../llvm/test/CodeGen/PowerPC/vec_constants.ll -march=ppc32
>> -mcpu=g5 |
>> ../build.clang-x86_64-darwin10-nobootstrap-RA/clang-build/Release+Asserts/bin/FileCheck
>> ../build.clang-x86_64-darwin10-nobootstrap-RA/llvm/test/CodeGen/PowerPC/vec_constants.ll
>> --
>> Exit Code: 1
>> Command Output (stderr):
>> --
>> ../build.clang-x86_64-darwin10-nobootstrap-RA/llvm/test/CodeGen/PowerPC/vec_constants.ll:36:15:
>> error: CHECK-NEXT: is not on the line after the previous match
>> ; CHECK-NEXT: vsubuwm
>> ^
>> <stdin>:41:2: note: 'next' match was here
>> vsubuwm v2, v3, v2
>> ^
>> <stdin>:39:10: note: previous match was here
>> vspltisw v3, 13
>> ^
>> --
>>
>>
>> On Feb 20, 2013, at 12:41 PM, Bill Schmidt
>> < wschmidt at linux.vnet.ibm.com > wrote:
>>
>>
>>
>> Author: wschmidt
>> Date: Wed Feb 20 14:41:42 2013
>> New Revision: 175663
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=175663&view=rev
>> Log:
>> Additional fixes for bug 15155.
>>
>> This handles the cases where the 6-bit splat element is odd,
>> converting
>> to a three-instruction sequence to add or subtract two splats. With
>> this
>> fix, the XFAIL in test/CodeGen/PowerPC/vec_constants.ll is removed.
>>
>> Modified:
>> llvm/trunk/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
>> llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp
>> llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h
>> llvm/trunk/test/CodeGen/PowerPC/vaddsplat.ll
>> llvm/trunk/test/CodeGen/PowerPC/vec_constants.ll
>>
>> Modified: llvm/trunk/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCISelDAGToDAG.cpp?rev=175663&r1=175662&r2=175663&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/Target/PowerPC/PPCISelDAGToDAG.cpp (original)
>> +++ llvm/trunk/lib/Target/PowerPC/PPCISelDAGToDAG.cpp Wed Feb 20
>> 14:41:42 2013
>> @@ -1323,34 +1323,75 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *
>> SDValue(Tmp, 0), GA);
>> }
>> case PPCISD::VADD_SPLAT: {
>> - // Convert: VADD_SPLAT elt, size
>> - // Into: tmp = VSPLTIS[BHW] elt
>> - // VADDU[BHW]M tmp, tmp
>> - // Where: [BHW] = B for size = 1, H for size = 2, W for size
>> = 4
>> + // This expands into one of three sequences, depending on
>> whether
>> + // the first operand is odd or even, positive or negative.
>> assert(isa<ConstantSDNode>(N->getOperand(0)) &&
>> isa<ConstantSDNode>(N->getOperand(1)) &&
>> "Invalid operand on VADD_SPLAT!");
>> +
>> + int Elt = N->getConstantOperandVal(0);
>> int EltSize = N->getConstantOperandVal(1);
>> - unsigned Opc1, Opc2;
>> + unsigned Opc1, Opc2, Opc3;
>> EVT VT;
>> +
>> if (EltSize == 1) {
>> Opc1 = PPC::VSPLTISB;
>> Opc2 = PPC::VADDUBM;
>> + Opc3 = PPC::VSUBUBM;
>> VT = MVT::v16i8;
>> } else if (EltSize == 2) {
>> Opc1 = PPC::VSPLTISH;
>> Opc2 = PPC::VADDUHM;
>> + Opc3 = PPC::VSUBUHM;
>> VT = MVT::v8i16;
>> } else {
>> assert(EltSize == 4 && "Invalid element size on VADD_SPLAT!");
>> Opc1 = PPC::VSPLTISW;
>> Opc2 = PPC::VADDUWM;
>> + Opc3 = PPC::VSUBUWM;
>> VT = MVT::v4i32;
>> }
>> - SDValue Elt = getI32Imm(N->getConstantOperandVal(0));
>> - SDNode *Tmp = CurDAG->getMachineNode(Opc1, dl, VT, Elt);
>> - SDValue TmpVal = SDValue(Tmp, 0);
>> - return CurDAG->getMachineNode(Opc2, dl, VT, TmpVal, TmpVal);
>> +
>> + if ((Elt & 1) == 0) {
>> + // Elt is even, in the range [-32,-18] + [16,30].
>> + //
>> + // Convert: VADD_SPLAT elt, size
>> + // Into: tmp = VSPLTIS[BHW] elt
>> + // VADDU[BHW]M tmp, tmp
>> + // Where: [BHW] = B for size = 1, H for size = 2, W for
>> size = 4
>> + SDValue EltVal = getI32Imm(Elt >> 1);
>> + SDNode *Tmp = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
>> + SDValue TmpVal = SDValue(Tmp, 0);
>> + return CurDAG->getMachineNode(Opc2, dl, VT, TmpVal, TmpVal);
>> +
>> + } else if (Elt > 0) {
>> + // Elt is odd and positive, in the range [17,31].
>> + //
>> + // Convert: VADD_SPLAT elt, size
>> + // Into: tmp1 = VSPLTIS[BHW] elt-16
>> + // tmp2 = VSPLTIS[BHW] -16
>> + // VSUBU[BHW]M tmp1, tmp2
>> + SDValue EltVal = getI32Imm(Elt - 16);
>> + SDNode *Tmp1 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
>> + EltVal = getI32Imm(-16);
>> + SDNode *Tmp2 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
>> + return CurDAG->getMachineNode(Opc3, dl, VT, SDValue(Tmp1, 0),
>> + SDValue(Tmp2, 0));
>> +
>> + } else {
>> + // Elt is odd and negative, in the range [-31,-17].
>> + //
>> + // Convert: VADD_SPLAT elt, size
>> + // Into: tmp1 = VSPLTIS[BHW] elt+16
>> + // tmp2 = VSPLTIS[BHW] -16
>> + // VADDU[BHW]M tmp1, tmp2
>> + SDValue EltVal = getI32Imm(Elt + 16);
>> + SDNode *Tmp1 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
>> + EltVal = getI32Imm(-16);
>> + SDNode *Tmp2 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal);
>> + return CurDAG->getMachineNode(Opc2, dl, VT, SDValue(Tmp1, 0),
>> + SDValue(Tmp2, 0));
>> + }
>> }
>> }
>>
>>
>> Modified: llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp?rev=175663&r1=175662&r2=175663&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp (original)
>> +++ llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp Wed Feb 20
>> 14:41:42 2013
>> @@ -5025,11 +5025,17 @@ SDValue PPCTargetLowering::LowerBUILD_VE
>> // Two instruction sequences.
>>
>> // If this value is in the range [-32,30] and is even, use:
>> - // tmp = VSPLTI[bhw], result = add tmp, tmp
>> - if (SextVal >= -32 && SextVal <= 30 && (SextVal & 1) == 0) {
>> - // To avoid having the optimization undone by constant folding,
>> we
>> - // convert to a pseudo that will be expanded later.
>> - SDValue Elt = DAG.getConstant(SextVal >> 1, MVT::i32);
>> + // VSPLTI[bhw](val/2) + VSPLTI[bhw](val/2)
>> + // If this value is in the range [17,31] and is odd, use:
>> + // VSPLTI[bhw](val-16) - VSPLTI[bhw](-16)
>> + // If this value is in the range [-31,-17] and is odd, use:
>> + // VSPLTI[bhw](val+16) + VSPLTI[bhw](-16)
>> + // Note the last two are three-instruction sequences.
>> + if (SextVal >= -32 && SextVal <= 31) {
>> + // To avoid having these optimizations undone by constant
>> folding,
>> + // we convert to a pseudo that will be expanded later into one
>> of
>> + // the above forms.
>> + SDValue Elt = DAG.getConstant(SextVal, MVT::i32);
>> EVT VT = Op.getValueType();
>> int Size = VT == MVT::v16i8 ? 1 : (VT == MVT::v8i16 ? 2 : 4);
>> SDValue EltSize = DAG.getConstant(Size, MVT::i32);
>> @@ -5129,25 +5135,6 @@ SDValue PPCTargetLowering::LowerBUILD_VE
>> }
>> }
>>
>> - // Three instruction sequences.
>> -
>> - // Odd, in range [17,31]: (vsplti C)-(vsplti -16).
>> - // FIXME: Disabled because the add gets constant folded.
>> - if (0 && SextVal >= 0 && SextVal <= 31) {
>> - SDValue LHS = BuildSplatI(SextVal-16, SplatSize, MVT::Other,
>> DAG, dl);
>> - SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
>> - LHS = DAG.getNode(ISD::SUB, dl, LHS.getValueType(), LHS, RHS);
>> - return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), LHS);
>> - }
>> - // Odd, in range [-31,-17]: (vsplti C)+(vsplti -16).
>> - // FIXME: Disabled because the add gets constant folded.
>> - if (0 && SextVal >= -31 && SextVal <= 0) {
>> - SDValue LHS = BuildSplatI(SextVal+16, SplatSize, MVT::Other,
>> DAG, dl);
>> - SDValue RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG, dl);
>> - LHS = DAG.getNode(ISD::ADD, dl, LHS.getValueType(), LHS, RHS);
>> - return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), LHS);
>> - }
>> -
>> return SDValue();
>> }
>>
>>
>> Modified: llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h?rev=175663&r1=175662&r2=175663&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h (original)
>> +++ llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h Wed Feb 20
>> 14:41:42 2013
>> @@ -238,8 +238,9 @@ namespace llvm {
>> ADDI_DTPREL_L,
>>
>> /// VRRC = VADD_SPLAT Elt, EltSize - Temporary node to be
>> expanded
>> - /// into an ADD of a VSPLTI with itself during instruction
>> selection.
>> - /// Necessary to avoid losing this optimization due to
>> constant folds.
>> + /// during instruction selection to optimize a BUILD_VECTOR
>> into
>> + /// operations on splats. This is necessary to avoid losing
>> these
>> + /// optimizations due to constant folding.
>> VADD_SPLAT,
>>
>> /// STD_32 - This is the STD instruction for use with "32-bit"
>> registers.
>>
>> Modified: llvm/trunk/test/CodeGen/PowerPC/vaddsplat.ll
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/vaddsplat.ll?rev=175663&r1=175662&r2=175663&view=diff
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/PowerPC/vaddsplat.ll (original)
>> +++ llvm/trunk/test/CodeGen/PowerPC/vaddsplat.ll Wed Feb 20 14:41:42
>> 2013
>> @@ -1,6 +1,6 @@
>> ; RUN: llc -O0 -mcpu=pwr7 <%s | FileCheck %s
>>
>> -; Test optimization of build_vector into vadd/vsplt for 6-bit
>> immediates.
>> +; Test optimizations of build_vector for 6-bit immediates.
>>
>> target datalayout =
>> "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
>> target triple = "powerpc64-unknown-linux-gnu"
>> @@ -9,69 +9,141 @@ target triple = "powerpc64-unknown-linux
>> %v8i16 = type <8 x i16>
>> %v16i8 = type <16 x i8>
>>
>> -define void @test_v4i32_pos(%v4i32* %P, %v4i32* %S) {
>> +define void @test_v4i32_pos_even(%v4i32* %P, %v4i32* %S) {
>> %p = load %v4i32* %P
>> %r = add %v4i32 %p, < i32 18, i32 18, i32 18, i32 18 >
>> store %v4i32 %r, %v4i32* %S
>> ret void
>> }
>>
>> -; CHECK: test_v4i32_pos:
>> +; CHECK: test_v4i32_pos_even:
>> ; CHECK: vspltisw [[REG1:[0-9]+]], 9
>> ; CHECK: vadduwm {{[0-9]+}}, [[REG1]], [[REG1]]
>>
>> -define void @test_v4i32_neg(%v4i32* %P, %v4i32* %S) {
>> +define void @test_v4i32_neg_even(%v4i32* %P, %v4i32* %S) {
>> %p = load %v4i32* %P
>> %r = add %v4i32 %p, < i32 -28, i32 -28, i32 -28, i32 -28 >
>> store %v4i32 %r, %v4i32* %S
>> ret void
>> }
>>
>> -; CHECK: test_v4i32_neg:
>> +; CHECK: test_v4i32_neg_even:
>> ; CHECK: vspltisw [[REG1:[0-9]+]], -14
>> ; CHECK: vadduwm {{[0-9]+}}, [[REG1]], [[REG1]]
>>
>> -define void @test_v8i16_pos(%v8i16* %P, %v8i16* %S) {
>> +define void @test_v8i16_pos_even(%v8i16* %P, %v8i16* %S) {
>> %p = load %v8i16* %P
>> %r = add %v8i16 %p, < i16 30, i16 30, i16 30, i16 30, i16 30,
>> i16 30, i16 30, i16 30 >
>> store %v8i16 %r, %v8i16* %S
>> ret void
>> }
>>
>> -; CHECK: test_v8i16_pos:
>> +; CHECK: test_v8i16_pos_even:
>> ; CHECK: vspltish [[REG1:[0-9]+]], 15
>> ; CHECK: vadduhm {{[0-9]+}}, [[REG1]], [[REG1]]
>>
>> -define void @test_v8i16_neg(%v8i16* %P, %v8i16* %S) {
>> +define void @test_v8i16_neg_even(%v8i16* %P, %v8i16* %S) {
>> %p = load %v8i16* %P
>> %r = add %v8i16 %p, < i16 -32, i16 -32, i16 -32, i16 -32, i16
>> -32, i16 -32, i16 -32, i16 -32 >
>> store %v8i16 %r, %v8i16* %S
>> ret void
>> }
>>
>> -; CHECK: test_v8i16_neg:
>> +; CHECK: test_v8i16_neg_even:
>> ; CHECK: vspltish [[REG1:[0-9]+]], -16
>> ; CHECK: vadduhm {{[0-9]+}}, [[REG1]], [[REG1]]
>>
>> -define void @test_v16i8_pos(%v16i8* %P, %v16i8* %S) {
>> +define void @test_v16i8_pos_even(%v16i8* %P, %v16i8* %S) {
>> %p = load %v16i8* %P
>> %r = add %v16i8 %p, < i8 16, i8 16, i8 16, i8 16, i8 16, i8
>> 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16,
>> i8 16 >
>> store %v16i8 %r, %v16i8* %S
>> ret void
>> }
>>
>> -; CHECK: test_v16i8_pos:
>> +; CHECK: test_v16i8_pos_even:
>> ; CHECK: vspltisb [[REG1:[0-9]+]], 8
>> ; CHECK: vaddubm {{[0-9]+}}, [[REG1]], [[REG1]]
>>
>> -define void @test_v16i8_neg(%v16i8* %P, %v16i8* %S) {
>> +define void @test_v16i8_neg_even(%v16i8* %P, %v16i8* %S) {
>> %p = load %v16i8* %P
>> %r = add %v16i8 %p, < i8 -18, i8 -18, i8 -18, i8 -18, i8 -18,
>> i8 -18, i8 -18, i8 -18, i8 -18, i8 -18, i8 -18, i8 -18, i8 -18, i8
>> -18, i8 -18, i8 -18 >
>> store %v16i8 %r, %v16i8* %S
>> ret void
>> }
>>
>> -; CHECK: test_v16i8_neg:
>> +; CHECK: test_v16i8_neg_even:
>> ; CHECK: vspltisb [[REG1:[0-9]+]], -9
>> ; CHECK: vaddubm {{[0-9]+}}, [[REG1]], [[REG1]]
>>
>> +define void @test_v4i32_pos_odd(%v4i32* %P, %v4i32* %S) {
>> + %p = load %v4i32* %P
>> + %r = add %v4i32 %p, < i32 27, i32 27, i32 27, i32 27 >
>> + store %v4i32 %r, %v4i32* %S
>> + ret void
>> +}
>> +
>> +; CHECK: test_v4i32_pos_odd:
>> +; CHECK: vspltisw [[REG2:[0-9]+]], -16
>> +; CHECK: vspltisw [[REG1:[0-9]+]], 11
>> +; CHECK: vsubuwm {{[0-9]+}}, [[REG1]], [[REG2]]
>> +
>> +define void @test_v4i32_neg_odd(%v4i32* %P, %v4i32* %S) {
>> + %p = load %v4i32* %P
>> + %r = add %v4i32 %p, < i32 -27, i32 -27, i32 -27, i32 -27 >
>> + store %v4i32 %r, %v4i32* %S
>> + ret void
>> +}
>> +
>> +; CHECK: test_v4i32_neg_odd:
>> +; CHECK: vspltisw [[REG2:[0-9]+]], -16
>> +; CHECK: vspltisw [[REG1:[0-9]+]], -11
>> +; CHECK: vadduwm {{[0-9]+}}, [[REG1]], [[REG2]]
>> +
>> +define void @test_v8i16_pos_odd(%v8i16* %P, %v8i16* %S) {
>> + %p = load %v8i16* %P
>> + %r = add %v8i16 %p, < i16 31, i16 31, i16 31, i16 31, i16
>> 31, i16 31, i16 31, i16 31 >
>> + store %v8i16 %r, %v8i16* %S
>> + ret void
>> +}
>> +
>> +; CHECK: test_v8i16_pos_odd:
>> +; CHECK: vspltish [[REG2:[0-9]+]], -16
>> +; CHECK: vspltish [[REG1:[0-9]+]], 15
>> +; CHECK: vsubuhm {{[0-9]+}}, [[REG1]], [[REG2]]
>> +
>> +define void @test_v8i16_neg_odd(%v8i16* %P, %v8i16* %S) {
>> + %p = load %v8i16* %P
>> + %r = add %v8i16 %p, < i16 -31, i16 -31, i16 -31, i16 -31,
>> i16 -31, i16 -31, i16 -31, i16 -31 >
>> + store %v8i16 %r, %v8i16* %S
>> + ret void
>> +}
>> +
>> +; CHECK: test_v8i16_neg_odd:
>> +; CHECK: vspltish [[REG2:[0-9]+]], -16
>> +; CHECK: vspltish [[REG1:[0-9]+]], -15
>> +; CHECK: vadduhm {{[0-9]+}}, [[REG1]], [[REG2]]
>> +
>> +define void @test_v16i8_pos_odd(%v16i8* %P, %v16i8* %S) {
>> + %p = load %v16i8* %P
>> + %r = add %v16i8 %p, < i8 17, i8 17, i8 17, i8 17, i8 17, i8
>> 17, i8 17, i8 17, i8 17, i8 17, i8 17, i8 17, i8 17, i8 17, i8 17,
>> i8 17 >
>> + store %v16i8 %r, %v16i8* %S
>> + ret void
>> +}
>> +
>> +; CHECK: test_v16i8_pos_odd:
>> +; CHECK: vspltisb [[REG2:[0-9]+]], -16
>> +; CHECK: vspltisb [[REG1:[0-9]+]], 1
>> +; CHECK: vsububm {{[0-9]+}}, [[REG1]], [[REG2]]
>> +
>> +define void @test_v16i8_neg_odd(%v16i8* %P, %v16i8* %S) {
>> + %p = load %v16i8* %P
>> + %r = add %v16i8 %p, < i8 -17, i8 -17, i8 -17, i8 -17, i8
>> -17, i8 -17, i8 -17, i8 -17, i8 -17, i8 -17, i8 -17, i8 -17, i8 -17,
>> i8 -17, i8 -17, i8 -17 >
>> + store %v16i8 %r, %v16i8* %S
>> + ret void
>> +}
>> +
>> +; CHECK: test_v16i8_neg_odd:
>> +; CHECK: vspltisb [[REG2:[0-9]+]], -16
>> +; CHECK: vspltisb [[REG1:[0-9]+]], -1
>> +; CHECK: vaddubm {{[0-9]+}}, [[REG1]], [[REG2]]
>> +
>>
>> Modified: llvm/trunk/test/CodeGen/PowerPC/vec_constants.ll
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/vec_constants.ll?rev=175663&r1=175662&r2=175663&view=diff
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/PowerPC/vec_constants.ll (original)
>> +++ llvm/trunk/test/CodeGen/PowerPC/vec_constants.ll Wed Feb 20
>> 14:41:42 2013
>> @@ -1,5 +1,4 @@
>> ; RUN: llc < %s -march=ppc32 -mcpu=g5 | FileCheck %s
>> -; XFAIL: *
>>
>> define void @test1(<4 x i32>* %P1, <4 x i32>* %P2, <4 x float>* %P3)
>> nounwind {
>> %tmp = load <4 x i32>* %P1 ; <<4 x i32>> [#uses=1]
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>>
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
More information about the llvm-commits
mailing list