[llvm] r319778 - [x86][AVX512] Lowering kunpack intrinsics to LLVM IR
Bill Seurer via llvm-commits
llvm-commits at lists.llvm.org
Wed Dec 6 11:35:26 PST 2017
Thanks, looks like it is working now.
On 12/06/2017 11:56 AM, Craig Topper wrote:
> I've attempted a fix for at least the ubsan failure in r319911
>
> ~Craig
>
> On Wed, Dec 6, 2017 at 9:49 AM, Bill Seurer via llvm-commits
> <llvm-commits at lists.llvm.org <mailto:llvm-commits at lists.llvm.org>> wrote:
>
> This is still causing failures on all the powerpc64 buildbots:
>
> ******************** TEST 'LLVM ::
> CodeGen/X86/avx512bw-intrinsics-upgrade.ll' FAILED ********************
> Script:
> --
> /home/buildbots/ppc64be-clang-test/clang-ppc64be/stage1/bin/llc <
> /home/buildbots/ppc64be-clang-test/clang-ppc64be/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
> -mtriple=x86_64-apple-darwin -mattr=+avx512f,+avx512bw |
> /home/buildbots/ppc64be-clang-test/clang-ppc64be/stage1/bin/FileCheck /home/buildbots/ppc64be-clang-test/clang-ppc64be/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
> --check-prefix=ALL --check-prefix=AVX512BW
> /home/buildbots/ppc64be-clang-test/clang-ppc64be/stage1/bin/llc <
> /home/buildbots/ppc64be-clang-test/clang-ppc64be/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
> -mtriple=i386-unknown-linux-gnu -mattr=+avx512f,+avx512bw |
> /home/buildbots/ppc64be-clang-test/clang-ppc64be/stage1/bin/FileCheck /home/buildbots/ppc64be-clang-test/clang-ppc64be/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
> --check-prefix=ALL --check-prefix=AVX512F-32
> --
> Exit Code: 1
>
> Command Output (stderr):
> --
> /home/buildbots/ppc64be-clang-test/clang-ppc64be/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll:32:18:
> error: AVX512BW-NEXT: is not on the line after the previous match
> ; AVX512BW-NEXT: movq %rsi, %rax
> ^
> <stdin>:20:2: note: 'next' match was here
> movq %rsi, %rax
> ^
> <stdin>:18:16: note: previous match ended here
> shlq $32, %rsi
> ^
> <stdin>:19:1: note: non-matching line after previous match is here
> orq %rdi, %rsi
> ^
>
> --
>
> ********************
>
>
> On 12/05/2017 09:42 AM, Jina Nahias via llvm-commits wrote:
>
> Author: jina.nahias
> Date: Tue Dec 5 07:42:56 2017
> New Revision: 319778
>
> URL: http://llvm.org/viewvc/llvm-project?rev=319778&view=rev
> <http://llvm.org/viewvc/llvm-project?rev=319778&view=rev>
> Log:
> [x86][AVX512] Lowering kunpack intrinsics to LLVM IR
>
> This patch, together with a matching clang patch
> (https://reviews.llvm.org/D39719
> <https://reviews.llvm.org/D39719>), implements the lowering of
> X86 kunpack intrinsics to IR.
>
> Differential Revision: https://reviews.llvm.org/D39720
> <https://reviews.llvm.org/D39720>
>
> Change-Id: I4088d9428478f9457f6afddc90bd3d66b3daf0a1
>
> Modified:
> llvm/trunk/include/llvm/IR/IntrinsicsX86.td
> llvm/trunk/lib/IR/AutoUpgrade.cpp
> llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
> llvm/trunk/lib/Target/X86/X86IntrinsicsInfo.h
> llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
> llvm/trunk/test/CodeGen/X86/avx512-intrinsics-upgrade.ll
> llvm/trunk/test/CodeGen/X86/avx512-intrinsics.ll
> llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll
> llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
> llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics.ll
>
> Modified: llvm/trunk/include/llvm/IR/IntrinsicsX86.td
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/IR/IntrinsicsX86.td?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/IR/IntrinsicsX86.td?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/include/llvm/IR/IntrinsicsX86.td (original)
> +++ llvm/trunk/include/llvm/IR/IntrinsicsX86.td Tue Dec 5
> 07:42:56 2017
> @@ -3738,15 +3738,6 @@ let TargetPrefix = "x86" in { // All in
> def int_x86_avx512_kxnor_w :
> GCCBuiltin<"__builtin_ia32_kxnorhi">,
> Intrinsic<[llvm_i16_ty], [llvm_i16_ty,
> llvm_i16_ty],
> [IntrNoMem]>;
> - def int_x86_avx512_kunpck_bw :
> GCCBuiltin<"__builtin_ia32_kunpckhi">,
> - Intrinsic<[llvm_i16_ty], [llvm_i16_ty, llvm_i16_ty],
> - [IntrNoMem]>;
> - def int_x86_avx512_kunpck_wd :
> GCCBuiltin<"__builtin_ia32_kunpcksi">,
> - Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
> - [IntrNoMem]>;
> - def int_x86_avx512_kunpck_dq :
> GCCBuiltin<"__builtin_ia32_kunpckdi">,
> - Intrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_i64_ty],
> - [IntrNoMem]>;
> def int_x86_avx512_kortestz_w :
> GCCBuiltin<"__builtin_ia32_kortestzhi">,
> Intrinsic<[llvm_i32_ty], [llvm_i16_ty,
> llvm_i16_ty],
> [IntrNoMem]>;
>
> Modified: llvm/trunk/lib/IR/AutoUpgrade.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/IR/AutoUpgrade.cpp?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/IR/AutoUpgrade.cpp?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/lib/IR/AutoUpgrade.cpp (original)
> +++ llvm/trunk/lib/IR/AutoUpgrade.cpp Tue Dec 5 07:42:56 2017
> @@ -78,6 +78,7 @@ static bool ShouldUpgradeX86Intrinsic(Fu
> Name=="ssse3.pabs.d.128" || // Added in 6.0
> Name.startswith("avx512.mask.shuf.i") || // Added in 6.0
> Name.startswith("avx512.mask.shuf.f") || // Added in 6.0
> + Name.startswith("avx512.kunpck") || //added in 6.0
> Name.startswith("avx2.pabs.") || // Added in 6.0
> Name.startswith("avx512.mask.pabs.") || // Added in 6.0
> Name.startswith("avx512.broadcastm") || // Added in 6.0
> @@ -1065,6 +1066,12 @@ void llvm::UpgradeIntrinsicCall(CallInst
> Rep = Builder.CreateVectorSplat(NumElts,
> CI->getArgOperand(0));
> Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
> CI->getArgOperand(1));
> + } else if (IsX86 && (Name.startswith("avx512.kunpck"))) {
> + uint64_t Shift = CI->getType()->getScalarSizeInBits() / 2;
> + uint64_t And = (1 << Shift) - 1;
> + Value* LowBits = Builder.CreateAnd(CI->getArgOperand(0),
> And);
> + Value* HighBits =
> Builder.CreateShl(CI->getArgOperand(1), Shift);
> + Rep = Builder.CreateOr(LowBits, HighBits);
> } else if (IsX86 && (Name == "sse.add.ss" || Name ==
> "sse2.add.sd <http://sse2.add.sd>")) {
> Type *I32Ty = Type::getInt32Ty(C);
> Value *Elt0 =
> Builder.CreateExtractElement(CI->getArgOperand(0),
>
> Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Tue Dec 5
> 07:42:56 2017
> @@ -30007,6 +30007,53 @@ static SDValue combineBitcastvxi1(Select
> SDValue N0 = BitCast.getOperand(0);
> EVT VecVT = N0->getValueType(0);
> + if (VT.isVector() && VecVT.isScalarInteger() &&
> Subtarget.hasAVX512() &&
> + N0->getOpcode() == ISD::OR) {
> + SDValue Op0 = N0->getOperand(0);
> + SDValue Op1 = N0->getOperand(1);
> + MVT TrunckVT;
> + MVT BitcastVT;
> + switch (VT.getSimpleVT().SimpleTy) {
> + default:
> + return SDValue();
> + case MVT::v16i1:
> + TrunckVT = MVT::i8;
> + BitcastVT = MVT::v8i1;
> + break;
> + case MVT::v32i1:
> + TrunckVT = MVT::i16;
> + BitcastVT = MVT::v16i1;
> + break;
> + case MVT::v64i1:
> + TrunckVT = MVT::i32;
> + BitcastVT = MVT::v32i1;
> + break;
> + }
> + bool isArg0UndefRight = Op0->getOpcode() == ISD::SHL;
> + bool isArg0UndefLeft =
> + Op0->getOpcode() == ISD::ZERO_EXTEND ||
> Op0->getOpcode() == ISD::AND;
> + bool isArg1UndefRight = Op1->getOpcode() == ISD::SHL;
> + bool isArg1UndefLeft =
> + Op1->getOpcode() == ISD::ZERO_EXTEND ||
> Op1->getOpcode() == ISD::AND;
> + SDValue OpLeft;
> + SDValue OpRight;
> + if (isArg0UndefRight && isArg1UndefLeft) {
> + OpLeft = Op0;
> + OpRight = Op1;
> + } else if (isArg1UndefRight && isArg0UndefLeft) {
> + OpLeft = Op1;
> + OpRight = Op0;
> + } else
> + return SDValue();
> + SDLoc DL(BitCast);
> + SDValue Shr = OpLeft->getOperand(0);
> + SDValue Trunc1 = DAG.getNode(ISD::TRUNCATE, DL, TrunckVT, Shr);
> + SDValue Bitcast1 = DAG.getBitcast(BitcastVT, Trunc1);
> + SDValue Trunc2 = DAG.getNode(ISD::TRUNCATE, DL, TrunckVT,
> OpRight);
> + SDValue Bitcast2 = DAG.getBitcast(BitcastVT, Trunc2);
> + return DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, Bitcast1,
> Bitcast2);
> + }
> +
> if (!VT.isScalarInteger() || !VecVT.isSimple())
> return SDValue();
>
> Modified: llvm/trunk/lib/Target/X86/X86IntrinsicsInfo.h
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86IntrinsicsInfo.h?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86IntrinsicsInfo.h?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86IntrinsicsInfo.h (original)
> +++ llvm/trunk/lib/Target/X86/X86IntrinsicsInfo.h Tue Dec 5
> 07:42:56 2017
> @@ -479,9 +479,6 @@ static const IntrinsicData IntrinsicsWi
> X86_INTRINSIC_DATA(avx512_exp2_ps, INTR_TYPE_1OP_MASK_RM,
> X86ISD::EXP2, 0),
> X86_INTRINSIC_DATA(avx512_kand_w, MASK_BINOP, ISD::AND, 0),
> X86_INTRINSIC_DATA(avx512_kor_w, MASK_BINOP, ISD::OR, 0),
> - X86_INTRINSIC_DATA(avx512_kunpck_bw, KUNPCK,
> ISD::CONCAT_VECTORS, 0),
> - X86_INTRINSIC_DATA(avx512_kunpck_dq, KUNPCK,
> ISD::CONCAT_VECTORS, 0),
> - X86_INTRINSIC_DATA(avx512_kunpck_wd, KUNPCK,
> ISD::CONCAT_VECTORS, 0),
> X86_INTRINSIC_DATA(avx512_kxor_w, MASK_BINOP, ISD::XOR, 0),
> X86_INTRINSIC_DATA(avx512_mask_add_pd_512,
> INTR_TYPE_2OP_MASK, ISD::FADD,
> X86ISD::FADD_RND),
>
> Modified: llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
> (original)
> +++ llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
> Tue Dec 5 07:42:56 2017
> @@ -5,6 +5,59 @@
> ; NOTE: This should use IR equivalent to what is generated by
> clang/test/CodeGen/avx512f-builtins.c
> +define zeroext i16 @test_mm512_kunpackb(<8 x i64> %__A, <8
> x i64> %__B, <8 x i64> %__C, <8 x i64> %__D, <8 x i64> %__E, <8
> x i64> %__F) local_unnamed_addr #0 {
> +; X32-LABEL: test_mm512_kunpackb:
> +; X32: # %bb.0: # %entry
> +; X32-NEXT: pushl %ebp
> +; X32-NEXT: .cfi_def_cfa_offset 8
> +; X32-NEXT: .cfi_offset %ebp, -8
> +; X32-NEXT: movl %esp, %ebp
> +; X32-NEXT: .cfi_def_cfa_register %ebp
> +; X32-NEXT: andl $-64, %esp
> +; X32-NEXT: subl $64, %esp
> +; X32-NEXT: vmovdqa64 136(%ebp), %zmm3
> +; X32-NEXT: vpcmpneqd %zmm1, %zmm0, %k0
> +; X32-NEXT: vpcmpneqd 8(%ebp), %zmm2, %k1
> +; X32-NEXT: kunpckbw %k0, %k1, %k1
> +; X32-NEXT: vpcmpneqd 72(%ebp), %zmm3, %k0 {%k1}
> +; X32-NEXT: kmovw %k0, %eax
> +; X32-NEXT: movzwl %ax, %eax
> +; X32-NEXT: movl %ebp, %esp
> +; X32-NEXT: popl %ebp
> +; X32-NEXT: vzeroupper
> +; X32-NEXT: retl
> +;
> +; X64-LABEL: test_mm512_kunpackb:
> +; X64: # %bb.0: # %entry
> +; X64-NEXT: vpcmpneqd %zmm1, %zmm0, %k0
> +; X64-NEXT: vpcmpneqd %zmm3, %zmm2, %k1
> +; X64-NEXT: kunpckbw %k0, %k1, %k1
> +; X64-NEXT: vpcmpneqd %zmm5, %zmm4, %k0 {%k1}
> +; X64-NEXT: kmovw %k0, %eax
> +; X64-NEXT: movzwl %ax, %eax
> +; X64-NEXT: vzeroupper
> +; X64-NEXT: retq
> +entry:
> + %0 = bitcast <8 x i64> %__A to <16 x i32>
> + %1 = bitcast <8 x i64> %__B to <16 x i32>
> + %2 = icmp ne <16 x i32> %0, %1
> + %3 = bitcast <16 x i1> %2 to i16
> + %4 = bitcast <8 x i64> %__C to <16 x i32>
> + %5 = bitcast <8 x i64> %__D to <16 x i32>
> + %6 = icmp ne <16 x i32> %4, %5
> + %7 = bitcast <16 x i1> %6 to i16
> + %8 = and i16 %7, 255
> + %shl.i = shl i16 %3, 8
> + %or.i = or i16 %8, %shl.i
> + %9 = bitcast <8 x i64> %__E to <16 x i32>
> + %10 = bitcast <8 x i64> %__F to <16 x i32>
> + %11 = icmp ne <16 x i32> %9, %10
> + %12 = bitcast i16 %or.i to <16 x i1>
> + %13 = and <16 x i1> %11, %12
> + %14 = bitcast <16 x i1> %13 to i16
> + ret i16 %14
> +}
> +
> define <16 x float> @test_mm512_shuffle_f32x4(<16 x float>
> %__A, <16 x float> %__B) {
> ; X32-LABEL: test_mm512_shuffle_f32x4:
> ; X32: # %bb.0: # %entry
>
> Modified: llvm/trunk/test/CodeGen/X86/avx512-intrinsics-upgrade.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-intrinsics-upgrade.ll?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-intrinsics-upgrade.ll?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/avx512-intrinsics-upgrade.ll
> (original)
> +++ llvm/trunk/test/CodeGen/X86/avx512-intrinsics-upgrade.ll Tue
> Dec 5 07:42:56 2017
> @@ -1,7 +1,21 @@
> ; NOTE: Assertions have been autogenerated by
> utils/update_llc_test_checks.py
> ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl |
> FileCheck %s
> - define <16 x
> i32>@test_int_x86_avx512_mask_pbroadcastd_gpr_512(i32 %x0, <16 x
> i32> %x1, i16 %mask) {
> +declare i16 @llvm.x86.avx512.kunpck.bw
> <http://llvm.x86.avx512.kunpck.bw>(i16, i16) nounwind readnone
> +
> +define i16 @unpckbw_test(i16 %a0, i16 %a1) {
> +; CHECK-LABEL: unpckbw_test:
> +; CHECK: ## %bb.0:
> +; CHECK-NEXT: movzbl %dil, %eax
> +; CHECK-NEXT: shll $8, %esi
> +; CHECK-NEXT: orl %esi, %eax
> +; CHECK-NEXT: ## kill: %ax<def> %ax<kill> %eax<kill>
> +; CHECK-NEXT: retq
> + %res = call i16 @llvm.x86.avx512.kunpck.bw
> <http://llvm.x86.avx512.kunpck.bw>(i16 %a0, i16 %a1)
> + ret i16 %res
> +}
> +
> +define <16 x
> i32>@test_int_x86_avx512_mask_pbroadcastd_gpr_512(i32 %x0, <16 x
> i32> %x1, i16 %mask) {
> ; CHECK-LABEL: test_int_x86_avx512_mask_pbroadcastd_gpr_512:
> ; CHECK: ## %bb.0:
> ; CHECK-NEXT: vpbroadcastd %edi, %zmm1
>
> Modified: llvm/trunk/test/CodeGen/X86/avx512-intrinsics.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-intrinsics.ll?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-intrinsics.ll?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/avx512-intrinsics.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/avx512-intrinsics.ll Tue Dec 5
> 07:42:56 2017
> @@ -96,21 +96,6 @@ define i16 @test_kor(i16 %a0, i16 %a1) {
> ret i16 %t2
> }
> -declare i16 @llvm.x86.avx512.kunpck.bw
> <http://llvm.x86.avx512.kunpck.bw>(i16, i16) nounwind readnone
> -
> -define i16 @unpckbw_test(i16 %a0, i16 %a1) {
> -; CHECK-LABEL: unpckbw_test:
> -; CHECK: ## %bb.0:
> -; CHECK-NEXT: kmovw %edi, %k0
> -; CHECK-NEXT: kmovw %esi, %k1
> -; CHECK-NEXT: kunpckbw %k1, %k0, %k0
> -; CHECK-NEXT: kmovw %k0, %eax
> -; CHECK-NEXT: ## kill: %ax<def> %ax<kill> %eax<kill>
> -; CHECK-NEXT: retq
> - %res = call i16 @llvm.x86.avx512.kunpck.bw
> <http://llvm.x86.avx512.kunpck.bw>(i16 %a0, i16 %a1)
> - ret i16 %res
> -}
> -
> declare i16 @llvm.x86.avx512.kxnor.w(i16, i16) nounwind readnone
> ; TODO: the two kxnor instructions here a no op and should be
> elimintaed,
> ; probably by FoldConstantArithmetic in SelectionDAG.
>
> Modified:
> llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll
> (original)
> +++ llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll
> Tue Dec 5 07:42:56 2017
> @@ -4,6 +4,117 @@
> ; NOTE: This should use IR equivalent to what is generated
> by clang/test/CodeGen/avx512bw-builtins.c
> +define i64 @test_mm512_kunpackd(<8 x i64> %__A, <8 x i64>
> %__B, <8 x i64> %__C, <8 x i64> %__D, <8 x i64> %__E, <8 x i64>
> %__F) {
> +; X32-LABEL: test_mm512_kunpackd:
> +; X32: # %bb.0: # %entry
> +; X32-NEXT: pushl %ebp
> +; X32-NEXT: .cfi_def_cfa_offset 8
> +; X32-NEXT: .cfi_offset %ebp, -8
> +; X32-NEXT: movl %esp, %ebp
> +; X32-NEXT: .cfi_def_cfa_register %ebp
> +; X32-NEXT: andl $-64, %esp
> +; X32-NEXT: subl $64, %esp
> +; X32-NEXT: vmovdqa64 136(%ebp), %zmm3
> +; X32-NEXT: vmovdqa64 72(%ebp), %zmm4
> +; X32-NEXT: vmovdqa64 8(%ebp), %zmm5
> +; X32-NEXT: vpcmpneqb %zmm0, %zmm1, %k0
> +; X32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
> +; X32-NEXT: vpcmpneqb %zmm5, %zmm2, %k0
> +; X32-NE
> <https://maps.google.com/?q=%2B;+X32-NE&entry=gmail&source=g>XT:
> kmovq %k0, {{[0-9]+}}(%esp)
> +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k0
> +; X32-NEXT: kmovd {{[0-9]+}}(%esp), %k1
> +; X32-NEXT: kunpckdq %k0, %k1, %k1
> +; X32-NEXT: vpcmpneqb %zmm3, %zmm4, %k0
> <https://maps.google.com/?q=%25zmm4,+%25k0&entry=gmail&source=g>
> {%k1}
> +; X32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
> +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
> +; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
> +; X32-NEXT: movl %ebp, %esp
> +; X32-NEXT: popl %ebp
> +; X32-NEXT: vzeroupper
> +; X32-NEXT: retl
> +;
> +; X64-LABEL: test_mm512_kunpackd:
> +; X64: # %bb.0: # %entry
> +; X64-NEXT: vpcmpneqb %zmm0, %zmm1, %k0
> +; X64-NEXT: vpcmpneqb %zmm3, %zmm2, %k1
> +; X64-NEXT: kunpckdq %k0, %k1, %k1
> +; X64-NEXT: vpcmpneqb %zmm5, %zmm4, %k0 {%k1}
> +; X64-NEXT: kmovq %k0, %rax
> +; X64-NEXT: vzeroupper
> +; X64-NEXT: retq
> +entry:
> + %0 = bitcast <8 x i64> %__B to <64 x i8>
> + %1 = bitcast <8 x i64> %__A to <64 x i8>
> + %2 = icmp ne <64 x i8> %0, %1
> + %3 = bitcast <64 x i1> %2 to i64
> + %4 = bitcast <8 x i64> %__C to <64 x i8>
> + %5 = bitcast <8 x i64> %__D to <64 x i8>
> + %6 = icmp ne <64 x i8> %4, %5
> + %7 = bitcast <64 x i1> %6 to i64
> + %and.i = and i64 %7, 4294967295
> + %shl.i = shl i64 %3, 32
> + %or.i = or i64 %and.i, %shl.i
> + %8 = bitcast <8 x i64> %__E to <64 x i8>
> + %9 = bitcast <8 x i64> %__F to <64 x i8>
> + %10 = icmp ne <64 x i8> %8, %9
> + %11 = bitcast i64 %or.i to <64 x i1>
> + %12 = and <64 x i1> %10, %11
> + %13 = bitcast <64 x i1> %12 to i64
> + ret i64 %13
> +}
> +
> +define i32 @test_mm512_kunpackw(<8 x i64> %__A, <8 x i64> %__B,
> <8 x i64> %__C, <8 x i64> %__D, <8 x i64> %__E, <8 x i64> %__F) {
> +; X32-LABEL: test_mm512_kunpackw:
> +; X32: # %bb.0: # %entry
> +; X32-NEXT: pushl %ebp
> +; X32-NEXT: .cfi_def_cfa_offset 8
> +; X32-NEXT: .cfi_offset %ebp, -8
> +; X32-NEXT: movl %esp, %ebp
> +; X32-NEXT: .cfi_def_cfa_register %ebp
> +; X32-NEXT: andl $-64, %esp
> +; X32-NEXT: subl $64, %esp
> +; X32-NEXT: vmovdqa64 136(%ebp), %zmm3
> +; X32-NEXT: vpcmpneqw %zmm0, %zmm1, %k0
> +; X32-NEXT: vpcmpneqw 8(%ebp), %zmm2, %k1
> +; X32-NEXT: kunpckwd %k0, %k1, %k1
> +; X32-NEXT: vpcmpneqw 72(%ebp), %zmm3, %k0 {%k1}
> +; X32-NEXT: kmovd %k0, %eax
> +; X32-NEXT: movl %ebp, %esp
> +; X32-NEXT: popl %ebp
> +; X32-NEXT: vzeroupper
> +; X32-NEXT: retl
> +;
> +; X64-LABEL: test_mm512_kunpackw:
> +; X64: # %bb.0: # %entry
> +; X64-NEXT: vpcmpneqw %zmm0, %zmm1, %k0
> +; X64-NEXT: vpcmpneqw %zmm3, %zmm2, %k1
> +; X64-NEXT: kunpckwd %k0, %k1, %k1
> +; X64-NEXT: vpcmpneqw %zmm5, %zmm4, %k0 {%k1}
> +; X64-NEXT: kmovd %k0, %eax
> +; X64-NEXT: vzeroupper
> +; X64-NEXT: retq
> +entry:
> + %0 = bitcast <8 x i64> %__B to <32 x i16>
> + %1 = bitcast <8 x i64> %__A to <32 x i16>
> + %2 = icmp ne <32 x i16> %0, %1
> + %3 = bitcast <32 x i1> %2 to i32
> + %4 = bitcast <8 x i64> %__C to <32 x i16>
> + %5 = bitcast <8 x i64> %__D to <32 x i16>
> + %6 = icmp ne <32 x i16> %4, %5
> + %7 = bitcast <32 x i1> %6 to i32
> + %and.i = and i32 %7, 65535
> + %shl.i = shl i32 %3, 16
> + %or.i = or i32 %and.i, %shl.i
> + %8 = bitcast <8 x i64> %__E to <32 x i16>
> + %9 = bitcast <8 x i64> %__F to <32 x i16>
> + %10 = icmp ne <32 x i16> %8, %9
> + %11 = bitcast i32 %or.i to <32 x i1>
> + %12 = and <32 x i1> %10, %11
> + %13 = bitcast <32 x i1> %12 to i32
> + ret i32 %13
> +}
> +
> +
> define <8 x i64> @test_mm512_mask_set1_epi8(<8 x i64> %__O,
> i64 %__M, i8 signext %__A) {
> ; X32-LABEL: test_mm512_mask_set1_epi8:
> ; X32: # %bb.0: # %entry
> @@ -694,13 +805,13 @@ define <8 x i64> @test_mm512_mask_set1_e
> ; X32-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
> ; X32-NEXT: vmovdqa {{.*#+}} ymm5 =
> [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,255,255]
> ; X32-NEXT: vpblendvb %ymm5, %ymm1, %ymm2, %ymm1
> -; X32-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
> -; X32-NEXT: vpmovb2m %zmm0, %k0
> -; X32-NEXT: vpmovm2b %k0, %zmm0
> -; X32-NEXT: vextracti64x4 $1, %zmm0, %ymm1
> ; X32-NEXT: movl %eax, %ecx
> ; X32-NEXT: shrl $30, %ecx
> ; X32-NEXT: kmovd %ecx, %k0
> +; X32-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
> +; X32-NEXT: vpmovb2m %zmm0, %k1
> +; X32-NEXT: vpmovm2b %k1, %zmm0
> +; X32-NEXT: vextracti64x4 $1, %zmm0, %ymm1
> ; X32-NEXT: vpmovm2b %k0, %zmm2
> ; X32-NEXT: vpbroadcastw %xmm2, %xmm2
> ; X32-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
> @@ -1422,13 +1533,13 @@ define <8 x i64> @test_mm512_maskz_set1_
> ; X32-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
> ; X32-NEXT: vmovdqa {{.*#+}} ymm4 =
> [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,255,255]
> ; X32-NEXT: vpblendvb %ymm4, %ymm1, %ymm2, %ymm1
> -; X32-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
> -; X32-NEXT: vpmovb2m %zmm0, %k0
> -; X32-NEXT: vpmovm2b %k0, %zmm0
> -; X32-NEXT: vextracti64x4 $1, %zmm0, %ymm1
> ; X32-NEXT: movl %eax, %ecx
> ; X32-NEXT: shrl $30, %ecx
> ; X32-NEXT: kmovd %ecx, %k0
> +; X32-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
> +; X32-NEXT: vpmovb2m %zmm0, %k1
> +; X32-NEXT: vpmovm2b %k1, %zmm0
> +; X32-NEXT: vextracti64x4 $1, %zmm0, %ymm1
> ; X32-NEXT: vpmovm2b %k0, %zmm2
> ; X32-NEXT: vpbroadcastw %xmm2, %xmm2
> ; X32-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
>
> Modified: llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
> (original)
> +++ llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
> Tue Dec 5 07:42:56 2017
> @@ -2,6 +2,45 @@
> ; RUN: llc < %s -mtriple=x86_64-apple-darwin
> -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=ALL
> --check-prefix=AVX512BW
> ; RUN: llc < %s -mtriple=i386-unknown-linux-gnu
> -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=ALL
> --check-prefix=AVX512F-32
> +declare i32 @llvm.x86.avx512.kunpck.wd(i32, i32)
> +
> +define i32 at test_int_x86_avx512_kunpck_wd(i32 %x0, i32 %x1) {
> +; AVX512BW-LABEL: test_int_x86_avx512_kunpck_wd:
> +; AVX512BW: ## %bb.0:
> +; AVX512BW-NEXT: movzwl %di, %eax
> +; AVX512BW-NEXT: shll $16, %esi
> +; AVX512BW-NEXT: orl %esi, %eax
> +; AVX512BW-NEXT: retq
> +;
> +; AVX512F-32-LABEL: test_int_x86_avx512_kunpck_wd:
> +; AVX512F-32: # %bb.0:
> +; AVX512F-32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
> +; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %eax
> +; AVX512F-32-NEXT: shll $16, %eax
> +; AVX512F-32-NEXT: orl %ecx, %eax
> +; AVX512F-32-NEXT: retl
> + %res = call i32 @llvm.x86.avx512.kunpck.wd(i32 %x0, i32 %x1)
> + ret i32 %res
> +}
> +
> +declare i64 @llvm.x86.avx512.kunpck.dq(i64, i64)
> +
> +define i64 at test_int_x86_avx512_kunpck_qd(i64 %x0, i64 %x1) {
> +; AVX512BW-LABEL: test_int_x86_avx512_kunpck_qd:
> +; AVX512BW: ## %bb.0:
> +; AVX512BW-NEXT: shlq $32, %rsi
> +; AVX512BW-NEXT: movq %rsi, %rax
> +; AVX512BW-NEXT: retq
> +;
> +; AVX512F-32-LABEL: test_int_x86_avx512_kunpck_qd:
> +; AVX512F-32: # %bb.0:
> +; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %edx
> +; AVX512F-32-NEXT: xorl %eax, %eax
> +; AVX512F-32-NEXT: retl
> + %res = call i64 @llvm.x86.avx512.kunpck.dq(i64 %x0, i64 %x1)
> + ret i64 %res
> +}
> +
> declare <64 x i8>
> @llvm.x86.avx512.mask.pbroadcast.b.gpr.512(i8, <64 x i8>, i64)
> define <64 x
> i8>@test_int_x86_avx512_mask_pbroadcast_b_gpr_512(i8 %x0, <64 x
> i8> %x1, i64 %mask) {
>
> Modified: llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics.ll?rev=319778&r1=319777&r2=319778&view=diff
> <http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics.ll?rev=319778&r1=319777&r2=319778&view=diff>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/avx512bw-intrinsics.ll Tue Dec
> 5 07:42:56 2017
> @@ -1455,55 +1455,6 @@ define <8 x i64>@test_int_x86_avx512_ma
> ret <8 x i64> %res2
> }
> -declare i32 @llvm.x86.avx512.kunpck.wd(i32, i32)
> -
> -define i32 at test_int_x86_avx512_kunpck_wd(i32 %x0, i32 %x1) {
> -; AVX512BW-LABEL: test_int_x86_avx512_kunpck_wd:
> -; AVX512BW: ## %bb.0:
> -; AVX512BW-NEXT: kmovd %edi, %k0
> -; AVX512BW-NEXT: kmovd %esi, %k1
> -; AVX512BW-NEXT: kunpckwd %k1, %k0, %k0
> -; AVX512BW-NEXT: kmovd %k0, %eax
> -; AVX512BW-NEXT: retq
> -;
> -; AVX512F-32-LABEL: test_int_x86_avx512_kunpck_wd:
> -; AVX512F-32: # %bb.0:
> -; AVX512F-32-NEXT: kmovw {{[0-9]+}}(%esp), %k0
> -; AVX512F-32-NEXT: kmovw {{[0-9]+}}(%esp), %k1
> -; AVX512F-32-NEXT: kunpckwd %k0, %k1, %k0
> -; AVX512F-32-NEXT: kmovd %k0, %eax
> -; AVX512F-32-NEXT: retl
> - %res = call i32 @llvm.x86.avx512.kunpck.wd(i32 %x0, i32 %x1)
> - ret i32 %res
> -}
> -
> -declare i64 @llvm.x86.avx512.kunpck.dq(i64, i64)
> -
> -define i64 at test_int_x86_avx512_kunpck_qd(i64 %x0, i64 %x1) {
> -; AVX512BW-LABEL: test_int_x86_avx512_kunpck_qd:
> -; AVX512BW: ## %bb.0:
> -; AVX512BW-NEXT: kmovq %rdi, %k0
> -; AVX512BW-NEXT: kmovq %rsi, %k1
> -; AVX512BW-NEXT: kunpckdq %k1, %k0, %k0
> -; AVX512BW-NEXT: kmovq %k0, %rax
> -; AVX512BW-NEXT: retq
> -;
> -; AVX512F-32-LABEL: test_int_x86_avx512_kunpck_qd:
> -; AVX512F-32: # %bb.0:
> -; AVX512F-32-NEXT: subl $12, %esp
> -; AVX512F-32-NEXT: .cfi_def_cfa_offset 16
> -; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k0
> -; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k1
> -; AVX512F-32-NEXT: kunpckdq %k0, %k1, %k0
> -; AVX512F-32-NEXT: kmovq %k0, (%esp)
> -; AVX512F-32-NEXT: movl (%esp), %eax
> -; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %edx
> -; AVX512F-32-NEXT: addl $12, %esp
> -; AVX512F-32-NEXT: retl
> - %res = call i64 @llvm.x86.avx512.kunpck.dq(i64 %x0, i64 %x1)
> - ret i64 %res
> -}
> -
> declare i64 @llvm.x86.avx512.cvtb2mask.512(<64 x i8>)
> define i64 at test_int_x86_avx512_cvtb2mask_512(<64 x i8> %x0) {
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org <mailto:llvm-commits at lists.llvm.org>
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
> <http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits>
>
>
>
> --
>
> -Bill Seurer
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org <mailto:llvm-commits at lists.llvm.org>
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
> <http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits>
>
>
--
-Bill Seurer
More information about the llvm-commits
mailing list