[llvm] r374579 - [X86][SSE] Add support for v4i8 add reduction
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Fri Oct 11 11:51:08 PDT 2019
Why do the load cases use a movzxdq after the movd? That seems unnecessary.
The movd should have generated 0s already.
~Craig
On Fri, Oct 11, 2019 at 10:51 AM Simon Pilgrim via llvm-commits <
llvm-commits at lists.llvm.org> wrote:
> Author: rksimon
> Date: Fri Oct 11 10:54:15 2019
> New Revision: 374579
>
> URL: http://llvm.org/viewvc/llvm-project?rev=374579&view=rev
> Log:
> [X86][SSE] Add support for v4i8 add reduction
>
> Modified:
> llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
> llvm/trunk/test/CodeGen/X86/vector-reduce-add.ll
>
> Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=374579&r1=374578&r2=374579&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Fri Oct 11 10:54:15 2019
> @@ -36239,10 +36239,15 @@ static SDValue combineReductionToHorizon
>
> SDLoc DL(ExtElt);
>
> - if (VecVT == MVT::v8i8) {
> + // vXi8 reduction - sub 128-bit vector.
> + if (VecVT == MVT::v4i8 || VecVT == MVT::v8i8) {
> + // Pad with zero.
> + if (VecVT == MVT::v4i8)
> + Rdx = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v8i8, Rdx,
> + DAG.getConstant(0, DL, VecVT));
> // Pad with undef.
> Rdx = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v16i8, Rdx,
> - DAG.getUNDEF(VecVT));
> + DAG.getUNDEF(MVT::v8i8));
> Rdx = DAG.getNode(X86ISD::PSADBW, DL, MVT::v2i64, Rdx,
> DAG.getConstant(0, DL, MVT::v16i8));
> Rdx = DAG.getBitcast(MVT::v16i8, Rdx);
>
> Modified: llvm/trunk/test/CodeGen/X86/vector-reduce-add.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-reduce-add.ll?rev=374579&r1=374578&r2=374579&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/vector-reduce-add.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/vector-reduce-add.ll Fri Oct 11 10:54:15
> 2019
> @@ -1029,44 +1029,36 @@ define i8 @test_v2i8_load(<2 x i8>* %p)
> define i8 @test_v4i8(<4 x i8> %a0) {
> ; SSE2-LABEL: test_v4i8:
> ; SSE2: # %bb.0:
> -; SSE2-NEXT: movdqa %xmm0, %xmm1
> -; SSE2-NEXT: psrld $16, %xmm1
> -; SSE2-NEXT: paddb %xmm0, %xmm1
> -; SSE2-NEXT: movdqa %xmm1, %xmm0
> -; SSE2-NEXT: psrlw $8, %xmm0
> -; SSE2-NEXT: paddb %xmm1, %xmm0
> +; SSE2-NEXT: pxor %xmm1, %xmm1
> +; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
> +; SSE2-NEXT: psadbw %xmm1, %xmm0
> ; SSE2-NEXT: movd %xmm0, %eax
> ; SSE2-NEXT: # kill: def $al killed $al killed $eax
> ; SSE2-NEXT: retq
> ;
> ; SSE41-LABEL: test_v4i8:
> ; SSE41: # %bb.0:
> -; SSE41-NEXT: movdqa %xmm0, %xmm1
> -; SSE41-NEXT: psrld $16, %xmm1
> -; SSE41-NEXT: paddb %xmm0, %xmm1
> -; SSE41-NEXT: movdqa %xmm1, %xmm0
> -; SSE41-NEXT: psrlw $8, %xmm0
> -; SSE41-NEXT: paddb %xmm1, %xmm0
> -; SSE41-NEXT: pextrb $0, %xmm0, %eax
> +; SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
> +; SSE41-NEXT: pxor %xmm1, %xmm1
> +; SSE41-NEXT: psadbw %xmm0, %xmm1
> +; SSE41-NEXT: pextrb $0, %xmm1, %eax
> ; SSE41-NEXT: # kill: def $al killed $al killed $eax
> ; SSE41-NEXT: retq
> ;
> ; AVX-LABEL: test_v4i8:
> ; AVX: # %bb.0:
> -; AVX-NEXT: vpsrld $16, %xmm0, %xmm1
> -; AVX-NEXT: vpaddb %xmm1, %xmm0, %xmm0
> -; AVX-NEXT: vpsrlw $8, %xmm0, %xmm1
> -; AVX-NEXT: vpaddb %xmm1, %xmm0, %xmm0
> +; AVX-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
> +; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
> +; AVX-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
> ; AVX-NEXT: vpextrb $0, %xmm0, %eax
> ; AVX-NEXT: # kill: def $al killed $al killed $eax
> ; AVX-NEXT: retq
> ;
> ; AVX512-LABEL: test_v4i8:
> ; AVX512: # %bb.0:
> -; AVX512-NEXT: vpsrld $16, %xmm0, %xmm1
> -; AVX512-NEXT: vpaddb %xmm1, %xmm0, %xmm0
> -; AVX512-NEXT: vpsrlw $8, %xmm0, %xmm1
> -; AVX512-NEXT: vpaddb %xmm1, %xmm0, %xmm0
> +; AVX512-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
> +; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
> +; AVX512-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
> ; AVX512-NEXT: vpextrb $0, %xmm0, %eax
> ; AVX512-NEXT: # kill: def $al killed $al killed $eax
> ; AVX512-NEXT: retq
> @@ -1078,36 +1070,28 @@ define i8 @test_v4i8_load(<4 x i8>* %p)
> ; SSE2-LABEL: test_v4i8_load:
> ; SSE2: # %bb.0:
> ; SSE2-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
> -; SSE2-NEXT: movdqa %xmm0, %xmm1
> -; SSE2-NEXT: psrld $16, %xmm1
> -; SSE2-NEXT: paddb %xmm0, %xmm1
> -; SSE2-NEXT: movdqa %xmm1, %xmm0
> -; SSE2-NEXT: psrlw $8, %xmm0
> -; SSE2-NEXT: paddb %xmm1, %xmm0
> -; SSE2-NEXT: movd %xmm0, %eax
> +; SSE2-NEXT: pxor %xmm1, %xmm1
> +; SSE2-NEXT: psadbw %xmm0, %xmm1
> +; SSE2-NEXT: movd %xmm1, %eax
> ; SSE2-NEXT: # kill: def $al killed $al killed $eax
> ; SSE2-NEXT: retq
> ;
> ; SSE41-LABEL: test_v4i8_load:
> ; SSE41: # %bb.0:
> ; SSE41-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
> -; SSE41-NEXT: movdqa %xmm0, %xmm1
> -; SSE41-NEXT: psrld $16, %xmm1
> -; SSE41-NEXT: paddb %xmm0, %xmm1
> -; SSE41-NEXT: movdqa %xmm1, %xmm0
> -; SSE41-NEXT: psrlw $8, %xmm0
> -; SSE41-NEXT: paddb %xmm1, %xmm0
> -; SSE41-NEXT: pextrb $0, %xmm0, %eax
> +; SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
> +; SSE41-NEXT: pxor %xmm1, %xmm1
> +; SSE41-NEXT: psadbw %xmm0, %xmm1
> +; SSE41-NEXT: pextrb $0, %xmm1, %eax
> ; SSE41-NEXT: # kill: def $al killed $al killed $eax
> ; SSE41-NEXT: retq
> ;
> ; AVX-LABEL: test_v4i8_load:
> ; AVX: # %bb.0:
> ; AVX-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
> -; AVX-NEXT: vpsrld $16, %xmm0, %xmm1
> -; AVX-NEXT: vpaddb %xmm1, %xmm0, %xmm0
> -; AVX-NEXT: vpsrlw $8, %xmm0, %xmm1
> -; AVX-NEXT: vpaddb %xmm1, %xmm0, %xmm0
> +; AVX-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
> +; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
> +; AVX-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
> ; AVX-NEXT: vpextrb $0, %xmm0, %eax
> ; AVX-NEXT: # kill: def $al killed $al killed $eax
> ; AVX-NEXT: retq
> @@ -1115,10 +1099,9 @@ define i8 @test_v4i8_load(<4 x i8>* %p)
> ; AVX512-LABEL: test_v4i8_load:
> ; AVX512: # %bb.0:
> ; AVX512-NEXT: vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
> -; AVX512-NEXT: vpsrld $16, %xmm0, %xmm1
> -; AVX512-NEXT: vpaddb %xmm1, %xmm0, %xmm0
> -; AVX512-NEXT: vpsrlw $8, %xmm0, %xmm1
> -; AVX512-NEXT: vpaddb %xmm1, %xmm0, %xmm0
> +; AVX512-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
> +; AVX512-NEXT: vpxor %xmm1, %xmm1, %xmm1
> +; AVX512-NEXT: vpsadbw %xmm1, %xmm0, %xmm0
> ; AVX512-NEXT: vpextrb $0, %xmm0, %eax
> ; AVX512-NEXT: # kill: def $al killed $al killed $eax
> ; AVX512-NEXT: retq
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20191011/b13eb1ba/attachment.html>
More information about the llvm-commits
mailing list