maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/X86/avx-cvt-3.ll

147 lines
6.1 KiB
LLVM
Raw Normal View History

[X86][AVX] Add 8i32 -> 8f32 sitofp tests with constant insertion Some elements are constant inserted into the source integer vector before conversion. llvm-svn: 294147
2017-02-06 05:40:25 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=avx | FileCheck %s --check-prefix=X86
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx | FileCheck %s --check-prefix=X64
; Insertion/shuffles of all-zero/all-bits/constants into v8i32->v8f32 sitofp conversion.
define <8 x float> @sitofp_insert_zero_v8i32(<8 x i32> %a0) {
; X86-LABEL: sitofp_insert_zero_v8i32:
; X86: # BB#0:
; X86-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X86-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6,7]
; X86-NEXT: vcvtdq2ps %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: sitofp_insert_zero_v8i32:
; X64: # BB#0:
; X64-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X64-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6,7]
; X64-NEXT: vcvtdq2ps %ymm0, %ymm0
; X64-NEXT: retq
%1 = insertelement <8 x i32> %a0, i32 0, i32 0
%2 = insertelement <8 x i32> %1, i32 0, i32 2
%3 = insertelement <8 x i32> %2, i32 0, i32 4
%4 = insertelement <8 x i32> %3, i32 0, i32 5
%5 = sitofp <8 x i32> %4 to <8 x float>
ret <8 x float> %5
}
define <8 x float> @sitofp_shuffle_zero_v8i32(<8 x i32> %a0) {
; X86-LABEL: sitofp_shuffle_zero_v8i32:
; X86: # BB#0:
; X86-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X86-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7]
; X86-NEXT: vcvtdq2ps %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: sitofp_shuffle_zero_v8i32:
; X64: # BB#0:
; X64-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X64-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7]
; X64-NEXT: vcvtdq2ps %ymm0, %ymm0
; X64-NEXT: retq
%1 = shufflevector <8 x i32> %a0, <8 x i32> <i32 0, i32 undef, i32 0, i32 undef, i32 0, i32 undef, i32 0, i32 undef>, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
%2 = sitofp <8 x i32> %1 to <8 x float>
ret <8 x float> %2
}
define <8 x float> @sitofp_insert_allbits_v8i32(<8 x i32> %a0) {
; X86-LABEL: sitofp_insert_allbits_v8i32:
; X86: # BB#0:
[x86, SSE] AVX1 PR28129 (256-bit all-ones rematerialization) Further perf tests on Jaguar indicate that: vxorps %ymm0, %ymm0, %ymm0 vcmpps $15, %ymm0, %ymm0, %ymm0 is consistently faster (by about 9%) than: vpcmpeqd %xmm0, %xmm0, %xmm0 vinsertf128 $1, %xmm0, %ymm0, %ymm0 Testing equivalent code on a SandyBridge (E5-2640) puts it slightly (~3%) faster as well. Committed on behalf of @dtemirbulatov Differential Revision: https://reviews.llvm.org/D32416 llvm-svn: 302989
2017-05-13 21:42:35 +08:00
; X86-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X86-NEXT: vcmptrueps %ymm1, %ymm1, %ymm1
[X86][SSE] Replace insert_vector_elt(vec, -1, idx) with shuffle Similar to what we already do for zero elt insertion, we can quickly rematerialize 'allbits' vectors so to avoid a unnecessary gpr value and insertion into a vector llvm-svn: 294162
2017-02-06 06:50:29 +08:00
; X86-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6,7]
[X86][AVX] Add 8i32 -> 8f32 sitofp tests with constant insertion Some elements are constant inserted into the source integer vector before conversion. llvm-svn: 294147
2017-02-06 05:40:25 +08:00
; X86-NEXT: vcvtdq2ps %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: sitofp_insert_allbits_v8i32:
; X64: # BB#0:
[x86, SSE] AVX1 PR28129 (256-bit all-ones rematerialization) Further perf tests on Jaguar indicate that: vxorps %ymm0, %ymm0, %ymm0 vcmpps $15, %ymm0, %ymm0, %ymm0 is consistently faster (by about 9%) than: vpcmpeqd %xmm0, %xmm0, %xmm0 vinsertf128 $1, %xmm0, %ymm0, %ymm0 Testing equivalent code on a SandyBridge (E5-2640) puts it slightly (~3%) faster as well. Committed on behalf of @dtemirbulatov Differential Revision: https://reviews.llvm.org/D32416 llvm-svn: 302989
2017-05-13 21:42:35 +08:00
; X64-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X64-NEXT: vcmptrueps %ymm1, %ymm1, %ymm1
[X86][SSE] Replace insert_vector_elt(vec, -1, idx) with shuffle Similar to what we already do for zero elt insertion, we can quickly rematerialize 'allbits' vectors so to avoid a unnecessary gpr value and insertion into a vector llvm-svn: 294162
2017-02-06 06:50:29 +08:00
; X64-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6,7]
[X86][AVX] Add 8i32 -> 8f32 sitofp tests with constant insertion Some elements are constant inserted into the source integer vector before conversion. llvm-svn: 294147
2017-02-06 05:40:25 +08:00
; X64-NEXT: vcvtdq2ps %ymm0, %ymm0
; X64-NEXT: retq
%1 = insertelement <8 x i32> %a0, i32 -1, i32 0
%2 = insertelement <8 x i32> %1, i32 -1, i32 2
%3 = insertelement <8 x i32> %2, i32 -1, i32 4
%4 = insertelement <8 x i32> %3, i32 -1, i32 5
%5 = sitofp <8 x i32> %4 to <8 x float>
ret <8 x float> %5
}
define <8 x float> @sitofp_shuffle_allbits_v8i32(<8 x i32> %a0) {
; X86-LABEL: sitofp_shuffle_allbits_v8i32:
; X86: # BB#0:
[x86, SSE] AVX1 PR28129 (256-bit all-ones rematerialization) Further perf tests on Jaguar indicate that: vxorps %ymm0, %ymm0, %ymm0 vcmpps $15, %ymm0, %ymm0, %ymm0 is consistently faster (by about 9%) than: vpcmpeqd %xmm0, %xmm0, %xmm0 vinsertf128 $1, %xmm0, %ymm0, %ymm0 Testing equivalent code on a SandyBridge (E5-2640) puts it slightly (~3%) faster as well. Committed on behalf of @dtemirbulatov Differential Revision: https://reviews.llvm.org/D32416 llvm-svn: 302989
2017-05-13 21:42:35 +08:00
; X86-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X86-NEXT: vcmptrueps %ymm1, %ymm1, %ymm1
[X86][AVX] Add 8i32 -> 8f32 sitofp tests with constant insertion Some elements are constant inserted into the source integer vector before conversion. llvm-svn: 294147
2017-02-06 05:40:25 +08:00
; X86-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7]
; X86-NEXT: vcvtdq2ps %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: sitofp_shuffle_allbits_v8i32:
; X64: # BB#0:
[x86, SSE] AVX1 PR28129 (256-bit all-ones rematerialization) Further perf tests on Jaguar indicate that: vxorps %ymm0, %ymm0, %ymm0 vcmpps $15, %ymm0, %ymm0, %ymm0 is consistently faster (by about 9%) than: vpcmpeqd %xmm0, %xmm0, %xmm0 vinsertf128 $1, %xmm0, %ymm0, %ymm0 Testing equivalent code on a SandyBridge (E5-2640) puts it slightly (~3%) faster as well. Committed on behalf of @dtemirbulatov Differential Revision: https://reviews.llvm.org/D32416 llvm-svn: 302989
2017-05-13 21:42:35 +08:00
; X64-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X64-NEXT: vcmptrueps %ymm1, %ymm1, %ymm1
[X86][AVX] Add 8i32 -> 8f32 sitofp tests with constant insertion Some elements are constant inserted into the source integer vector before conversion. llvm-svn: 294147
2017-02-06 05:40:25 +08:00
; X64-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7]
; X64-NEXT: vcvtdq2ps %ymm0, %ymm0
; X64-NEXT: retq
%1 = shufflevector <8 x i32> %a0, <8 x i32> <i32 -1, i32 undef, i32 -1, i32 undef, i32 -1, i32 undef, i32 -1, i32 undef>, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
%2 = sitofp <8 x i32> %1 to <8 x float>
ret <8 x float> %2
}
define <8 x float> @sitofp_insert_constants_v8i32(<8 x i32> %a0) {
; X86-LABEL: sitofp_insert_constants_v8i32:
; X86: # BB#0:
; X86-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X86-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7]
[x86, SSE] AVX1 PR28129 (256-bit all-ones rematerialization) Further perf tests on Jaguar indicate that: vxorps %ymm0, %ymm0, %ymm0 vcmpps $15, %ymm0, %ymm0, %ymm0 is consistently faster (by about 9%) than: vpcmpeqd %xmm0, %xmm0, %xmm0 vinsertf128 $1, %xmm0, %ymm0, %ymm0 Testing equivalent code on a SandyBridge (E5-2640) puts it slightly (~3%) faster as well. Committed on behalf of @dtemirbulatov Differential Revision: https://reviews.llvm.org/D32416 llvm-svn: 302989
2017-05-13 21:42:35 +08:00
; X86-NEXT: vcmptrueps %ymm1, %ymm1, %ymm1
[X86][SSE] Replace insert_vector_elt(vec, -1, idx) with shuffle Similar to what we already do for zero elt insertion, we can quickly rematerialize 'allbits' vectors so to avoid a unnecessary gpr value and insertion into a vector llvm-svn: 294162
2017-02-06 06:50:29 +08:00
; X86-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4,5,6,7]
[X86][AVX] Add 8i32 -> 8f32 sitofp tests with constant insertion Some elements are constant inserted into the source integer vector before conversion. llvm-svn: 294147
2017-02-06 05:40:25 +08:00
; X86-NEXT: vextractf128 $1, %ymm0, %xmm1
; X86-NEXT: movl $2, %eax
; X86-NEXT: vpinsrd $0, %eax, %xmm1, %xmm1
; X86-NEXT: movl $-3, %eax
; X86-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1
; X86-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; X86-NEXT: vcvtdq2ps %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: sitofp_insert_constants_v8i32:
; X64: # BB#0:
; X64-NEXT: vxorps %ymm1, %ymm1, %ymm1
; X64-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7]
[x86, SSE] AVX1 PR28129 (256-bit all-ones rematerialization) Further perf tests on Jaguar indicate that: vxorps %ymm0, %ymm0, %ymm0 vcmpps $15, %ymm0, %ymm0, %ymm0 is consistently faster (by about 9%) than: vpcmpeqd %xmm0, %xmm0, %xmm0 vinsertf128 $1, %xmm0, %ymm0, %ymm0 Testing equivalent code on a SandyBridge (E5-2640) puts it slightly (~3%) faster as well. Committed on behalf of @dtemirbulatov Differential Revision: https://reviews.llvm.org/D32416 llvm-svn: 302989
2017-05-13 21:42:35 +08:00
; X64-NEXT: vcmptrueps %ymm1, %ymm1, %ymm1
[X86][SSE] Replace insert_vector_elt(vec, -1, idx) with shuffle Similar to what we already do for zero elt insertion, we can quickly rematerialize 'allbits' vectors so to avoid a unnecessary gpr value and insertion into a vector llvm-svn: 294162
2017-02-06 06:50:29 +08:00
; X64-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4,5,6,7]
[X86][AVX] Add 8i32 -> 8f32 sitofp tests with constant insertion Some elements are constant inserted into the source integer vector before conversion. llvm-svn: 294147
2017-02-06 05:40:25 +08:00
; X64-NEXT: vextractf128 $1, %ymm0, %xmm1
; X64-NEXT: movl $2, %eax
; X64-NEXT: vpinsrd $0, %eax, %xmm1, %xmm1
; X64-NEXT: movl $-3, %eax
; X64-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1
; X64-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; X64-NEXT: vcvtdq2ps %ymm0, %ymm0
; X64-NEXT: retq
%1 = insertelement <8 x i32> %a0, i32 0, i32 0
%2 = insertelement <8 x i32> %1, i32 -1, i32 2
%3 = insertelement <8 x i32> %2, i32 2, i32 4
%4 = insertelement <8 x i32> %3, i32 -3, i32 5
%5 = sitofp <8 x i32> %4 to <8 x float>
ret <8 x float> %5
}
define <8 x float> @sitofp_shuffle_constants_v8i32(<8 x i32> %a0) {
; X86-LABEL: sitofp_shuffle_constants_v8i32:
; X86: # BB#0:
; X86-NEXT: vblendps {{.*#+}} ymm0 = mem[0],ymm0[1],mem[2],ymm0[3],mem[4],ymm0[5],mem[6],ymm0[7]
; X86-NEXT: vcvtdq2ps %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: sitofp_shuffle_constants_v8i32:
; X64: # BB#0:
; X64-NEXT: vblendps {{.*#+}} ymm0 = mem[0],ymm0[1],mem[2],ymm0[3],mem[4],ymm0[5],mem[6],ymm0[7]
; X64-NEXT: vcvtdq2ps %ymm0, %ymm0
; X64-NEXT: retq
%1 = shufflevector <8 x i32> %a0, <8 x i32> <i32 0, i32 undef, i32 -1, i32 undef, i32 2, i32 undef, i32 -3, i32 undef>, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
%2 = sitofp <8 x i32> %1 to <8 x float>
ret <8 x float> %2
}