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/vector-shuffle-combining-av...

256 lines
14 KiB
LLVM
Raw Normal View History

[X86] Don't assume that shuffle non-mask operands starts at #0. That's not the case for VPERMV/VPERMV3, which cover all possible combinations (the C intrinsics use a different order; the AVX vs AVX512 intrinsics are different still). Since: r246981 AVX-512: Lowering for 512-bit vector shuffles. VPERMV is recognized in getTargetShuffleMask. This breaks assumptions in most callers, as they expect the non-mask operands to start at index 0. VPERMV has the mask as operand #0; VPERMV3 has it in the middle. Instead of the faulty assumption, have getTargetShuffleMask return its operands as well. One alternative we considered was to change the operand order of VPERMV, but we agreed to stick to the instruction order, as there are more AVX512 weirdness to cover (vpermt2/vpermi2 in particular). Differential Revision: http://reviews.llvm.org/D17041 llvm-svn: 262627
2016-03-04 00:53:50 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s
declare <8 x i32> @llvm.x86.avx2.permd(<8 x i32>, <8 x i32>)
declare <8 x float> @llvm.x86.avx2.permps(<8 x float>, <8 x i32>)
[X86][AVX2] Add support for target shuffle combining to BROADCAST Only support broadcast from vector register so far - memory folding support will have to wait. llvm-svn: 274572
2016-07-06 04:11:29 +08:00
declare <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8>, <16 x i8>)
[X86][AVX2] Added 128-bit wide shuffle test Demonstrate missing 128-bit wide shuffle combine support llvm-svn: 268290
2016-05-03 03:46:58 +08:00
declare <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8>, <32 x i8>)
[X86] Don't assume that shuffle non-mask operands starts at #0. That's not the case for VPERMV/VPERMV3, which cover all possible combinations (the C intrinsics use a different order; the AVX vs AVX512 intrinsics are different still). Since: r246981 AVX-512: Lowering for 512-bit vector shuffles. VPERMV is recognized in getTargetShuffleMask. This breaks assumptions in most callers, as they expect the non-mask operands to start at index 0. VPERMV has the mask as operand #0; VPERMV3 has it in the middle. Instead of the faulty assumption, have getTargetShuffleMask return its operands as well. One alternative we considered was to change the operand order of VPERMV, but we agreed to stick to the instruction order, as there are more AVX512 weirdness to cover (vpermt2/vpermi2 in particular). Differential Revision: http://reviews.llvm.org/D17041 llvm-svn: 262627
2016-03-04 00:53:50 +08:00
[X86][AVX2] Added PSLLDQ/PSRLDQ shuffle combining tests llvm-svn: 272469
2016-06-11 21:18:21 +08:00
define <32 x i8> @combine_pshufb_pslldq(<32 x i8> %a0) {
; CHECK-LABEL: combine_pshufb_pslldq:
; CHECK: # BB#0:
[X86][SSE] Added PSLLDQ/PSRLDQ as a target shuffle type Ensure that PALIGNR/PSLLDQ/PSRLDQ are byte vectors so that they can be correctly decoded for target shuffle combining llvm-svn: 272471
2016-06-11 21:38:28 +08:00
; CHECK-NEXT: vxorps %ymm0, %ymm0, %ymm0
[X86][AVX2] Added PSLLDQ/PSRLDQ shuffle combining tests llvm-svn: 272469
2016-06-11 21:18:21 +08:00
; CHECK-NEXT: retq
%1 = tail call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %a0, <32 x i8> <i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7>)
%2 = shufflevector <32 x i8> %1, <32 x i8> zeroinitializer, <32 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
ret <32 x i8> %2
}
define <32 x i8> @combine_pshufb_psrldq(<32 x i8> %a0) {
; CHECK-LABEL: combine_pshufb_psrldq:
; CHECK: # BB#0:
[X86][SSE] Added PSLLDQ/PSRLDQ as a target shuffle type Ensure that PALIGNR/PSLLDQ/PSRLDQ are byte vectors so that they can be correctly decoded for target shuffle combining llvm-svn: 272471
2016-06-11 21:38:28 +08:00
; CHECK-NEXT: vxorps %ymm0, %ymm0, %ymm0
[X86][AVX2] Added PSLLDQ/PSRLDQ shuffle combining tests llvm-svn: 272469
2016-06-11 21:18:21 +08:00
; CHECK-NEXT: retq
%1 = tail call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %a0, <32 x i8> <i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128>)
%2 = shufflevector <32 x i8> %1, <32 x i8> zeroinitializer, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
ret <32 x i8> %2
}
[X86] Don't assume that shuffle non-mask operands starts at #0. That's not the case for VPERMV/VPERMV3, which cover all possible combinations (the C intrinsics use a different order; the AVX vs AVX512 intrinsics are different still). Since: r246981 AVX-512: Lowering for 512-bit vector shuffles. VPERMV is recognized in getTargetShuffleMask. This breaks assumptions in most callers, as they expect the non-mask operands to start at index 0. VPERMV has the mask as operand #0; VPERMV3 has it in the middle. Instead of the faulty assumption, have getTargetShuffleMask return its operands as well. One alternative we considered was to change the operand order of VPERMV, but we agreed to stick to the instruction order, as there are more AVX512 weirdness to cover (vpermt2/vpermi2 in particular). Differential Revision: http://reviews.llvm.org/D17041 llvm-svn: 262627
2016-03-04 00:53:50 +08:00
define <32 x i8> @combine_pshufb_vpermd(<8 x i32> %a) {
; CHECK-LABEL: combine_pshufb_vpermd:
; CHECK: # BB#0:
; CHECK-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,16,17,18,18]
; CHECK-NEXT: retq
%tmp0 = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %a, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 4>)
%tmp1 = bitcast <8 x i32> %tmp0 to <32 x i8>
%tmp2 = shufflevector <32 x i8> %tmp1, <32 x i8> undef, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 30>
ret <32 x i8> %tmp2
}
define <32 x i8> @combine_pshufb_vpermps(<8 x float> %a) {
; CHECK-LABEL: combine_pshufb_vpermps:
; CHECK: # BB#0:
; CHECK-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,16,17,18,18]
; CHECK-NEXT: retq
%tmp0 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %a, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 4>)
%tmp1 = bitcast <8 x float> %tmp0 to <32 x i8>
%tmp2 = shufflevector <32 x i8> %tmp1, <32 x i8> undef, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 30>
ret <32 x i8> %tmp2
}
[X86][AVX2] Added 128-bit wide shuffle test Demonstrate missing 128-bit wide shuffle combine support llvm-svn: 268290
2016-05-03 03:46:58 +08:00
define <4 x i64> @combine_permq_pshufb(<4 x i64> %a0) {
; CHECK-LABEL: combine_permq_pshufb:
; CHECK: # BB#0:
; CHECK-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,2,1,0]
[X86][SSE] Added support for combining target shuffles to (V)PSHUFD/VPERMILPD/VPERMILPS immediate permutes This patch allows target shuffles to be combined to single input immediate permute instructions - (V)PSHUFD/VPERMILPD/VPERMILPS - allowing more general pattern matching than what we current do and improves the likelihood of memory folding compared to existing patterns which tend to reuse the input in multiple arguments. Further permute instructions (V)PSHUFLW/(V)PSHUFHW/(V)PERMQ/(V)PERMPD may be added in the future but its proven tricky to create tests cases for them so far. (V)PSHUFLW/(V)PSHUFHW is already handled quite well in combineTargetShuffle so it may be that removing some of that code may allow us to perform more of the combining in one place without duplication. Differential Revision: http://reviews.llvm.org/D21148 llvm-svn: 273999
2016-06-28 16:08:15 +08:00
; CHECK-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
[X86][AVX2] Added 128-bit wide shuffle test Demonstrate missing 128-bit wide shuffle combine support llvm-svn: 268290
2016-05-03 03:46:58 +08:00
; CHECK-NEXT: retq
%1 = shufflevector <4 x i64> %a0, <4 x i64> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
%2 = bitcast <4 x i64> %1 to <32 x i8>
%3 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %2, <32 x i8> <i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7>)
%4 = bitcast <32 x i8> %3 to <4 x i64>
ret <4 x i64> %4
}
[X86][AVX2] Add support for target shuffle combining to BROADCAST Only support broadcast from vector register so far - memory folding support will have to wait. llvm-svn: 274572
2016-07-06 04:11:29 +08:00
define <16 x i8> @combine_pshufb_as_vpbroadcastb128(<16 x i8> %a) {
; CHECK-LABEL: combine_pshufb_as_vpbroadcastb128:
; CHECK: # BB#0:
; CHECK-NEXT: vpbroadcastb %xmm0, %xmm0
; CHECK-NEXT: retq
%1 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a, <16 x i8> zeroinitializer)
ret <16 x i8> %1
}
define <32 x i8> @combine_pshufb_as_vpbroadcastb256(<2 x i64> %a) {
; CHECK-LABEL: combine_pshufb_as_vpbroadcastb256:
; CHECK: # BB#0:
VirtRegMap: Replace some identity copies with KILL instructions. An identity COPY like this: %AL = COPY %AL, %EAX<imp-def> has no semantic effect, but encodes liveness information: Further users of %EAX only depend on this instruction even though it does not define the full register. Replace the COPY with a KILL instruction in those cases to maintain this liveness information. (This reverts a small part of r238588 but this time adds a comment explaining why a KILL instruction is useful). llvm-svn: 274952
2016-07-09 08:19:07 +08:00
; CHECK-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<def>
[X86][AVX2] Add support for target shuffle combining to BROADCAST Only support broadcast from vector register so far - memory folding support will have to wait. llvm-svn: 274572
2016-07-06 04:11:29 +08:00
; CHECK-NEXT: vpbroadcastb %xmm0, %ymm0
; CHECK-NEXT: retq
%1 = shufflevector <2 x i64> %a, <2 x i64> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
%2 = bitcast <4 x i64> %1 to <32 x i8>
%3 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %2, <32 x i8> zeroinitializer)
%4 = bitcast <32 x i8> %3 to <8 x i32>
%5 = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %4, <8 x i32> zeroinitializer)
%6 = bitcast <8 x i32> %5 to <32 x i8>
ret <32 x i8> %6
}
define <16 x i8> @combine_pshufb_as_vpbroadcastw128(<16 x i8> %a) {
; CHECK-LABEL: combine_pshufb_as_vpbroadcastw128:
; CHECK: # BB#0:
; CHECK-NEXT: vpbroadcastw %xmm0, %xmm0
; CHECK-NEXT: retq
%1 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a, <16 x i8> <i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1>)
ret <16 x i8> %1
}
define <32 x i8> @combine_pshufb_as_vpbroadcastw256(<2 x i64> %a) {
; CHECK-LABEL: combine_pshufb_as_vpbroadcastw256:
; CHECK: # BB#0:
VirtRegMap: Replace some identity copies with KILL instructions. An identity COPY like this: %AL = COPY %AL, %EAX<imp-def> has no semantic effect, but encodes liveness information: Further users of %EAX only depend on this instruction even though it does not define the full register. Replace the COPY with a KILL instruction in those cases to maintain this liveness information. (This reverts a small part of r238588 but this time adds a comment explaining why a KILL instruction is useful). llvm-svn: 274952
2016-07-09 08:19:07 +08:00
; CHECK-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<def>
[X86][AVX2] Add support for target shuffle combining to BROADCAST Only support broadcast from vector register so far - memory folding support will have to wait. llvm-svn: 274572
2016-07-06 04:11:29 +08:00
; CHECK-NEXT: vpbroadcastw %xmm0, %ymm0
; CHECK-NEXT: retq
%1 = shufflevector <2 x i64> %a, <2 x i64> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
%2 = bitcast <4 x i64> %1 to <32 x i8>
%3 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %2, <32 x i8> <i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1>)
%4 = bitcast <32 x i8> %3 to <8 x i32>
%5 = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %4, <8 x i32> zeroinitializer)
%6 = bitcast <8 x i32> %5 to <32 x i8>
ret <32 x i8> %6
}
define <16 x i8> @combine_pshufb_as_vpbroadcastd128(<16 x i8> %a) {
; CHECK-LABEL: combine_pshufb_as_vpbroadcastd128:
; CHECK: # BB#0:
; CHECK-NEXT: vpbroadcastd %xmm0, %xmm0
; CHECK-NEXT: vpaddb {{.*}}(%rip), %xmm0, %xmm0
; CHECK-NEXT: retq
%1 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3>)
%2 = add <16 x i8> %1, <i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3>
ret <16 x i8> %2
}
define <8 x i32> @combine_permd_as_vpbroadcastd256(<4 x i32> %a) {
; CHECK-LABEL: combine_permd_as_vpbroadcastd256:
; CHECK: # BB#0:
VirtRegMap: Replace some identity copies with KILL instructions. An identity COPY like this: %AL = COPY %AL, %EAX<imp-def> has no semantic effect, but encodes liveness information: Further users of %EAX only depend on this instruction even though it does not define the full register. Replace the COPY with a KILL instruction in those cases to maintain this liveness information. (This reverts a small part of r238588 but this time adds a comment explaining why a KILL instruction is useful). llvm-svn: 274952
2016-07-09 08:19:07 +08:00
; CHECK-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<def>
[X86][AVX2] Add support for target shuffle combining to BROADCAST Only support broadcast from vector register so far - memory folding support will have to wait. llvm-svn: 274572
2016-07-06 04:11:29 +08:00
; CHECK-NEXT: vpbroadcastd %xmm0, %ymm0
; CHECK-NEXT: vpaddd {{.*}}(%rip), %ymm0, %ymm0
; CHECK-NEXT: retq
%1 = shufflevector <4 x i32> %a, <4 x i32> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%2 = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %1, <8 x i32> zeroinitializer)
%3 = add <8 x i32> %2, <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %3
}
define <16 x i8> @combine_pshufb_as_vpbroadcastq128(<16 x i8> %a) {
; CHECK-LABEL: combine_pshufb_as_vpbroadcastq128:
; CHECK: # BB#0:
; CHECK-NEXT: vpbroadcastq %xmm0, %xmm0
; CHECK-NEXT: retq
%1 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7>)
ret <16 x i8> %1
}
define <8 x i32> @combine_permd_as_vpbroadcastq256(<4 x i32> %a) {
; CHECK-LABEL: combine_permd_as_vpbroadcastq256:
; CHECK: # BB#0:
VirtRegMap: Replace some identity copies with KILL instructions. An identity COPY like this: %AL = COPY %AL, %EAX<imp-def> has no semantic effect, but encodes liveness information: Further users of %EAX only depend on this instruction even though it does not define the full register. Replace the COPY with a KILL instruction in those cases to maintain this liveness information. (This reverts a small part of r238588 but this time adds a comment explaining why a KILL instruction is useful). llvm-svn: 274952
2016-07-09 08:19:07 +08:00
; CHECK-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<def>
[X86][AVX2] Add support for target shuffle combining to BROADCAST Only support broadcast from vector register so far - memory folding support will have to wait. llvm-svn: 274572
2016-07-06 04:11:29 +08:00
; CHECK-NEXT: vpbroadcastq %xmm0, %ymm0
; CHECK-NEXT: vpaddd {{.*}}(%rip), %ymm0, %ymm0
; CHECK-NEXT: retq
%1 = shufflevector <4 x i32> %a, <4 x i32> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%2 = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %1, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>)
%3 = add <8 x i32> %2, <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %3
}
define <4 x float> @combine_pshufb_as_vpbroadcastss128(<4 x float> %a) {
; CHECK-LABEL: combine_pshufb_as_vpbroadcastss128:
; CHECK: # BB#0:
; CHECK-NEXT: vbroadcastss %xmm0, %xmm0
; CHECK-NEXT: retq
%1 = bitcast <4 x float> %a to <16 x i8>
%2 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %1, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3, i8 0, i8 1, i8 2, i8 3>)
%3 = bitcast <16 x i8> %2 to <4 x float>
ret <4 x float> %3
}
define <8 x float> @combine_permd_as_vpbroadcastss256(<4 x float> %a) {
; CHECK-LABEL: combine_permd_as_vpbroadcastss256:
; CHECK: # BB#0:
VirtRegMap: Replace some identity copies with KILL instructions. An identity COPY like this: %AL = COPY %AL, %EAX<imp-def> has no semantic effect, but encodes liveness information: Further users of %EAX only depend on this instruction even though it does not define the full register. Replace the COPY with a KILL instruction in those cases to maintain this liveness information. (This reverts a small part of r238588 but this time adds a comment explaining why a KILL instruction is useful). llvm-svn: 274952
2016-07-09 08:19:07 +08:00
; CHECK-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<def>
[X86][AVX2] Add support for target shuffle combining to BROADCAST Only support broadcast from vector register so far - memory folding support will have to wait. llvm-svn: 274572
2016-07-06 04:11:29 +08:00
; CHECK-NEXT: vbroadcastss %xmm0, %ymm0
; CHECK-NEXT: retq
%1 = shufflevector <4 x float> %a, <4 x float> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
%2 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %1, <8 x i32> zeroinitializer)
ret <8 x float> %2
}
define <4 x double> @combine_permd_as_vpbroadcastsd256(<2 x double> %a) {
; CHECK-LABEL: combine_permd_as_vpbroadcastsd256:
; CHECK: # BB#0:
VirtRegMap: Replace some identity copies with KILL instructions. An identity COPY like this: %AL = COPY %AL, %EAX<imp-def> has no semantic effect, but encodes liveness information: Further users of %EAX only depend on this instruction even though it does not define the full register. Replace the COPY with a KILL instruction in those cases to maintain this liveness information. (This reverts a small part of r238588 but this time adds a comment explaining why a KILL instruction is useful). llvm-svn: 274952
2016-07-09 08:19:07 +08:00
; CHECK-NEXT: # kill: %XMM0<def> %XMM0<kill> %YMM0<def>
[X86][AVX2] Add support for target shuffle combining to BROADCAST Only support broadcast from vector register so far - memory folding support will have to wait. llvm-svn: 274572
2016-07-06 04:11:29 +08:00
; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0
; CHECK-NEXT: retq
%1 = shufflevector <2 x double> %a, <2 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
%2 = bitcast <4 x double> %1 to <8 x float>
%3 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %2, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>)
%4 = bitcast <8 x float> %3 to <4 x double>
ret <4 x double> %4
}
[X86][SSE] Added test cases for missed opportunities to combine pshufb to pslldq/psrldq llvm-svn: 274631
2016-07-06 23:09:48 +08:00
[X86][AVX2] Add support for target shuffle combining to VPERMPD/VPERMQ llvm-svn: 274908
2016-07-09 03:23:29 +08:00
define <8 x i32> @combine_permd_as_permq(<8 x i32> %a) {
; CHECK-LABEL: combine_permd_as_permq:
; CHECK: # BB#0:
; CHECK-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,1]
; CHECK-NEXT: retq
%1 = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %a, <8 x i32> <i32 0, i32 1, i32 4, i32 5, i32 4, i32 5, i32 2, i32 3>)
ret <8 x i32> %1
}
define <8 x float> @combine_permps_as_permpd(<8 x float> %a) {
; CHECK-LABEL: combine_permps_as_permpd:
; CHECK: # BB#0:
; CHECK-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,2,0,1]
; CHECK-NEXT: retq
%1 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %a, <8 x i32> <i32 6, i32 7, i32 4, i32 5, i32 0, i32 1, i32 2, i32 3>)
ret <8 x float> %1
}
[X86][SSE] Added test cases for missed opportunities to combine pshufb to pslldq/psrldq llvm-svn: 274631
2016-07-06 23:09:48 +08:00
define <32 x i8> @combine_pshufb_as_pslldq(<32 x i8> %a0) {
; CHECK-LABEL: combine_pshufb_as_pslldq:
; CHECK: # BB#0:
; CHECK-NEXT: vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[0,1,2,3,4,5],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,18,19,20,21]
; CHECK-NEXT: retq
%res0 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %a0, <32 x i8> <i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5>)
ret <32 x i8> %res0
}
define <32 x i8> @combine_pshufb_as_psrldq(<32 x i8> %a0) {
; CHECK-LABEL: combine_pshufb_as_psrldq:
; CHECK: # BB#0:
; CHECK-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[31],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
; CHECK-NEXT: retq
%res0 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %a0, <32 x i8> <i8 15, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 15, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128, i8 128>)
ret <32 x i8> %res0
}
[X86][SSE] Added tests for combining shuffles to PSHUFLW/PSHUFHW llvm-svn: 275019
2016-07-11 04:19:56 +08:00
define <32 x i8> @combine_pshufb_as_pshuflw(<32 x i8> %a0) {
; CHECK-LABEL: combine_pshufb_as_pshuflw:
; CHECK: # BB#0:
; CHECK-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5,8,9,10,11,12,13,14,15,18,19,16,17,22,23,20,21,24,25,26,27,28,29,30,31]
; CHECK-NEXT: retq
%res0 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %a0, <32 x i8> <i8 2, i8 3, i8 0, i8 1, i8 6, i8 7, i8 4, i8 5, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 2, i8 3, i8 0, i8 1, i8 6, i8 7, i8 4, i8 5, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>)
ret <32 x i8> %res0
}
define <32 x i8> @combine_pshufb_as_pshufhw(<32 x i8> %a0) {
; CHECK-LABEL: combine_pshufb_as_pshufhw:
; CHECK: # BB#0:
; CHECK-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6,7,10,11,8,9,14,15,12,13,16,17,18,19,20,21,22,23,26,27,24,25,30,31,28,29]
; CHECK-NEXT: retq
%res0 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %a0, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 10, i8 11, i8 8, i8 9, i8 14, i8 15, i8 12, i8 13, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 10, i8 11, i8 8, i8 9, i8 14, i8 15, i8 12, i8 13>)
ret <32 x i8> %res0
}
define <32 x i8> @combine_pshufb_as_pshufw(<32 x i8> %a0) {
; CHECK-LABEL: combine_pshufb_as_pshufw:
; CHECK: # BB#0:
; CHECK-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13,18,19,16,17,22,23,20,21,26,27,24,25,30,31,28,29]
; CHECK-NEXT: retq
%res0 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %a0, <32 x i8> <i8 2, i8 3, i8 0, i8 1, i8 6, i8 7, i8 4, i8 5, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 2, i8 3, i8 0, i8 1, i8 6, i8 7, i8 4, i8 5, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>)
%res1 = call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %res0, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 10, i8 11, i8 8, i8 9, i8 14, i8 15, i8 12, i8 13, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 10, i8 11, i8 8, i8 9, i8 14, i8 15, i8 12, i8 13>)
ret <32 x i8> %res1
}