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/bitcast-and-setcc-512.ll

660 lines
26 KiB
LLVM
Raw Normal View History

Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX12,AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX12,AVX2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512BW
define i8 @v8i64(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {
; SSE-LABEL: v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; SSE: # %bb.0:
[X86][SSE] Add support for lowering unary shuffles to PACKSS/PACKUS Extension to D38472 llvm-svn: 314901
2017-10-04 21:12:08 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm11
[X86][SSE] Add support for lowering unary shuffles to PACKSS/PACKUS Extension to D38472 llvm-svn: 314901
2017-10-04 21:12:08 +08:00
; SSE-NEXT: pcmpgtq %xmm7, %xmm3
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,2,4,5,6,7]
[X86][SSE] Add support for lowering unary shuffles to PACKSS/PACKUS Extension to D38472 llvm-svn: 314901
2017-10-04 21:12:08 +08:00
; SSE-NEXT: pcmpgtq %xmm6, %xmm2
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,1,0,2,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
[SelectionDAG] When legalizing the result type of CONCAT_VECTORS, take into account whether the input type also needs to be promoted. If so go ahead and get the promoted input vector to extract from. Previously, we would create a bunch of any_extends of extract_vector_elts with illegal input type that needs to be promoted. The legalization of those extract_vector_elts would then potentially introduce a truncate. So now we have a bunch of any_extends of truncates. By legalizing both parts together we avoid creating these extra nodes. The test changes seem to be because we were previously combining the build_vector with the any_extend before the any_extend got combined with the truncate. llvm-svn: 320669
2017-12-14 14:49:07 +08:00
; SSE-NEXT: pcmpgtq %xmm5, %xmm1
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
[SelectionDAG] When legalizing the result type of CONCAT_VECTORS, take into account whether the input type also needs to be promoted. If so go ahead and get the promoted input vector to extract from. Previously, we would create a bunch of any_extends of extract_vector_elts with illegal input type that needs to be promoted. The legalization of those extract_vector_elts would then potentially introduce a truncate. So now we have a bunch of any_extends of truncates. By legalizing both parts together we avoid creating these extra nodes. The test changes seem to be because we were previously combining the build_vector with the any_extend before the any_extend got combined with the truncate. llvm-svn: 320669
2017-12-14 14:49:07 +08:00
; SSE-NEXT: pcmpgtq %xmm4, %xmm0
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtq {{[0-9]+}}(%rsp), %xmm11
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm11[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,2,4,5,6,7]
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: pcmpgtq {{[0-9]+}}(%rsp), %xmm10
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufd {{.*#+}} xmm2 = xmm10[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[0,1,0,2,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: pcmpgtq {{[0-9]+}}(%rsp), %xmm9
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm9[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
[X86][SSE] Add support for lowering unary shuffles to PACKSS/PACKUS Extension to D38472 llvm-svn: 314901
2017-10-04 21:12:08 +08:00
; SSE-NEXT: pcmpgtq {{[0-9]+}}(%rsp), %xmm8
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufd {{.*#+}} xmm3 = xmm8[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; SSE-NEXT: pand %xmm0, %xmm3
; SSE-NEXT: packsswb %xmm0, %xmm3
; SSE-NEXT: pmovmskb %xmm3, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; SSE-NEXT: # kill: def $al killed $al killed $eax
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; SSE-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX1-LABEL: v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX1: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm8
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm9
; AVX1-NEXT: vpcmpgtq %xmm8, %xmm9, %xmm8
; AVX1-NEXT: vpcmpgtq %xmm3, %xmm1, %xmm1
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX1-NEXT: vpackssdw %xmm8, %xmm1, %xmm8
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT: vpcmpgtq %xmm3, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtq %xmm2, %xmm0, %xmm0
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX1-NEXT: vpackssdw %xmm1, %xmm0, %xmm0
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX1-NEXT: vpackssdw %xmm8, %xmm0, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm7, %xmm1
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm2
; AVX1-NEXT: vpcmpgtq %xmm1, %xmm2, %xmm1
; AVX1-NEXT: vpcmpgtq %xmm7, %xmm5, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX1-NEXT: vpackssdw %xmm1, %xmm2, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm3
; AVX1-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpcmpgtq %xmm6, %xmm4, %xmm3
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX1-NEXT: vpackssdw %xmm2, %xmm3, %xmm2
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX1-NEXT: vpackssdw %xmm1, %xmm2, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX1-NEXT: vpsllw $15, %xmm0, %xmm0
; AVX1-NEXT: vpsraw $15, %xmm0, %xmm0
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; AVX1-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vpmovmskb %xmm0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX1-NEXT: # kill: def $al killed $al killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX1-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX2-LABEL: v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX2: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpcmpgtq %ymm3, %ymm1, %ymm1
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm3
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX2-NEXT: vpackssdw %xmm3, %xmm1, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpcmpgtq %ymm2, %ymm0, %ymm0
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX2-NEXT: vpackssdw %xmm2, %xmm0, %xmm0
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX2-NEXT: vpackssdw %xmm1, %xmm0, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpcmpgtq %ymm7, %ymm5, %ymm1
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX2-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpcmpgtq %ymm6, %ymm4, %ymm2
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT: vpackssdw %xmm3, %xmm2, %xmm2
; AVX2-NEXT: vpackssdw %xmm1, %xmm2, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX2-NEXT: vpsllw $15, %xmm0, %xmm0
; AVX2-NEXT: vpsraw $15, %xmm0, %xmm0
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; AVX2-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpmovmskb %xmm0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX2-NEXT: # kill: def $al killed $al killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX2-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512F-LABEL: v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512F: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vpcmpgtq %zmm1, %zmm0, %k1
; AVX512F-NEXT: vpcmpgtq %zmm3, %zmm2, %k0 {%k1}
; AVX512F-NEXT: kmovw %k0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512F-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512BW-LABEL: v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vpcmpgtq %zmm1, %zmm0, %k1
; AVX512BW-NEXT: vpcmpgtq %zmm3, %zmm2, %k0 {%k1}
; AVX512BW-NEXT: kmovd %k0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512BW-NEXT: # kill: def $al killed $al killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512BW-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
%x0 = icmp sgt <8 x i64> %a, %b
%x1 = icmp sgt <8 x i64> %c, %d
%y = and <8 x i1> %x0, %x1
%res = bitcast <8 x i1> %y to i8
ret i8 %res
}
define i8 @v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x double> %d) {
; SSE-LABEL: v8f64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; SSE: # %bb.0:
[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types. This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type. Differential Revision: https://reviews.llvm.org/D39289 llvm-svn: 316831
2017-10-28 22:27:53 +08:00
; SSE-NEXT: movapd {{[0-9]+}}(%rsp), %xmm8
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movapd {{[0-9]+}}(%rsp), %xmm9
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: movapd {{[0-9]+}}(%rsp), %xmm10
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movapd {{[0-9]+}}(%rsp), %xmm11
[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types. This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type. Differential Revision: https://reviews.llvm.org/D39289 llvm-svn: 316831
2017-10-28 22:27:53 +08:00
; SSE-NEXT: cmpltpd %xmm3, %xmm7
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm7[0,1,0,2,4,5,6,7]
[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types. This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type. Differential Revision: https://reviews.llvm.org/D39289 llvm-svn: 316831
2017-10-28 22:27:53 +08:00
; SSE-NEXT: cmpltpd %xmm2, %xmm6
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm6[0,1,0,2,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
[SelectionDAG] When legalizing the result type of CONCAT_VECTORS, take into account whether the input type also needs to be promoted. If so go ahead and get the promoted input vector to extract from. Previously, we would create a bunch of any_extends of extract_vector_elts with illegal input type that needs to be promoted. The legalization of those extract_vector_elts would then potentially introduce a truncate. So now we have a bunch of any_extends of truncates. By legalizing both parts together we avoid creating these extra nodes. The test changes seem to be because we were previously combining the build_vector with the any_extend before the any_extend got combined with the truncate. llvm-svn: 320669
2017-12-14 14:49:07 +08:00
; SSE-NEXT: cmpltpd %xmm1, %xmm5
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm5[0,2,2,3,4,5,6,7]
[SelectionDAG] When legalizing the result type of CONCAT_VECTORS, take into account whether the input type also needs to be promoted. If so go ahead and get the promoted input vector to extract from. Previously, we would create a bunch of any_extends of extract_vector_elts with illegal input type that needs to be promoted. The legalization of those extract_vector_elts would then potentially introduce a truncate. So now we have a bunch of any_extends of truncates. By legalizing both parts together we avoid creating these extra nodes. The test changes seem to be because we were previously combining the build_vector with the any_extend before the any_extend got combined with the truncate. llvm-svn: 320669
2017-12-14 14:49:07 +08:00
; SSE-NEXT: cmpltpd %xmm0, %xmm4
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm4[0,2,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm11
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: shufps {{.*#+}} xmm11 = xmm11[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm11[0,1,0,2,4,5,6,7]
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm10
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: shufps {{.*#+}} xmm10 = xmm10[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm2 = xmm10[0,1,0,2,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm9
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm1 = xmm9[0,2,2,3,4,5,6,7]
[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types. This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type. Differential Revision: https://reviews.llvm.org/D39289 llvm-svn: 316831
2017-10-28 22:27:53 +08:00
; SSE-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm8
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,2,2,3]
; SSE-NEXT: pshuflw {{.*#+}} xmm3 = xmm8[0,2,2,3,4,5,6,7]
; SSE-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
; SSE-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; SSE-NEXT: pand %xmm0, %xmm3
; SSE-NEXT: packsswb %xmm0, %xmm3
; SSE-NEXT: pmovmskb %xmm3, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; SSE-NEXT: # kill: def $al killed $al killed $eax
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; SSE-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX12-LABEL: v8f64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX12: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vcmpltpd %ymm1, %ymm3, %ymm1
; AVX12-NEXT: vextractf128 $1, %ymm1, %xmm3
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX12-NEXT: vpackssdw %xmm3, %xmm1, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vcmpltpd %ymm0, %ymm2, %ymm0
; AVX12-NEXT: vextractf128 $1, %ymm0, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX12-NEXT: vpackssdw %xmm2, %xmm0, %xmm0
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX12-NEXT: vpackssdw %xmm1, %xmm0, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vcmpltpd %ymm5, %ymm7, %ymm1
; AVX12-NEXT: vextractf128 $1, %ymm1, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX12-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vcmpltpd %ymm4, %ymm6, %ymm2
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX12-NEXT: vextractf128 $1, %ymm2, %xmm3
; AVX12-NEXT: vpackssdw %xmm3, %xmm2, %xmm2
; AVX12-NEXT: vpackssdw %xmm1, %xmm2, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vpand %xmm1, %xmm0, %xmm0
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; AVX12-NEXT: vpsllw $15, %xmm0, %xmm0
; AVX12-NEXT: vpsraw $15, %xmm0, %xmm0
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; AVX12-NEXT: vpacksswb %xmm0, %xmm0, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vpmovmskb %xmm0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX12-NEXT: # kill: def $al killed $al killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX12-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512F-LABEL: v8f64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512F: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vcmpltpd %zmm0, %zmm1, %k1
; AVX512F-NEXT: vcmpltpd %zmm2, %zmm3, %k0 {%k1}
; AVX512F-NEXT: kmovw %k0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512F-NEXT: # kill: def $al killed $al killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512F-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512BW-LABEL: v8f64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vcmpltpd %zmm0, %zmm1, %k1
; AVX512BW-NEXT: vcmpltpd %zmm2, %zmm3, %k0 {%k1}
; AVX512BW-NEXT: kmovd %k0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512BW-NEXT: # kill: def $al killed $al killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512BW-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
%x0 = fcmp ogt <8 x double> %a, %b
%x1 = fcmp ogt <8 x double> %c, %d
%y = and <8 x i1> %x0, %x1
%res = bitcast <8 x i1> %y to i8
ret i8 %res
}
define i32 @v32i16(<32 x i16> %a, <32 x i16> %b, <32 x i16> %c, <32 x i16> %d) {
; SSE-LABEL: v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; SSE: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm11
; SSE-NEXT: pcmpgtw %xmm5, %xmm1
; SSE-NEXT: pcmpgtw %xmm4, %xmm0
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; SSE-NEXT: packsswb %xmm1, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtw %xmm7, %xmm3
; SSE-NEXT: pcmpgtw %xmm6, %xmm2
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; SSE-NEXT: packsswb %xmm3, %xmm2
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtw {{[0-9]+}}(%rsp), %xmm11
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: pcmpgtw {{[0-9]+}}(%rsp), %xmm10
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; SSE-NEXT: packsswb %xmm11, %xmm10
; SSE-NEXT: pand %xmm0, %xmm10
; SSE-NEXT: pcmpgtw {{[0-9]+}}(%rsp), %xmm9
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtw {{[0-9]+}}(%rsp), %xmm8
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; SSE-NEXT: packsswb %xmm9, %xmm8
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: pand %xmm2, %xmm8
[X86][SSE] Add support for lowering shuffles to PACKSS/PACKUS If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles. Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773 Differential Revision: https://reviews.llvm.org/D38472 llvm-svn: 314788
2017-10-03 20:01:31 +08:00
; SSE-NEXT: pmovmskb %xmm10, %ecx
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: pmovmskb %xmm8, %eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: shll $16, %eax
; SSE-NEXT: orl %ecx, %eax
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; SSE-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX1-LABEL: v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX1: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm8
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm9
; AVX1-NEXT: vpcmpgtw %xmm8, %xmm9, %xmm8
; AVX1-NEXT: vpcmpgtw %xmm3, %xmm1, %xmm1
; AVX1-NEXT: vpacksswb %xmm8, %xmm1, %xmm8
; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT: vpcmpgtw %xmm3, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtw %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vextractf128 $1, %ymm7, %xmm1
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm2
; AVX1-NEXT: vpcmpgtw %xmm1, %xmm2, %xmm1
; AVX1-NEXT: vpcmpgtw %xmm7, %xmm5, %xmm2
; AVX1-NEXT: vpacksswb %xmm1, %xmm2, %xmm1
; AVX1-NEXT: vpand %xmm1, %xmm8, %xmm1
; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm3
; AVX1-NEXT: vpcmpgtw %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpcmpgtw %xmm6, %xmm4, %xmm3
; AVX1-NEXT: vpacksswb %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpmovmskb %xmm0, %ecx
; AVX1-NEXT: vpmovmskb %xmm1, %eax
; AVX1-NEXT: shll $16, %eax
; AVX1-NEXT: orl %ecx, %eax
; AVX1-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX1-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX2-LABEL: v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX2: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpcmpgtw %ymm3, %ymm1, %ymm1
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm3
; AVX2-NEXT: vpacksswb %xmm3, %xmm1, %xmm1
; AVX2-NEXT: vpcmpgtw %ymm2, %ymm0, %ymm0
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-NEXT: vpacksswb %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm0
; AVX2-NEXT: vpcmpgtw %ymm7, %ymm5, %ymm1
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-NEXT: vpacksswb %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpcmpgtw %ymm6, %ymm4, %ymm2
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT: vpacksswb %xmm3, %xmm2, %xmm2
; AVX2-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1
; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
; AVX2-NEXT: vpmovmskb %ymm0, %eax
; AVX2-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX2-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512F-LABEL: v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512F: # %bb.0:
[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8. Summary: For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register. I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes. There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined. I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc)) Overall this removes something like 13000 CHECK lines from lit tests. Reviewers: zvi, RKSimon, delena, spatel Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42031 llvm-svn: 323201
2018-01-23 22:25:39 +08:00
; AVX512F-NEXT: vpcmpgtw %ymm3, %ymm1, %ymm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vpcmpgtw %ymm2, %ymm0, %ymm0
[DAGCombiner] allow hoisting vector bitwise logic ahead of truncates The transform performs a bitwise logic op in a wider type followed by truncate when both inputs are truncated from the same source type: logic_op (truncate x), (truncate y) --> truncate (logic_op x, y) There are a bunch of other checks that should prevent doing this when it might be harmful. We already do this transform for scalars in this spot. The vector limitation was shared with a check for the case when the operands are extended. I'm not sure if that limit is needed either, but that would be a separate patch. Differential Revision: https://reviews.llvm.org/D55448 llvm-svn: 349303
2018-12-16 22:57:04 +08:00
; AVX512F-NEXT: vpcmpgtw %ymm7, %ymm5, %ymm2
; AVX512F-NEXT: vpand %ymm2, %ymm1, %ymm1
; AVX512F-NEXT: vpcmpgtw %ymm6, %ymm4, %ymm2
; AVX512F-NEXT: vpand %ymm2, %ymm0, %ymm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0
[DAGCombiner] allow hoisting vector bitwise logic ahead of truncates The transform performs a bitwise logic op in a wider type followed by truncate when both inputs are truncated from the same source type: logic_op (truncate x), (truncate y) --> truncate (logic_op x, y) There are a bunch of other checks that should prevent doing this when it might be harmful. We already do this transform for scalars in this spot. The vector limitation was shared with a check for the case when the operands are extended. I'm not sure if that limit is needed either, but that would be a separate patch. Differential Revision: https://reviews.llvm.org/D55448 llvm-svn: 349303
2018-12-16 22:57:04 +08:00
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8. Summary: For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register. I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes. There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined. I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc)) Overall this removes something like 13000 CHECK lines from lit tests. Reviewers: zvi, RKSimon, delena, spatel Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42031 llvm-svn: 323201
2018-01-23 22:25:39 +08:00
; AVX512F-NEXT: kmovw %k0, %ecx
[DAGCombiner] allow hoisting vector bitwise logic ahead of truncates The transform performs a bitwise logic op in a wider type followed by truncate when both inputs are truncated from the same source type: logic_op (truncate x), (truncate y) --> truncate (logic_op x, y) There are a bunch of other checks that should prevent doing this when it might be harmful. We already do this transform for scalars in this spot. The vector limitation was shared with a check for the case when the operands are extended. I'm not sure if that limit is needed either, but that would be a separate patch. Differential Revision: https://reviews.llvm.org/D55448 llvm-svn: 349303
2018-12-16 22:57:04 +08:00
; AVX512F-NEXT: vpmovsxwd %ymm1, %zmm0
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8. Summary: For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register. I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes. There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined. I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc)) Overall this removes something like 13000 CHECK lines from lit tests. Reviewers: zvi, RKSimon, delena, spatel Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42031 llvm-svn: 323201
2018-01-23 22:25:39 +08:00
; AVX512F-NEXT: kmovw %k0, %eax
; AVX512F-NEXT: shll $16, %eax
; AVX512F-NEXT: orl %ecx, %eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512F-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512BW-LABEL: v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vpcmpgtw %zmm1, %zmm0, %k1
; AVX512BW-NEXT: vpcmpgtw %zmm3, %zmm2, %k0 {%k1}
; AVX512BW-NEXT: kmovd %k0, %eax
; AVX512BW-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512BW-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
%x0 = icmp sgt <32 x i16> %a, %b
%x1 = icmp sgt <32 x i16> %c, %d
%y = and <32 x i1> %x0, %x1
%res = bitcast <32 x i1> %y to i32
ret i32 %res
}
define i16 @v16i32(<16 x i32> %a, <16 x i32> %b, <16 x i32> %c, <16 x i32> %d) {
; SSE-LABEL: v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; SSE: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8
[X86][SSE] Add support for lowering v8i16 binary shuffles to PACKSS/PACKUS Missed in D38472 llvm-svn: 314916
2017-10-05 01:31:28 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm11
; SSE-NEXT: pcmpgtd %xmm7, %xmm3
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: movdqa {{.*#+}} xmm7 = <u,u,u,u,0,4,8,12,u,u,u,u,u,u,u,u>
; SSE-NEXT: pshufb %xmm7, %xmm3
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtd %xmm6, %xmm2
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm7, %xmm2
; SSE-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtd %xmm5, %xmm1
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: movdqa {{.*#+}} xmm3 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
; SSE-NEXT: pshufb %xmm3, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtd %xmm4, %xmm0
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm3, %xmm0
; SSE-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtd {{[0-9]+}}(%rsp), %xmm11
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm7, %xmm11
[X86][SSE] Add support for lowering v8i16 binary shuffles to PACKSS/PACKUS Missed in D38472 llvm-svn: 314916
2017-10-05 01:31:28 +08:00
; SSE-NEXT: pcmpgtd {{[0-9]+}}(%rsp), %xmm9
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm7, %xmm9
; SSE-NEXT: punpckldq {{.*#+}} xmm9 = xmm9[0],xmm11[0],xmm9[1],xmm11[1]
; SSE-NEXT: pcmpgtd {{[0-9]+}}(%rsp), %xmm10
; SSE-NEXT: pshufb %xmm3, %xmm10
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtd {{[0-9]+}}(%rsp), %xmm8
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm3, %xmm8
; SSE-NEXT: punpckldq {{.*#+}} xmm8 = xmm8[0],xmm10[0],xmm8[1],xmm10[1]
; SSE-NEXT: pblendw {{.*#+}} xmm8 = xmm8[0,1,2,3],xmm9[4,5,6,7]
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: pand %xmm0, %xmm8
; SSE-NEXT: pmovmskb %xmm8, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; SSE-NEXT: # kill: def $ax killed $ax killed $eax
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; SSE-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX1-LABEL: v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX1: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm8
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm9
; AVX1-NEXT: vpcmpgtd %xmm8, %xmm9, %xmm8
; AVX1-NEXT: vpcmpgtd %xmm3, %xmm1, %xmm1
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX1-NEXT: vpackssdw %xmm8, %xmm1, %xmm8
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT: vpcmpgtd %xmm3, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtd %xmm2, %xmm0, %xmm0
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX1-NEXT: vpackssdw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpacksswb %xmm8, %xmm0, %xmm0
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm7, %xmm1
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm2
; AVX1-NEXT: vpcmpgtd %xmm1, %xmm2, %xmm1
; AVX1-NEXT: vpcmpgtd %xmm7, %xmm5, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX1-NEXT: vpackssdw %xmm1, %xmm2, %xmm1
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm3
; AVX1-NEXT: vpcmpgtd %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpcmpgtd %xmm6, %xmm4, %xmm3
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX1-NEXT: vpackssdw %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpacksswb %xmm1, %xmm2, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpmovmskb %xmm0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX1-NEXT: # kill: def $ax killed $ax killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX1-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX2-LABEL: v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX2: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpcmpgtd %ymm3, %ymm1, %ymm1
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm3
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX2-NEXT: vpackssdw %xmm3, %xmm1, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpcmpgtd %ymm2, %ymm0, %ymm0
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX2-NEXT: vpackssdw %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; AVX2-NEXT: vpcmpgtd %ymm7, %ymm5, %ymm1
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX2-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; AVX2-NEXT: vpcmpgtd %ymm6, %ymm4, %ymm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT: vpackssdw %xmm3, %xmm2, %xmm2
; AVX2-NEXT: vpacksswb %xmm1, %xmm2, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpmovmskb %xmm0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX2-NEXT: # kill: def $ax killed $ax killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX2-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512F-LABEL: v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512F: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vpcmpgtd %zmm1, %zmm0, %k1
; AVX512F-NEXT: vpcmpgtd %zmm3, %zmm2, %k0 {%k1}
; AVX512F-NEXT: kmovw %k0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512F-NEXT: # kill: def $ax killed $ax killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512F-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512BW-LABEL: v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vpcmpgtd %zmm1, %zmm0, %k1
; AVX512BW-NEXT: vpcmpgtd %zmm3, %zmm2, %k0 {%k1}
; AVX512BW-NEXT: kmovd %k0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512BW-NEXT: # kill: def $ax killed $ax killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512BW-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
%x0 = icmp sgt <16 x i32> %a, %b
%x1 = icmp sgt <16 x i32> %c, %d
%y = and <16 x i1> %x0, %x1
%res = bitcast <16 x i1> %y to i16
ret i16 %res
}
define i16 @v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x float> %d) {
; SSE-LABEL: v16f32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; SSE: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movaps {{[0-9]+}}(%rsp), %xmm8
[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types. This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type. Differential Revision: https://reviews.llvm.org/D39289 llvm-svn: 316831
2017-10-28 22:27:53 +08:00
; SSE-NEXT: movaps {{[0-9]+}}(%rsp), %xmm10
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: movaps {{[0-9]+}}(%rsp), %xmm9
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movaps {{[0-9]+}}(%rsp), %xmm11
; SSE-NEXT: cmpltps %xmm3, %xmm7
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: movdqa {{.*#+}} xmm3 = <u,u,u,u,0,4,8,12,u,u,u,u,u,u,u,u>
; SSE-NEXT: pshufb %xmm3, %xmm7
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: cmpltps %xmm2, %xmm6
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm3, %xmm6
; SSE-NEXT: punpckldq {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1]
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: cmpltps %xmm1, %xmm5
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: movdqa {{.*#+}} xmm1 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
; SSE-NEXT: pshufb %xmm1, %xmm5
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: cmpltps %xmm0, %xmm4
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm1, %xmm4
; SSE-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
; SSE-NEXT: pblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm6[4,5,6,7]
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: cmpltps {{[0-9]+}}(%rsp), %xmm11
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm3, %xmm11
[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types. This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type. Differential Revision: https://reviews.llvm.org/D39289 llvm-svn: 316831
2017-10-28 22:27:53 +08:00
; SSE-NEXT: cmpltps {{[0-9]+}}(%rsp), %xmm9
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm3, %xmm9
; SSE-NEXT: punpckldq {{.*#+}} xmm9 = xmm9[0],xmm11[0],xmm9[1],xmm11[1]
; SSE-NEXT: cmpltps {{[0-9]+}}(%rsp), %xmm10
; SSE-NEXT: pshufb %xmm1, %xmm10
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: cmpltps {{[0-9]+}}(%rsp), %xmm8
[SelectionDAG][ARM][X86] Teach PromoteIntRes_SETCC to do a better job picking the result type for the setcc. Previously if getSetccResultType returned an illegal type we just fell back to using the default promoted type. This appears to have been to handle the case where for vectors getSetccResultType returns the input type, but the input type itself isn't legal and will need to be promoted. Without the legality check we would never reach a legal type. But just picking the promoted type to be the setcc type can create strange setccs where the result type is 128 bits and the operand type is 256 bits. If for example the result type was promoted to v8i16 from v8i1, but the input type was promoted from v8i23 to v8i32. We currently handle this with custom lowering code in X86. This legality check also caused us reject the getSetccResultType when the input type needed to be widened or split. Even though that result wouldn't have caused legalization to get stuck. This patch tries to fix this by detecting the getSetccResultType needs to be promoted. If its input type also needs to be promoted we'll try a ask for a new setcc result type based on its eventual promoted value. Otherwise we fall back to default type to promote to. For any other illegal values we might get back from the initial call to getSetccResultType we just keep and allow it to be re-legalized later via splitting or widening or scalarizing. llvm-svn: 327683
2018-03-16 07:04:11 +08:00
; SSE-NEXT: pshufb %xmm1, %xmm8
; SSE-NEXT: punpckldq {{.*#+}} xmm8 = xmm8[0],xmm10[0],xmm8[1],xmm10[1]
; SSE-NEXT: pblendw {{.*#+}} xmm8 = xmm8[0,1,2,3],xmm9[4,5,6,7]
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; SSE-NEXT: pand %xmm4, %xmm8
; SSE-NEXT: pmovmskb %xmm8, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; SSE-NEXT: # kill: def $ax killed $ax killed $eax
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; SSE-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX12-LABEL: v16f32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX12: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vcmpltps %ymm1, %ymm3, %ymm1
; AVX12-NEXT: vextractf128 $1, %ymm1, %xmm3
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX12-NEXT: vpackssdw %xmm3, %xmm1, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vcmpltps %ymm0, %ymm2, %ymm0
; AVX12-NEXT: vextractf128 $1, %ymm0, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX12-NEXT: vpackssdw %xmm2, %xmm0, %xmm0
[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types. This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type. Differential Revision: https://reviews.llvm.org/D39289 llvm-svn: 316831
2017-10-28 22:27:53 +08:00
; AVX12-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; AVX12-NEXT: vcmpltps %ymm5, %ymm7, %ymm1
; AVX12-NEXT: vextractf128 $1, %ymm1, %xmm2
[X86] truncateVectorCompareWithPACKSS - use PACKSSDW/PACKSSWB instead of just PACKSSWB. By using the widest type possible for PACKSS truncation we have a better chance of being able to peek through bitcasts and improves other combines driven by ComputeNumSignBits. llvm-svn: 316448
2017-10-24 23:38:16 +08:00
; AVX12-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
[SelectionDAG] Add BITCAST handling to ComputeNumSignBits for splatted sign bits. For cases where we are BITCASTing to vectors of smaller elements, then if the entire source was a splatted sign (src's NumSignBits == SrcBitWidth) we can say that the dst's NumSignBit == DstBitWidth, as we're just splitting those sign bits across multiple elements. We could generalize this but at the moment the only use case I have is to peek through bitcasts to vector comparison results. Differential Revision: https://reviews.llvm.org/D37849 llvm-svn: 313543
2017-09-19 00:45:05 +08:00
; AVX12-NEXT: vcmpltps %ymm4, %ymm6, %ymm2
[SelectionDAG] Support 'bit preserving' floating points bitcasts on computeKnownBits/ComputeNumSignBits For cases where we know the floating point representations match the bitcasted integer equivalent, allow bitcasting to these types. This is especially useful for the X86 floating point compare results which return all/zero bits but as a floating point type. Differential Revision: https://reviews.llvm.org/D39289 llvm-svn: 316831
2017-10-28 22:27:53 +08:00
; AVX12-NEXT: vextractf128 $1, %ymm2, %xmm3
; AVX12-NEXT: vpackssdw %xmm3, %xmm2, %xmm2
; AVX12-NEXT: vpacksswb %xmm1, %xmm2, %xmm1
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX12-NEXT: vpmovmskb %xmm0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX12-NEXT: # kill: def $ax killed $ax killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX12-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX12-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512F-LABEL: v16f32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512F: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vcmpltps %zmm0, %zmm1, %k1
; AVX512F-NEXT: vcmpltps %zmm2, %zmm3, %k0 {%k1}
; AVX512F-NEXT: kmovw %k0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512F-NEXT: # kill: def $ax killed $ax killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512F-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512BW-LABEL: v16f32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vcmpltps %zmm0, %zmm1, %k1
; AVX512BW-NEXT: vcmpltps %zmm2, %zmm3, %k0 {%k1}
; AVX512BW-NEXT: kmovd %k0, %eax
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512BW-NEXT: # kill: def $ax killed $ax killed $eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512BW-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
%x0 = fcmp ogt <16 x float> %a, %b
%x1 = fcmp ogt <16 x float> %c, %d
%y = and <16 x i1> %x0, %x1
%res = bitcast <16 x i1> %y to i16
ret i16 %res
}
define i64 @v64i8(<64 x i8> %a, <64 x i8> %b, <64 x i8> %c, <64 x i8> %d) {
; SSE-LABEL: v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; SSE: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm11
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm8
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; SSE-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtb %xmm7, %xmm3
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; SSE-NEXT: pcmpgtb %xmm6, %xmm2
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtb %xmm5, %xmm1
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; SSE-NEXT: pcmpgtb %xmm4, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtb {{[0-9]+}}(%rsp), %xmm9
; SSE-NEXT: pand %xmm3, %xmm9
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; SSE-NEXT: pcmpgtb {{[0-9]+}}(%rsp), %xmm8
; SSE-NEXT: pand %xmm2, %xmm8
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: pcmpgtb {{[0-9]+}}(%rsp), %xmm11
; SSE-NEXT: pand %xmm1, %xmm11
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; SSE-NEXT: pcmpgtb {{[0-9]+}}(%rsp), %xmm10
; SSE-NEXT: pand %xmm0, %xmm10
; SSE-NEXT: pmovmskb %xmm10, %eax
; SSE-NEXT: pmovmskb %xmm11, %ecx
; SSE-NEXT: shll $16, %ecx
[SelectionDAG] Fix codegen of vector stores with non byte-sized elements. This was completely broken, but hopefully fixed by this patch. In cases where it is needed, a vector with non byte-sized elements is stored by extracting, zero-extending, shift:ing and or:ing the elements into an integer of the same width as the vector, which is then stored. Review: Eli Friedman, Ulrich Weigand https://reviews.llvm.org/D42100#inline-369520 https://bugs.llvm.org/show_bug.cgi?id=35520 llvm-svn: 323042
2018-01-21 00:05:10 +08:00
; SSE-NEXT: orl %eax, %ecx
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; SSE-NEXT: pmovmskb %xmm8, %edx
; SSE-NEXT: pmovmskb %xmm9, %eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; SSE-NEXT: shll $16, %eax
; SSE-NEXT: orl %edx, %eax
; SSE-NEXT: shlq $32, %rax
; SSE-NEXT: orq %rcx, %rax
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; SSE-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX1-LABEL: v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX1: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm3, %xmm8
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm9
; AVX1-NEXT: vpcmpgtb %xmm8, %xmm9, %xmm8
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; AVX1-NEXT: vpcmpgtb %xmm3, %xmm1, %xmm9
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm2, %xmm3
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT: vpcmpgtb %xmm3, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtb %xmm2, %xmm0, %xmm0
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; AVX1-NEXT: vextractf128 $1, %ymm7, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm5, %xmm3
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: vpcmpgtb %xmm2, %xmm3, %xmm2
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; AVX1-NEXT: vpand %xmm2, %xmm8, %xmm2
; AVX1-NEXT: vpcmpgtb %xmm7, %xmm5, %xmm3
; AVX1-NEXT: vpand %xmm3, %xmm9, %xmm3
; AVX1-NEXT: vextractf128 $1, %ymm6, %xmm5
; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm7
; AVX1-NEXT: vpcmpgtb %xmm5, %xmm7, %xmm5
; AVX1-NEXT: vpand %xmm5, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtb %xmm6, %xmm4, %xmm4
; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm0
; AVX1-NEXT: vpmovmskb %xmm0, %eax
; AVX1-NEXT: vpmovmskb %xmm1, %ecx
[SelectionDAG] Fix codegen of vector stores with non byte-sized elements. This was completely broken, but hopefully fixed by this patch. In cases where it is needed, a vector with non byte-sized elements is stored by extracting, zero-extending, shift:ing and or:ing the elements into an integer of the same width as the vector, which is then stored. Review: Eli Friedman, Ulrich Weigand https://reviews.llvm.org/D42100#inline-369520 https://bugs.llvm.org/show_bug.cgi?id=35520 llvm-svn: 323042
2018-01-21 00:05:10 +08:00
; AVX1-NEXT: shll $16, %ecx
; AVX1-NEXT: orl %eax, %ecx
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; AVX1-NEXT: vpmovmskb %xmm3, %edx
; AVX1-NEXT: vpmovmskb %xmm2, %eax
; AVX1-NEXT: shll $16, %eax
; AVX1-NEXT: orl %edx, %eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX1-NEXT: shlq $32, %rax
; AVX1-NEXT: orq %rcx, %rax
; AVX1-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX1-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX2-LABEL: v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX2: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: vpcmpgtb %ymm3, %ymm1, %ymm1
[X86] Add a custom legalization for (i16 (bitcast v16i1)) and (i32 (bitcast v32i1)) without AVX512 to prevent scalarization Summary: We have an early DAG combine to turn these patterns into MOVMSK, but that combine doesn't work if the vXi1 type has more elements than the widest legal vXi8 type. Type legalization will eventually split it down to v16i1 or v32i1 and then the bitcast gets legalized to a truncstore and a scalar load. The truncstore will get lowered to a series of extracts and bit math. This patch adds a custom legalization to use a sign extend and MOVMSK instead. This prevents the eventual scalarization. Reviewers: spatel, RKSimon, zvi Reviewed By: RKSimon Subscribers: mgorny, llvm-commits Differential Revision: https://reviews.llvm.org/D43593 llvm-svn: 326119
2018-02-27 04:32:27 +08:00
; AVX2-NEXT: vpcmpgtb %ymm2, %ymm0, %ymm0
; AVX2-NEXT: vpcmpgtb %ymm7, %ymm5, %ymm2
; AVX2-NEXT: vpand %ymm2, %ymm1, %ymm1
; AVX2-NEXT: vpcmpgtb %ymm6, %ymm4, %ymm2
; AVX2-NEXT: vpand %ymm2, %ymm0, %ymm0
; AVX2-NEXT: vpmovmskb %ymm0, %ecx
; AVX2-NEXT: vpmovmskb %ymm1, %eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX2-NEXT: shlq $32, %rax
; AVX2-NEXT: orq %rcx, %rax
; AVX2-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX2-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512F-LABEL: v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512F: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vpcmpgtb %ymm3, %ymm1, %ymm1
[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8. Summary: For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register. I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes. There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined. I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc)) Overall this removes something like 13000 CHECK lines from lit tests. Reviewers: zvi, RKSimon, delena, spatel Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42031 llvm-svn: 323201
2018-01-23 22:25:39 +08:00
; AVX512F-NEXT: vextracti128 $1, %ymm1, %xmm3
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vpcmpgtb %ymm2, %ymm0, %ymm0
[DAGCombiner] allow hoisting vector bitwise logic ahead of truncates The transform performs a bitwise logic op in a wider type followed by truncate when both inputs are truncated from the same source type: logic_op (truncate x), (truncate y) --> truncate (logic_op x, y) There are a bunch of other checks that should prevent doing this when it might be harmful. We already do this transform for scalars in this spot. The vector limitation was shared with a check for the case when the operands are extended. I'm not sure if that limit is needed either, but that would be a separate patch. Differential Revision: https://reviews.llvm.org/D55448 llvm-svn: 349303
2018-12-16 22:57:04 +08:00
; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX512F-NEXT: vpcmpgtb %ymm7, %ymm5, %ymm5
; AVX512F-NEXT: vextracti128 $1, %ymm5, %xmm7
; AVX512F-NEXT: vpand %xmm7, %xmm3, %xmm3
; AVX512F-NEXT: vpcmpgtb %ymm6, %ymm4, %ymm4
; AVX512F-NEXT: vextracti128 $1, %ymm4, %xmm6
; AVX512F-NEXT: vpand %xmm6, %xmm2, %xmm2
; AVX512F-NEXT: vpand %xmm4, %xmm0, %xmm0
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0
[DAGCombiner] allow hoisting vector bitwise logic ahead of truncates The transform performs a bitwise logic op in a wider type followed by truncate when both inputs are truncated from the same source type: logic_op (truncate x), (truncate y) --> truncate (logic_op x, y) There are a bunch of other checks that should prevent doing this when it might be harmful. We already do this transform for scalars in this spot. The vector limitation was shared with a check for the case when the operands are extended. I'm not sure if that limit is needed either, but that would be a separate patch. Differential Revision: https://reviews.llvm.org/D55448 llvm-svn: 349303
2018-12-16 22:57:04 +08:00
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8. Summary: For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register. I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes. There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined. I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc)) Overall this removes something like 13000 CHECK lines from lit tests. Reviewers: zvi, RKSimon, delena, spatel Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42031 llvm-svn: 323201
2018-01-23 22:25:39 +08:00
; AVX512F-NEXT: kmovw %k0, %eax
[DAGCombiner] allow hoisting vector bitwise logic ahead of truncates The transform performs a bitwise logic op in a wider type followed by truncate when both inputs are truncated from the same source type: logic_op (truncate x), (truncate y) --> truncate (logic_op x, y) There are a bunch of other checks that should prevent doing this when it might be harmful. We already do this transform for scalars in this spot. The vector limitation was shared with a check for the case when the operands are extended. I'm not sure if that limit is needed either, but that would be a separate patch. Differential Revision: https://reviews.llvm.org/D55448 llvm-svn: 349303
2018-12-16 22:57:04 +08:00
; AVX512F-NEXT: vpmovsxbd %xmm2, %zmm0
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8. Summary: For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register. I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes. There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined. I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc)) Overall this removes something like 13000 CHECK lines from lit tests. Reviewers: zvi, RKSimon, delena, spatel Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42031 llvm-svn: 323201
2018-01-23 22:25:39 +08:00
; AVX512F-NEXT: kmovw %k0, %ecx
; AVX512F-NEXT: shll $16, %ecx
; AVX512F-NEXT: orl %eax, %ecx
[DAGCombiner] allow hoisting vector bitwise logic ahead of truncates The transform performs a bitwise logic op in a wider type followed by truncate when both inputs are truncated from the same source type: logic_op (truncate x), (truncate y) --> truncate (logic_op x, y) There are a bunch of other checks that should prevent doing this when it might be harmful. We already do this transform for scalars in this spot. The vector limitation was shared with a check for the case when the operands are extended. I'm not sure if that limit is needed either, but that would be a separate patch. Differential Revision: https://reviews.llvm.org/D55448 llvm-svn: 349303
2018-12-16 22:57:04 +08:00
; AVX512F-NEXT: vpand %xmm5, %xmm1, %xmm0
; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8. Summary: For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register. I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes. There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined. I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc)) Overall this removes something like 13000 CHECK lines from lit tests. Reviewers: zvi, RKSimon, delena, spatel Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42031 llvm-svn: 323201
2018-01-23 22:25:39 +08:00
; AVX512F-NEXT: kmovw %k0, %edx
[DAGCombiner] allow hoisting vector bitwise logic ahead of truncates The transform performs a bitwise logic op in a wider type followed by truncate when both inputs are truncated from the same source type: logic_op (truncate x), (truncate y) --> truncate (logic_op x, y) There are a bunch of other checks that should prevent doing this when it might be harmful. We already do this transform for scalars in this spot. The vector limitation was shared with a check for the case when the operands are extended. I'm not sure if that limit is needed either, but that would be a separate patch. Differential Revision: https://reviews.llvm.org/D55448 llvm-svn: 349303
2018-12-16 22:57:04 +08:00
; AVX512F-NEXT: vpmovsxbd %xmm3, %zmm0
; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0
[X86] Legalize v32i1 without BWI via splitting to v16i1 rather than the default of promoting to v32i8. Summary: For the most part its better to keep v32i1 as a mask type of a narrower width than trying to promote it to a ymm register. I had to add some overrides to the methods that get the types for the calling convention so that we still use v32i8 for argument/return purposes. There are still some regressions in here. I definitely saw some around shuffles. I think we probably should move vXi1 shuffle from lowering to a DAG combine where I think the extend and truncate we have to emit would be better combined. I think we also need a DAG combine to remove trunc from (extract_vector_elt (trunc)) Overall this removes something like 13000 CHECK lines from lit tests. Reviewers: zvi, RKSimon, delena, spatel Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42031 llvm-svn: 323201
2018-01-23 22:25:39 +08:00
; AVX512F-NEXT: kmovw %k0, %eax
; AVX512F-NEXT: shll $16, %eax
; AVX512F-NEXT: orl %edx, %eax
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512F-NEXT: shlq $32, %rax
; AVX512F-NEXT: orq %rcx, %rax
; AVX512F-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512F-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
;
; AVX512BW-LABEL: v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
; AVX512BW-NEXT: vpcmpgtb %zmm1, %zmm0, %k1
; AVX512BW-NEXT: vpcmpgtb %zmm3, %zmm2, %k0 {%k1}
; AVX512BW-NEXT: kmovq %k0, %rax
; AVX512BW-NEXT: vzeroupper
[X86][AVX512] Regenerate tests to remove retl/retq regex These are only testing 64-bit targets so we don't need the regex llvm-svn: 317021
2017-11-01 02:43:24 +08:00
; AVX512BW-NEXT: retq
Fix line endings. NFCI. llvm-svn: 313246
2017-09-14 18:30:22 +08:00
%x0 = icmp sgt <64 x i8> %a, %b
%x1 = icmp sgt <64 x i8> %c, %d
%y = and <64 x i1> %x0, %x1
%res = bitcast <64 x i1> %y to i64
ret i64 %res
}