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/insert-into-constant-vector.ll

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

525 lines
20 KiB
LLVM
Raw Normal View History

[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
Recommit r354363 "[X86][SSE] Generalize X86ISD::BLENDI support to more value types" And its follow ups r354511, r354640. A follow patch will fix the issue that caused it to be reverted. llvm-svn: 354737
2019-02-24 05:41:42 +08:00
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X32SSE --check-prefix=X32SSE2
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64SSE --check-prefix=X64SSE2
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=X32SSE --check-prefix=X32SSE4
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=X64SSE --check-prefix=X64SSE4
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32AVX --check-prefix=X32AVX1
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64AVX --check-prefix=X64AVX1
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X32AVX --check-prefix=X32AVX2
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X64AVX --check-prefix=X64AVX2
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=X32AVX --check-prefix=X32AVX512F
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=X64AVX --check-prefix=X64AVX512F
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
define <16 x i8> @elt0_v16i8(i8 %x) {
; X32SSE2-LABEL: elt0_v16i8:
[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
; X32SSE2: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE2-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; X32SSE2-NEXT: movaps {{.*#+}} xmm0 = [0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
; X32SSE2-NEXT: andnps %xmm1, %xmm0
; X32SSE2-NEXT: orps {{\.LCPI.*}}, %xmm0
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE2-NEXT: retl
;
; X64SSE2-LABEL: elt0_v16i8:
[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
; X64SSE2: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-NEXT: movd %edi, %xmm1
; X64SSE2-NEXT: movdqa {{.*#+}} xmm0 = [0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
; X64SSE2-NEXT: pandn %xmm1, %xmm0
; X64SSE2-NEXT: por {{.*}}(%rip), %xmm0
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE2-NEXT: retq
;
; X32SSE4-LABEL: elt0_v16i8:
[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
; X32SSE4: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE4-NEXT: movdqa {{.*#+}} xmm0 = <u,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15>
; X32SSE4-NEXT: pinsrb $0, {{[0-9]+}}(%esp), %xmm0
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE4-NEXT: retl
;
; X64SSE4-LABEL: elt0_v16i8:
[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
; X64SSE4: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE4-NEXT: movdqa {{.*#+}} xmm0 = <u,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15>
; X64SSE4-NEXT: pinsrb $0, %edi, %xmm0
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE4-NEXT: retq
;
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-LABEL: elt0_v16i8:
[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
; X32AVX: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32AVX-NEXT: vmovdqa {{.*#+}} xmm0 = <u,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15>
; X32AVX-NEXT: vpinsrb $0, {{[0-9]+}}(%esp), %xmm0, %xmm0
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-NEXT: retl
;
; X64AVX-LABEL: elt0_v16i8:
[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
; X64AVX: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64AVX-NEXT: vmovdqa {{.*#+}} xmm0 = <u,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15>
; X64AVX-NEXT: vpinsrb $0, %edi, %xmm0, %xmm0
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64AVX-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
%ins = insertelement <16 x i8> <i8 42, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, i8 %x, i32 0
ret <16 x i8> %ins
}
define <8 x i16> @elt5_v8i16(i16 %x) {
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE-LABEL: elt5_v8i16:
[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
; X32SSE: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE-NEXT: movdqa {{.*#+}} xmm0 = <42,1,2,3,4,u,6,7>
; X32SSE-NEXT: pinsrw $5, {{[0-9]+}}(%esp), %xmm0
; X32SSE-NEXT: retl
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
;
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE-LABEL: elt5_v8i16:
[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
; X64SSE: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE-NEXT: movdqa {{.*#+}} xmm0 = <42,1,2,3,4,u,6,7>
; X64SSE-NEXT: pinsrw $5, %edi, %xmm0
; X64SSE-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
;
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-LABEL: elt5_v8i16:
[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
; X32AVX: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32AVX-NEXT: vmovdqa {{.*#+}} xmm0 = <42,1,2,3,4,u,6,7>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-NEXT: vpinsrw $5, {{[0-9]+}}(%esp), %xmm0, %xmm0
; X32AVX-NEXT: retl
;
; X64AVX-LABEL: elt5_v8i16:
[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
; X64AVX: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64AVX-NEXT: vmovdqa {{.*#+}} xmm0 = <42,1,2,3,4,u,6,7>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64AVX-NEXT: vpinsrw $5, %edi, %xmm0, %xmm0
; X64AVX-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
%ins = insertelement <8 x i16> <i16 42, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, i16 %x, i32 5
ret <8 x i16> %ins
}
define <4 x i32> @elt3_v4i32(i32 %x) {
; X32SSE2-LABEL: elt3_v4i32:
[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
; X32SSE2: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE2-NEXT: movaps {{.*#+}} xmm0 = <42,1,2,u>
; X32SSE2-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; X32SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
; X32SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE2-NEXT: retl
;
; X64SSE2-LABEL: elt3_v4i32:
[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
; X64SSE2: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-NEXT: movd %edi, %xmm1
; X64SSE2-NEXT: movaps {{.*#+}} xmm0 = <42,1,2,u>
; X64SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[2,0]
; X64SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE2-NEXT: retq
;
; X32SSE4-LABEL: elt3_v4i32:
[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
; X32SSE4: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE4-NEXT: movdqa {{.*#+}} xmm0 = <42,1,2,u>
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE4-NEXT: pinsrd $3, {{[0-9]+}}(%esp), %xmm0
; X32SSE4-NEXT: retl
;
; X64SSE4-LABEL: elt3_v4i32:
[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
; X64SSE4: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE4-NEXT: movdqa {{.*#+}} xmm0 = <42,1,2,u>
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE4-NEXT: pinsrd $3, %edi, %xmm0
; X64SSE4-NEXT: retq
;
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-LABEL: elt3_v4i32:
[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
; X32AVX: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32AVX-NEXT: vmovdqa {{.*#+}} xmm0 = <42,1,2,u>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-NEXT: vpinsrd $3, {{[0-9]+}}(%esp), %xmm0, %xmm0
; X32AVX-NEXT: retl
;
; X64AVX-LABEL: elt3_v4i32:
[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
; X64AVX: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64AVX-NEXT: vmovdqa {{.*#+}} xmm0 = <42,1,2,u>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64AVX-NEXT: vpinsrd $3, %edi, %xmm0, %xmm0
; X64AVX-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
%ins = insertelement <4 x i32> <i32 42, i32 1, i32 2, i32 3>, i32 %x, i32 3
ret <4 x i32> %ins
}
define <2 x i64> @elt0_v2i64(i64 %x) {
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32SSE-LABEL: elt0_v2i64:
[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
; X32SSE: # %bb.0:
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32SSE-NEXT: movl $1, %eax
; X32SSE-NEXT: movd %eax, %xmm1
; X32SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
; X32SSE-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; X32SSE-NEXT: retl
;
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-LABEL: elt0_v2i64:
[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
; X64SSE2: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-NEXT: movq %rdi, %xmm1
; X64SSE2-NEXT: movapd {{.*#+}} xmm0 = <u,1>
; X64SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
; X64SSE2-NEXT: retq
;
; X64SSE4-LABEL: elt0_v2i64:
[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
; X64SSE4: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE4-NEXT: movdqa {{.*#+}} xmm0 = <u,1>
; X64SSE4-NEXT: pinsrq $0, %rdi, %xmm0
; X64SSE4-NEXT: retq
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
;
; X32AVX-LABEL: elt0_v2i64:
[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
; X32AVX: # %bb.0:
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-NEXT: movl $1, %eax
; X32AVX-NEXT: vmovd %eax, %xmm0
; X32AVX-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
; X32AVX-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; X32AVX-NEXT: retl
;
; X64AVX-LABEL: elt0_v2i64:
[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
; X64AVX: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64AVX-NEXT: vmovdqa {{.*#+}} xmm0 = <u,1>
; X64AVX-NEXT: vpinsrq $0, %rdi, %xmm0, %xmm0
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64AVX-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
%ins = insertelement <2 x i64> <i64 42, i64 1>, i64 %x, i32 0
ret <2 x i64> %ins
}
define <4 x float> @elt1_v4f32(float %x) {
; X32SSE2-LABEL: elt1_v4f32:
[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
; X32SSE2: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32SSE2-NEXT: movaps {{.*#+}} xmm1 = <4.2E+1,u,2.0E+0,3.0E+0>
[X86] Add patterns to select (scalar_to_vector (loadf32)) as (V)MOVSSrm instead of COPY_TO_REGCLASS + (V)MOVSSrm_alt. Similar for (V)MOVSD. Ultimately, I'd like to see about folding scalar_to_vector+load to vzload. Which would select as (V)MOVSSrm so this is closer to that. llvm-svn: 364948
2019-07-03 01:51:02 +08:00
; X32SSE2-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
; X32SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE2-NEXT: retl
;
; X64SSE2-LABEL: elt1_v4f32:
[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
; X64SSE2: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE2-NEXT: movaps {{.*#+}} xmm1 = <4.2E+1,u,2.0E+0,3.0E+0>
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[0,0]
; X64SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[2,3]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE2-NEXT: retq
;
; X32SSE4-LABEL: elt1_v4f32:
[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
; X32SSE4: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32SSE4-NEXT: movaps {{.*#+}} xmm0 = <4.2E+1,u,2.0E+0,3.0E+0>
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE4-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
; X32SSE4-NEXT: retl
;
; X64SSE4-LABEL: elt1_v4f32:
[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
; X64SSE4: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE4-NEXT: movaps {{.*#+}} xmm1 = <4.2E+1,u,2.0E+0,3.0E+0>
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE4-NEXT: insertps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[2,3]
; X64SSE4-NEXT: movaps %xmm1, %xmm0
; X64SSE4-NEXT: retq
;
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-LABEL: elt1_v4f32:
[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
; X32AVX: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32AVX-NEXT: vmovaps {{.*#+}} xmm0 = <4.2E+1,u,2.0E+0,3.0E+0>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
; X32AVX-NEXT: retl
;
; X64AVX-LABEL: elt1_v4f32:
[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
; X64AVX: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64AVX-NEXT: vmovaps {{.*#+}} xmm1 = <4.2E+1,u,2.0E+0,3.0E+0>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64AVX-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[2,3]
; X64AVX-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
%ins = insertelement <4 x float> <float 42.0, float 1.0, float 2.0, float 3.0>, float %x, i32 1
ret <4 x float> %ins
}
define <2 x double> @elt1_v2f64(double %x) {
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32SSE-LABEL: elt1_v2f64:
[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
; X32SSE: # %bb.0:
[X86] Remove patterns from MOVLPSmr and MOVHPSmr instructions. These patterns are the same as the MOVLPDmr and MOVHPDmr patterns, but with a bitcast at the end. We can just select the PD instruction and let execution domain fixing switch to PS. llvm-svn: 365267
2019-07-07 01:59:51 +08:00
; X32SSE-NEXT: movaps {{.*#+}} xmm0 = <4.2E+1,u>
; X32SSE-NEXT: movhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32SSE-NEXT: retl
;
; X64SSE-LABEL: elt1_v2f64:
[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
; X64SSE: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE-NEXT: movaps {{.*#+}} xmm1 = <4.2E+1,u>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64SSE-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
; X64SSE-NEXT: movaps %xmm1, %xmm0
; X64SSE-NEXT: retq
;
; X32AVX-LABEL: elt1_v2f64:
[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
; X32AVX: # %bb.0:
[X86] Remove patterns from MOVLPSmr and MOVHPSmr instructions. These patterns are the same as the MOVLPDmr and MOVHPDmr patterns, but with a bitcast at the end. We can just select the PD instruction and let execution domain fixing switch to PS. llvm-svn: 365267
2019-07-07 01:59:51 +08:00
; X32AVX-NEXT: vmovaps {{.*#+}} xmm0 = <4.2E+1,u>
; X32AVX-NEXT: vmovhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32AVX-NEXT: retl
;
; X64AVX-LABEL: elt1_v2f64:
[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
; X64AVX: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64AVX-NEXT: vmovaps {{.*#+}} xmm1 = <4.2E+1,u>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64AVX-NEXT: vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; X64AVX-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
%ins = insertelement <2 x double> <double 42.0, double 1.0>, double %x, i32 1
ret <2 x double> %ins
}
define <8 x i32> @elt7_v8i32(i32 %x) {
; X32SSE2-LABEL: elt7_v8i32:
[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
; X32SSE2: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE2-NEXT: movaps {{.*#+}} xmm1 = <4,5,6,u>
; X32SSE2-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X32SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[2,0]
; X32SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE2-NEXT: movaps {{.*#+}} xmm0 = [42,1,2,3]
; X32SSE2-NEXT: retl
;
; X64SSE2-LABEL: elt7_v8i32:
[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
; X64SSE2: # %bb.0:
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE2-NEXT: movd %edi, %xmm0
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-NEXT: movaps {{.*#+}} xmm1 = <4,5,6,u>
; X64SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[2,0]
; X64SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE2-NEXT: movaps {{.*#+}} xmm0 = [42,1,2,3]
; X64SSE2-NEXT: retq
;
; X32SSE4-LABEL: elt7_v8i32:
[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
; X32SSE4: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE4-NEXT: movdqa {{.*#+}} xmm1 = <4,5,6,u>
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE4-NEXT: pinsrd $3, {{[0-9]+}}(%esp), %xmm1
; X32SSE4-NEXT: movaps {{.*#+}} xmm0 = [42,1,2,3]
; X32SSE4-NEXT: retl
;
; X64SSE4-LABEL: elt7_v8i32:
[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
; X64SSE4: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE4-NEXT: movdqa {{.*#+}} xmm1 = <4,5,6,u>
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE4-NEXT: pinsrd $3, %edi, %xmm1
; X64SSE4-NEXT: movaps {{.*#+}} xmm0 = [42,1,2,3]
; X64SSE4-NEXT: retq
;
[X86][AVX] insert_subvector(bitcast(v), bitcast(s), c1) -> bitcast(insert_subvector(v,s,c2)) Similar to what we already do in DAGCombiner, but this version also handles bitcasts from types with different scalar sizes, which x86 is better at handling. Differential Revision: https://reviews.llvm.org/D57514 llvm-svn: 352773
2019-02-01 01:38:10 +08:00
; X32AVX-LABEL: elt7_v8i32:
; X32AVX: # %bb.0:
; X32AVX-NEXT: vbroadcastss {{[0-9]+}}(%esp), %ymm0
; X32AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4,5,6],ymm0[7]
; X32AVX-NEXT: retl
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
;
; X64AVX1-LABEL: elt7_v8i32:
; X64AVX1: # %bb.0:
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX1-NEXT: vmovd %edi, %xmm0
; X64AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,2,0]
; X64AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X64AVX1-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4,5,6],ymm0[7]
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X64AVX1-NEXT: retq
;
; X64AVX2-LABEL: elt7_v8i32:
; X64AVX2: # %bb.0:
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX2-NEXT: vmovd %edi, %xmm0
[X86][AVX] Fold vt1 concat_vectors(vt2 undef, vt2 broadcast(x)) --> vt1 broadcast(x) If we're not inserting the broadcast into the lowest subvector then we can avoid the insertion by just performing a larger broadcast. Avoids a regression when we enable AVX1 broadcasts in shuffle combining llvm-svn: 352742
2019-01-31 19:15:05 +08:00
; X64AVX2-NEXT: vpbroadcastd %xmm0, %ymm0
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX2-NEXT: vpblendd {{.*#+}} ymm0 = mem[0,1,2,3,4,5,6],ymm0[7]
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X64AVX2-NEXT: retq
;
; X64AVX512F-LABEL: elt7_v8i32:
; X64AVX512F: # %bb.0:
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX512F-NEXT: vmovd %edi, %xmm0
[X86][AVX] Fold vt1 concat_vectors(vt2 undef, vt2 broadcast(x)) --> vt1 broadcast(x) If we're not inserting the broadcast into the lowest subvector then we can avoid the insertion by just performing a larger broadcast. Avoids a regression when we enable AVX1 broadcasts in shuffle combining llvm-svn: 352742
2019-01-31 19:15:05 +08:00
; X64AVX512F-NEXT: vpbroadcastd %xmm0, %ymm0
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX512F-NEXT: vpblendd {{.*#+}} ymm0 = mem[0,1,2,3,4,5,6],ymm0[7]
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X64AVX512F-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
%ins = insertelement <8 x i32> <i32 42, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, i32 %x, i32 7
ret <8 x i32> %ins
}
define <8 x float> @elt6_v8f32(float %x) {
; X32SSE2-LABEL: elt6_v8f32:
[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
; X32SSE2: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32SSE2-NEXT: movaps {{.*#+}} xmm1 = <4.0E+0,5.0E+0,u,7.0E+0>
[X86] Add patterns to select (scalar_to_vector (loadf32)) as (V)MOVSSrm instead of COPY_TO_REGCLASS + (V)MOVSSrm_alt. Similar for (V)MOVSD. Ultimately, I'd like to see about folding scalar_to_vector+load to vzload. Which would select as (V)MOVSSrm so this is closer to that. llvm-svn: 364948
2019-07-03 01:51:02 +08:00
; X32SSE2-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X32SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
; X32SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32SSE2-NEXT: movaps {{.*#+}} xmm0 = [4.2E+1,1.0E+0,2.0E+0,3.0E+0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE2-NEXT: retl
;
; X64SSE2-LABEL: elt6_v8f32:
[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
; X64SSE2: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE2-NEXT: movaps {{.*#+}} xmm1 = <4.0E+0,5.0E+0,u,7.0E+0>
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[3,0]
; X64SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0,2]
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE2-NEXT: movaps {{.*#+}} xmm0 = [4.2E+1,1.0E+0,2.0E+0,3.0E+0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE2-NEXT: retq
;
; X32SSE4-LABEL: elt6_v8f32:
[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
; X32SSE4: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32SSE4-NEXT: movaps {{.*#+}} xmm1 = <4.0E+0,5.0E+0,u,7.0E+0>
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE4-NEXT: insertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3]
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32SSE4-NEXT: movaps {{.*#+}} xmm0 = [4.2E+1,1.0E+0,2.0E+0,3.0E+0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32SSE4-NEXT: retl
;
; X64SSE4-LABEL: elt6_v8f32:
[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
; X64SSE4: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE4-NEXT: movaps {{.*#+}} xmm1 = <4.0E+0,5.0E+0,u,7.0E+0>
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE4-NEXT: insertps {{.*#+}} xmm1 = xmm1[0,1],xmm0[0],xmm1[3]
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE4-NEXT: movaps {{.*#+}} xmm0 = [4.2E+1,1.0E+0,2.0E+0,3.0E+0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64SSE4-NEXT: retq
;
[x86] try to form broadcast before widening shuffle elements I noticed that we weren't generating broadcasts as much I thought we would with D54271, and this is part of the problem. Widening the shuffle elements means adding bitcasts and hiding the relationship between a splatted scalar and the vector. If we can form a broadcast, do that before going through the rest of the shuffle lowering because broadcasts should be cheap and can often be load-folded. Differential Revision: https://reviews.llvm.org/D54280 llvm-svn: 346498
2018-11-09 22:54:58 +08:00
; X32AVX-LABEL: elt6_v8f32:
; X32AVX: # %bb.0:
; X32AVX-NEXT: vbroadcastss {{[0-9]+}}(%esp), %ymm0
; X32AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4,5],ymm0[6],mem[7]
; X32AVX-NEXT: retl
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
;
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX1-LABEL: elt6_v8f32:
; X64AVX1: # %bb.0:
; X64AVX1-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; X64AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X64AVX1-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4,5],ymm0[6],mem[7]
; X64AVX1-NEXT: retq
;
; X64AVX2-LABEL: elt6_v8f32:
; X64AVX2: # %bb.0:
[x86] try to form broadcast before widening shuffle elements I noticed that we weren't generating broadcasts as much I thought we would with D54271, and this is part of the problem. Widening the shuffle elements means adding bitcasts and hiding the relationship between a splatted scalar and the vector. If we can form a broadcast, do that before going through the rest of the shuffle lowering because broadcasts should be cheap and can often be load-folded. Differential Revision: https://reviews.llvm.org/D54280 llvm-svn: 346498
2018-11-09 22:54:58 +08:00
; X64AVX2-NEXT: vbroadcastss %xmm0, %ymm0
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX2-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4,5],ymm0[6],mem[7]
; X64AVX2-NEXT: retq
;
; X64AVX512F-LABEL: elt6_v8f32:
; X64AVX512F: # %bb.0:
[x86] try to form broadcast before widening shuffle elements I noticed that we weren't generating broadcasts as much I thought we would with D54271, and this is part of the problem. Widening the shuffle elements means adding bitcasts and hiding the relationship between a splatted scalar and the vector. If we can form a broadcast, do that before going through the rest of the shuffle lowering because broadcasts should be cheap and can often be load-folded. Differential Revision: https://reviews.llvm.org/D54280 llvm-svn: 346498
2018-11-09 22:54:58 +08:00
; X64AVX512F-NEXT: vbroadcastss %xmm0, %ymm0
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX512F-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4,5],ymm0[6],mem[7]
; X64AVX512F-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
%ins = insertelement <8 x float> <float 42.0, float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0>, float %x, i32 6
ret <8 x float> %ins
}
define <8 x i64> @elt5_v8i64(i64 %x) {
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32SSE-LABEL: elt5_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
; X32SSE: # %bb.0:
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32SSE-NEXT: movl $4, %eax
; X32SSE-NEXT: movd %eax, %xmm2
; X32SSE-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
; X32SSE-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
; X32SSE-NEXT: movaps {{.*#+}} xmm0 = [42,0,1,0]
; X32SSE-NEXT: movaps {{.*#+}} xmm1 = [2,0,3,0]
; X32SSE-NEXT: movaps {{.*#+}} xmm3 = [6,0,7,0]
; X32SSE-NEXT: retl
;
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-LABEL: elt5_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
; X64SSE2: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE2-NEXT: movq %rdi, %xmm0
; X64SSE2-NEXT: movdqa {{.*#+}} xmm2 = <4,u>
; X64SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
; X64SSE2-NEXT: movaps {{.*#+}} xmm0 = [42,1]
; X64SSE2-NEXT: movaps {{.*#+}} xmm1 = [2,3]
; X64SSE2-NEXT: movaps {{.*#+}} xmm3 = [6,7]
; X64SSE2-NEXT: retq
;
; X64SSE4-LABEL: elt5_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
; X64SSE4: # %bb.0:
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64SSE4-NEXT: movdqa {{.*#+}} xmm2 = <4,u>
; X64SSE4-NEXT: pinsrq $1, %rdi, %xmm2
; X64SSE4-NEXT: movaps {{.*#+}} xmm0 = [42,1]
; X64SSE4-NEXT: movaps {{.*#+}} xmm1 = [2,3]
; X64SSE4-NEXT: movaps {{.*#+}} xmm3 = [6,7]
; X64SSE4-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
;
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X32AVX1-LABEL: elt5_v8i64:
; X32AVX1: # %bb.0:
; X32AVX1-NEXT: movl $4, %eax
; X32AVX1-NEXT: vmovd %eax, %xmm0
; X32AVX1-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
; X32AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; X32AVX1-NEXT: vinsertf128 $1, {{\.LCPI.*}}, %ymm0, %ymm1
; X32AVX1-NEXT: vmovaps {{.*#+}} ymm0 = [42,0,1,0,2,0,3,0]
; X32AVX1-NEXT: retl
;
; X64AVX1-LABEL: elt5_v8i64:
; X64AVX1: # %bb.0:
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X64AVX1-NEXT: vmovdqa {{.*#+}} xmm0 = <4,u,6,7>
; X64AVX1-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0
; X64AVX1-NEXT: vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],mem[4,5,6,7]
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X64AVX1-NEXT: vmovaps {{.*#+}} ymm0 = [42,1,2,3]
; X64AVX1-NEXT: retq
;
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32AVX2-LABEL: elt5_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
; X32AVX2: # %bb.0:
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32AVX2-NEXT: movl $4, %eax
; X32AVX2-NEXT: vmovd %eax, %xmm0
; X32AVX2-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero
; X32AVX2-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; X32AVX2-NEXT: vinserti128 $1, {{\.LCPI.*}}, %ymm0, %ymm1
; X32AVX2-NEXT: vmovaps {{.*#+}} ymm0 = [42,0,1,0,2,0,3,0]
; X32AVX2-NEXT: retl
;
; X64AVX2-LABEL: elt5_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
; X64AVX2: # %bb.0:
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X64AVX2-NEXT: vmovdqa {{.*#+}} xmm0 = <4,u,6,7>
; X64AVX2-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0
; X64AVX2-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],mem[4,5,6,7]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64AVX2-NEXT: vmovaps {{.*#+}} ymm0 = [42,1,2,3]
; X64AVX2-NEXT: retq
;
; X32AVX512F-LABEL: elt5_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
; X32AVX512F: # %bb.0:
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32AVX512F-NEXT: vmovdqa {{.*#+}} ymm0 = [42,0,1,0,2,0,3,0]
; X32AVX512F-NEXT: movl $4, %eax
; X32AVX512F-NEXT: vmovd %eax, %xmm1
; X32AVX512F-NEXT: vmovq {{.*#+}} xmm2 = mem[0],zero
; X32AVX512F-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
; X32AVX512F-NEXT: vinserti128 $1, {{\.LCPI.*}}, %ymm1, %ymm1
; X32AVX512F-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; X32AVX512F-NEXT: retl
;
; X64AVX512F-LABEL: elt5_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
; X64AVX512F: # %bb.0:
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX512F-NEXT: vmovq %rdi, %xmm1
; X64AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,1,2,3,4,8,6,7]
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm0 = <42,1,2,3,4,u,6,7>
[x86] use shuffles for scalar insertion into high elements of a constant vector As discussed in D54073, we have a potential regression from more aggressive vector narrowing here, so let's try to avoid that by changing build-vector lowering slightly. Insert-vector-element lowering always does this since there's no "pinsr" for ymm/zmm: // If the vector is wider than 128 bits, extract the 128-bit subvector, insert // into that, and then insert the subvector back into the result. ...but we can sometimes do better for insert-into-constant-vector by using shuffle lowering. Differential Revision: https://reviews.llvm.org/D54271 llvm-svn: 346433
2018-11-09 03:16:27 +08:00
; X64AVX512F-NEXT: vpermt2q %zmm1, %zmm2, %zmm0
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64AVX512F-NEXT: retq
%ins = insertelement <8 x i64> <i64 42, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7>, i64 %x, i32 5
ret <8 x i64> %ins
}
define <8 x double> @elt1_v8f64(double %x) {
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32SSE-LABEL: elt1_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
; X32SSE: # %bb.0:
[X86] Remove patterns from MOVLPSmr and MOVHPSmr instructions. These patterns are the same as the MOVLPDmr and MOVHPDmr patterns, but with a bitcast at the end. We can just select the PD instruction and let execution domain fixing switch to PS. llvm-svn: 365267
2019-07-07 01:59:51 +08:00
; X32SSE-NEXT: movaps {{.*#+}} xmm0 = <4.2E+1,u>
; X32SSE-NEXT: movhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32SSE-NEXT: movaps {{.*#+}} xmm1 = [2.0E+0,3.0E+0]
; X32SSE-NEXT: movaps {{.*#+}} xmm2 = [4.0E+0,5.0E+0]
; X32SSE-NEXT: movaps {{.*#+}} xmm3 = [6.0E+0,7.0E+0]
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X32SSE-NEXT: retl
;
; X64SSE-LABEL: elt1_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
; X64SSE: # %bb.0:
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE-NEXT: movaps {{.*#+}} xmm4 = <4.2E+1,u>
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64SSE-NEXT: movlhps {{.*#+}} xmm4 = xmm4[0],xmm0[0]
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64SSE-NEXT: movaps {{.*#+}} xmm1 = [2.0E+0,3.0E+0]
; X64SSE-NEXT: movaps {{.*#+}} xmm2 = [4.0E+0,5.0E+0]
; X64SSE-NEXT: movaps {{.*#+}} xmm3 = [6.0E+0,7.0E+0]
[x86] fix prefix typos for CHECK lines; NFC llvm-svn: 315368
2017-10-11 05:12:47 +08:00
; X64SSE-NEXT: movaps %xmm4, %xmm0
; X64SSE-NEXT: retq
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
;
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X32AVX1-LABEL: elt1_v8f64:
; X32AVX1: # %bb.0:
[X86] Remove patterns from MOVLPSmr and MOVHPSmr instructions. These patterns are the same as the MOVLPDmr and MOVHPDmr patterns, but with a bitcast at the end. We can just select the PD instruction and let execution domain fixing switch to PS. llvm-svn: 365267
2019-07-07 01:59:51 +08:00
; X32AVX1-NEXT: vmovaps {{.*#+}} xmm0 = <4.2E+1,u,2.0E+0,3.0E+0>
; X32AVX1-NEXT: vmovhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
; X32AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],mem[4,5,6,7]
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X32AVX1-NEXT: vmovaps {{.*#+}} ymm1 = [4.0E+0,5.0E+0,6.0E+0,7.0E+0]
; X32AVX1-NEXT: retl
;
; X64AVX1-LABEL: elt1_v8f64:
; X64AVX1: # %bb.0:
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X64AVX1-NEXT: vmovaps {{.*#+}} xmm1 = <4.2E+1,u,2.0E+0,3.0E+0>
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X64AVX1-NEXT: vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X64AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],mem[4,5,6,7]
[x86] add RUNs for AVX1; NFC Differences in splat-ability might be reason to differentiate some cases. llvm-svn: 346426
2018-11-09 02:18:20 +08:00
; X64AVX1-NEXT: vmovaps {{.*#+}} ymm1 = [4.0E+0,5.0E+0,6.0E+0,7.0E+0]
; X64AVX1-NEXT: retq
;
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32AVX2-LABEL: elt1_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
; X32AVX2: # %bb.0:
[X86] Remove patterns from MOVLPSmr and MOVHPSmr instructions. These patterns are the same as the MOVLPDmr and MOVHPDmr patterns, but with a bitcast at the end. We can just select the PD instruction and let execution domain fixing switch to PS. llvm-svn: 365267
2019-07-07 01:59:51 +08:00
; X32AVX2-NEXT: vmovaps {{.*#+}} xmm0 = <4.2E+1,u,2.0E+0,3.0E+0>
; X32AVX2-NEXT: vmovhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
; X32AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],mem[4,5,6,7]
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X32AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [4.0E+0,5.0E+0,6.0E+0,7.0E+0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32AVX2-NEXT: retl
;
; X64AVX2-LABEL: elt1_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
; X64AVX2: # %bb.0:
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X64AVX2-NEXT: vmovaps {{.*#+}} xmm1 = <4.2E+1,u,2.0E+0,3.0E+0>
[X86][SSE] Add custom execution domain fixing for BLENDPD/BLENDPS/PBLENDD/PBLENDW (PR34873) Add support for custom execution domain fixing and implement support for BLENDPD/BLENDPS/PBLENDD/PBLENDW. Differential Revision: https://reviews.llvm.org/D42042 llvm-svn: 322524
2018-01-16 06:18:45 +08:00
; X64AVX2-NEXT: vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X64AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],mem[4,5,6,7]
[X86] Force floating point values in constant pool decoding to print in scientific notation so they can't be confused with integers. When the floating point constants are whole numbers they have no decimal point so look like integers, but mean something very different in something like an 'and' instruction. Ideally we would just print a decimal point and a 0, but I couldn't see how to make APFloat::toString do that. llvm-svn: 345488
2018-10-29 12:52:04 +08:00
; X64AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [4.0E+0,5.0E+0,6.0E+0,7.0E+0]
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64AVX2-NEXT: retq
;
; X32AVX512F-LABEL: elt1_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
; X32AVX512F: # %bb.0:
[X86] Remove patterns from MOVLPSmr and MOVHPSmr instructions. These patterns are the same as the MOVLPDmr and MOVHPDmr patterns, but with a bitcast at the end. We can just select the PD instruction and let execution domain fixing switch to PS. llvm-svn: 365267
2019-07-07 01:59:51 +08:00
; X32AVX512F-NEXT: vmovaps {{.*#+}} xmm0 = <4.2E+1,u,2.0E+0,3.0E+0,4.0E+0,5.0E+0,6.0E+0,7.0E+0>
; X32AVX512F-NEXT: vmovhps {{.*#+}} xmm0 = xmm0[0,1],mem[0,1]
; X32AVX512F-NEXT: vmovaps {{.*#+}} zmm1 = <4.2E+1,u,2.0E+0,3.0E+0,4.0E+0,5.0E+0,6.0E+0,7.0E+0>
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X32AVX512F-NEXT: vinsertf32x4 $0, %xmm0, %zmm1, %zmm0
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X32AVX512F-NEXT: retl
;
; X64AVX512F-LABEL: elt1_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
; X64AVX512F: # %bb.0:
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X64AVX512F-NEXT: vmovaps {{.*#+}} xmm1 = <4.2E+1,u,2.0E+0,3.0E+0,4.0E+0,5.0E+0,6.0E+0,7.0E+0>
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64AVX512F-NEXT: vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]
[x86] allow vector load narrowing with multi-use values This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs. Apart from 2-3 strange cases, these are all wins. I've structured this to be no-functional-change-intended for any target except for x86 because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those targets have existing regression tests (4, 4, 10 files respectively) that would be affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show any regression test diffs. The trade-off is deciding if an extra vector load is better than a single wide load + extract_subvector. For x86, this is almost always better (on paper at least) because we often can fold loads into subsequent ops and not increase the official instruction count. There's also some unknown -- but potentially large -- benefit from using narrower vector ops if wide ops are implemented with multiple uops and/or frequency throttling is avoided. Differential Revision: https://reviews.llvm.org/D54073 llvm-svn: 346595
2018-11-11 04:05:31 +08:00
; X64AVX512F-NEXT: vmovaps {{.*#+}} zmm1 = <4.2E+1,u,2.0E+0,3.0E+0,4.0E+0,5.0E+0,6.0E+0,7.0E+0>
[x86] use an insert op to put one variable element into a constant of vectors Instead of loading (a potential ton of) scalar constants, load those as a vector and then insert into it. Differential Revision: https://reviews.llvm.org/D38756 llvm-svn: 316685
2017-10-27 02:27:55 +08:00
; X64AVX512F-NEXT: vinsertf32x4 $0, %xmm0, %zmm1, %zmm0
[x86] add tests for insertelement; NFC llvm-svn: 315312
2017-10-10 21:45:25 +08:00
; X64AVX512F-NEXT: retq
%ins = insertelement <8 x double> <double 42.0, double 1.0, double 2.0, double 3.0, double 4.0, double 5.0, double 6.0, double 7.0>, double %x, i32 1
ret <8 x double> %ins
}