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/vec_cmp_uint-128.ll

1130 lines
34 KiB
LLVM
Raw Normal View History

Make utils/update_llc_test_checks.py note that the assertions are autogenerated. Also update existing test cases which appear to be generated by it and weren't modified (other than addition of the header) by rerunning it. llvm-svn: 253917
2015-11-24 05:33:58 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE41
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE42
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefix=ALL --check-prefix=XOP --check-prefix=XOPAVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+xop | FileCheck %s --check-prefix=ALL --check-prefix=XOP --check-prefix=XOPAVX2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512F
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512BW
;
; Equal
;
define <2 x i64> @eq_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
; SSE2-LABEL: eq_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
; SSE2: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
; SSE2-NEXT: pand %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: eq_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
; SSE41: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pcmpeqq %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: eq_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: pcmpeqq %xmm1, %xmm0
; SSE42-NEXT: retq
;
; AVX-LABEL: eq_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: eq_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomeqq %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp eq <2 x i64> %a, %b
%2 = sext <2 x i1> %1 to <2 x i64>
ret <2 x i64> %2
}
define <4 x i32> @eq_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
; SSE-LABEL: eq_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
; SSE: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: pcmpeqd %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: eq_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: eq_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomeqd %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp eq <4 x i32> %a, %b
%2 = sext <4 x i1> %1 to <4 x i32>
ret <4 x i32> %2
}
define <8 x i16> @eq_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
; SSE-LABEL: eq_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
; SSE: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: pcmpeqw %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: eq_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: eq_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomeqw %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp eq <8 x i16> %a, %b
%2 = sext <8 x i1> %1 to <8 x i16>
ret <8 x i16> %2
}
define <16 x i8> @eq_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
; SSE-LABEL: eq_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
; SSE: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: pcmpeqb %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: eq_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: eq_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomeqb %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp eq <16 x i8> %a, %b
%2 = sext <16 x i1> %1 to <16 x i8>
ret <16 x i8> %2
}
;
; Not Equal
;
define <2 x i64> @ne_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
; SSE2-LABEL: ne_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
; SSE2: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
; SSE2-NEXT: pand %xmm1, %xmm0
; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
; SSE2-NEXT: pxor %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: ne_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
; SSE41: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pcmpeqq %xmm1, %xmm0
; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
; SSE41-NEXT: pxor %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: ne_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: pcmpeqq %xmm1, %xmm0
; SSE42-NEXT: pcmpeqd %xmm1, %xmm1
; SSE42-NEXT: pxor %xmm1, %xmm0
; SSE42-NEXT: retq
;
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
; AVX1-LABEL: ne_v2i64:
; AVX1: # %bb.0:
; AVX1-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: ne_v2i64:
; AVX2: # %bb.0:
; AVX2-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: ne_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomneqq %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
;
; AVX512-LABEL: ne_v2i64:
; AVX512: # %bb.0:
; AVX512-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ne <2 x i64> %a, %b
%2 = sext <2 x i1> %1 to <2 x i64>
ret <2 x i64> %2
}
define <4 x i32> @ne_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
; SSE-LABEL: ne_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
; SSE: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: pcmpeqd %xmm1, %xmm0
; SSE-NEXT: pcmpeqd %xmm1, %xmm1
; SSE-NEXT: pxor %xmm1, %xmm0
; SSE-NEXT: retq
;
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
; AVX1-LABEL: ne_v4i32:
; AVX1: # %bb.0:
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: ne_v4i32:
; AVX2: # %bb.0:
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: ne_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomneqd %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
;
; AVX512-LABEL: ne_v4i32:
; AVX512: # %bb.0:
; AVX512-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ne <4 x i32> %a, %b
%2 = sext <4 x i1> %1 to <4 x i32>
ret <4 x i32> %2
}
define <8 x i16> @ne_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
; SSE-LABEL: ne_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
; SSE: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: pcmpeqw %xmm1, %xmm0
; SSE-NEXT: pcmpeqd %xmm1, %xmm1
; SSE-NEXT: pxor %xmm1, %xmm0
; SSE-NEXT: retq
;
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
; AVX1-LABEL: ne_v8i16:
; AVX1: # %bb.0:
; AVX1-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: ne_v8i16:
; AVX2: # %bb.0:
; AVX2-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: ne_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomneqw %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
;
; AVX512-LABEL: ne_v8i16:
; AVX512: # %bb.0:
; AVX512-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ne <8 x i16> %a, %b
%2 = sext <8 x i1> %1 to <8 x i16>
ret <8 x i16> %2
}
define <16 x i8> @ne_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
; SSE-LABEL: ne_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
; SSE: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: pcmpeqb %xmm1, %xmm0
; SSE-NEXT: pcmpeqd %xmm1, %xmm1
; SSE-NEXT: pxor %xmm1, %xmm0
; SSE-NEXT: retq
;
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
; AVX1-LABEL: ne_v16i8:
; AVX1: # %bb.0:
; AVX1-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: ne_v16i8:
; AVX2: # %bb.0:
; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: ne_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomneqb %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
;
; AVX512-LABEL: ne_v16i8:
; AVX512: # %bb.0:
; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
[X86] Use vpternlog to implement vector not under AVX512. Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation. This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner. llvm-svn: 323572
2018-01-27 06:17:40 +08:00
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ne <16 x i8> %a, %b
%2 = sext <16 x i1> %1 to <16 x i8>
ret <16 x i8> %2
}
;
; Greater Than Or Equal
;
define <2 x i64> @ge_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
; SSE2-LABEL: ge_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
; SSE2: # %bb.0:
[X86] When lowering unsigned v2i64 setcc without SSE42, flip the sign bits in the v2i64 type then bitcast to v4i32. This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step. Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast. There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it. This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues. llvm-svn: 344071
2018-10-10 03:05:50 +08:00
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: pxor %xmm2, %xmm1
; SSE2-NEXT: movdqa %xmm1, %xmm2
; SSE2-NEXT: pcmpgtd %xmm0, %xmm2
; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
; SSE2-NEXT: pcmpeqd %xmm0, %xmm1
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,3,3]
; SSE2-NEXT: pand %xmm3, %xmm0
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; SSE2-NEXT: por %xmm0, %xmm1
; SSE2-NEXT: pcmpeqd %xmm0, %xmm0
; SSE2-NEXT: pxor %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: ge_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
; SSE41: # %bb.0:
[X86] When lowering unsigned v2i64 setcc without SSE42, flip the sign bits in the v2i64 type then bitcast to v4i32. This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step. Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast. There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it. This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues. llvm-svn: 344071
2018-10-10 03:05:50 +08:00
; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pxor %xmm2, %xmm0
; SSE41-NEXT: pxor %xmm2, %xmm1
; SSE41-NEXT: movdqa %xmm1, %xmm2
; SSE41-NEXT: pcmpgtd %xmm0, %xmm2
; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
; SSE41-NEXT: pcmpeqd %xmm0, %xmm1
; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,3,3]
; SSE41-NEXT: pand %xmm3, %xmm0
; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; SSE41-NEXT: por %xmm0, %xmm1
; SSE41-NEXT: pcmpeqd %xmm0, %xmm0
; SSE41-NEXT: pxor %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: ge_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; SSE42-NEXT: pxor %xmm2, %xmm0
; SSE42-NEXT: pxor %xmm1, %xmm2
; SSE42-NEXT: pcmpgtq %xmm0, %xmm2
; SSE42-NEXT: pcmpeqd %xmm0, %xmm0
; SSE42-NEXT: pxor %xmm2, %xmm0
; SSE42-NEXT: retq
;
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
; AVX1-LABEL: ge_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
; AVX1: # %bb.0:
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX1-NEXT: vpxor %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpxor %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: ge_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
; AVX2: # %bb.0:
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX2-NEXT: vpxor %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vpxor %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: ge_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomgeuq %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
;
; AVX512-LABEL: ge_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
; AVX512: # %bb.0:
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
; AVX512-NEXT: vpmaxuq %zmm1, %zmm0, %zmm1
; AVX512-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp uge <2 x i64> %a, %b
%2 = sext <2 x i1> %1 to <2 x i64>
ret <2 x i64> %2
}
define <4 x i32> @ge_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
; SSE2-LABEL: ge_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
; SSE2: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: pxor %xmm1, %xmm2
; SSE2-NEXT: pcmpgtd %xmm0, %xmm2
; SSE2-NEXT: pcmpeqd %xmm0, %xmm0
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: ge_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
; SSE41: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pmaxud %xmm0, %xmm1
; SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: ge_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: pmaxud %xmm0, %xmm1
; SSE42-NEXT: pcmpeqd %xmm1, %xmm0
; SSE42-NEXT: retq
;
; AVX-LABEL: ge_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpmaxud %xmm1, %xmm0, %xmm1
; AVX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: ge_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomgeud %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp uge <4 x i32> %a, %b
%2 = sext <4 x i1> %1 to <4 x i32>
ret <4 x i32> %2
}
define <8 x i16> @ge_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
; SSE2-LABEL: ge_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
; SSE2: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: psubusw %xmm0, %xmm1
; SSE2-NEXT: pxor %xmm0, %xmm0
; SSE2-NEXT: pcmpeqw %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: ge_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
; SSE41: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pmaxuw %xmm0, %xmm1
; SSE41-NEXT: pcmpeqw %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: ge_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: pmaxuw %xmm0, %xmm1
; SSE42-NEXT: pcmpeqw %xmm1, %xmm0
; SSE42-NEXT: retq
;
; AVX-LABEL: ge_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpmaxuw %xmm1, %xmm0, %xmm1
; AVX-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: ge_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomgeuw %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp uge <8 x i16> %a, %b
%2 = sext <8 x i1> %1 to <8 x i16>
ret <8 x i16> %2
}
define <16 x i8> @ge_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
; SSE-LABEL: ge_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
; SSE: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: pmaxub %xmm0, %xmm1
; SSE-NEXT: pcmpeqb %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: ge_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpmaxub %xmm1, %xmm0, %xmm1
; AVX-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: ge_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomgeub %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp uge <16 x i8> %a, %b
%2 = sext <16 x i1> %1 to <16 x i8>
ret <16 x i8> %2
}
;
; Greater Than
;
define <2 x i64> @gt_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
; SSE2-LABEL: gt_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
; SSE2: # %bb.0:
[X86] When lowering unsigned v2i64 setcc without SSE42, flip the sign bits in the v2i64 type then bitcast to v4i32. This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step. Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast. There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it. This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues. llvm-svn: 344071
2018-10-10 03:05:50 +08:00
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: pxor %xmm2, %xmm1
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: movdqa %xmm0, %xmm2
; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; SSE2-NEXT: pand %xmm3, %xmm1
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
; SSE2-NEXT: por %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: gt_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
; SSE41: # %bb.0:
[X86] When lowering unsigned v2i64 setcc without SSE42, flip the sign bits in the v2i64 type then bitcast to v4i32. This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step. Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast. There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it. This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues. llvm-svn: 344071
2018-10-10 03:05:50 +08:00
; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pxor %xmm2, %xmm1
; SSE41-NEXT: pxor %xmm2, %xmm0
; SSE41-NEXT: movdqa %xmm0, %xmm2
; SSE41-NEXT: pcmpgtd %xmm1, %xmm2
; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
; SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; SSE41-NEXT: pand %xmm3, %xmm1
; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
; SSE41-NEXT: por %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: gt_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; SSE42-NEXT: pxor %xmm2, %xmm1
; SSE42-NEXT: pxor %xmm2, %xmm0
; SSE42-NEXT: pcmpgtq %xmm1, %xmm0
; SSE42-NEXT: retq
;
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX1-LABEL: gt_v2i64:
; AVX1: # %bb.0:
; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX1-NEXT: vpxor %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: gt_v2i64:
; AVX2: # %bb.0:
; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX2-NEXT: vpxor %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: gt_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomgtuq %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
;
; AVX512-LABEL: gt_v2i64:
; AVX512: # %bb.0:
; AVX512-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm1
; AVX512-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ugt <2 x i64> %a, %b
%2 = sext <2 x i1> %1 to <2 x i64>
ret <2 x i64> %2
}
define <4 x i32> @gt_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; SSE2-LABEL: gt_v4i32:
; SSE2: # %bb.0:
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
; SSE2-NEXT: pxor %xmm2, %xmm1
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: pcmpgtd %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: gt_v4i32:
; SSE41: # %bb.0:
; SSE41-NEXT: pminud %xmm0, %xmm1
; SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
; SSE41-NEXT: pxor %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: gt_v4i32:
; SSE42: # %bb.0:
; SSE42-NEXT: pminud %xmm0, %xmm1
; SSE42-NEXT: pcmpeqd %xmm1, %xmm0
; SSE42-NEXT: pcmpeqd %xmm1, %xmm1
; SSE42-NEXT: pxor %xmm1, %xmm0
; SSE42-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; AVX1-LABEL: gt_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
; AVX1: # %bb.0:
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm1
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX1-NEXT: retq
;
; AVX2-LABEL: gt_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
; AVX2: # %bb.0:
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX2-NEXT: vpminud %xmm1, %xmm0, %xmm1
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX2-NEXT: retq
;
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-LABEL: gt_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomgtud %xmm1, %xmm0, %xmm0
; XOP-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; AVX512-LABEL: gt_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
; AVX512: # %bb.0:
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX512-NEXT: vpminud %xmm1, %xmm0, %xmm1
; AVX512-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT: vzeroupper
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX512-NEXT: retq
%1 = icmp ugt <4 x i32> %a, %b
%2 = sext <4 x i1> %1 to <4 x i32>
ret <4 x i32> %2
}
define <8 x i16> @gt_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; SSE2-LABEL: gt_v8i16:
; SSE2: # %bb.0:
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
; SSE2-NEXT: pxor %xmm2, %xmm1
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: pcmpgtw %xmm1, %xmm0
; SSE2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; SSE41-LABEL: gt_v8i16:
; SSE41: # %bb.0:
; SSE41-NEXT: pminuw %xmm0, %xmm1
; SSE41-NEXT: pcmpeqw %xmm1, %xmm0
; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
; SSE41-NEXT: pxor %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: gt_v8i16:
; SSE42: # %bb.0:
; SSE42-NEXT: pminuw %xmm0, %xmm1
; SSE42-NEXT: pcmpeqw %xmm1, %xmm0
; SSE42-NEXT: pcmpeqd %xmm1, %xmm1
; SSE42-NEXT: pxor %xmm1, %xmm0
; SSE42-NEXT: retq
;
; AVX1-LABEL: gt_v8i16:
; AVX1: # %bb.0:
; AVX1-NEXT: vpminuw %xmm1, %xmm0, %xmm1
; AVX1-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: gt_v8i16:
; AVX2: # %bb.0:
; AVX2-NEXT: vpminuw %xmm1, %xmm0, %xmm1
; AVX2-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: gt_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomgtuw %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
;
; AVX512-LABEL: gt_v8i16:
; AVX512: # %bb.0:
; AVX512-NEXT: vpminuw %xmm1, %xmm0, %xmm1
; AVX512-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ugt <8 x i16> %a, %b
%2 = sext <8 x i1> %1 to <8 x i16>
ret <8 x i16> %2
}
define <16 x i8> @gt_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
; SSE-LABEL: gt_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
; SSE: # %bb.0:
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; SSE-NEXT: pminub %xmm0, %xmm1
; SSE-NEXT: pcmpeqb %xmm1, %xmm0
; SSE-NEXT: pcmpeqd %xmm1, %xmm1
; SSE-NEXT: pxor %xmm1, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: retq
;
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX1-LABEL: gt_v16i8:
; AVX1: # %bb.0:
; AVX1-NEXT: vpminub %xmm1, %xmm0, %xmm1
; AVX1-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: gt_v16i8:
; AVX2: # %bb.0:
; AVX2-NEXT: vpminub %xmm1, %xmm0, %xmm1
; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: gt_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomgtub %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
;
; AVX512-LABEL: gt_v16i8:
; AVX512: # %bb.0:
; AVX512-NEXT: vpminub %xmm1, %xmm0, %xmm1
; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ugt <16 x i8> %a, %b
%2 = sext <16 x i1> %1 to <16 x i8>
ret <16 x i8> %2
}
;
; Less Than Or Equal
;
define <2 x i64> @le_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
; SSE2-LABEL: le_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
; SSE2: # %bb.0:
[X86] When lowering unsigned v2i64 setcc without SSE42, flip the sign bits in the v2i64 type then bitcast to v4i32. This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step. Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast. There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it. This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues. llvm-svn: 344071
2018-10-10 03:05:50 +08:00
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: pxor %xmm2, %xmm1
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: movdqa %xmm0, %xmm2
; SSE2-NEXT: pcmpgtd %xmm1, %xmm2
; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
; SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSE2-NEXT: pand %xmm3, %xmm0
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; SSE2-NEXT: por %xmm0, %xmm1
; SSE2-NEXT: pcmpeqd %xmm0, %xmm0
; SSE2-NEXT: pxor %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: le_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
; SSE41: # %bb.0:
[X86] When lowering unsigned v2i64 setcc without SSE42, flip the sign bits in the v2i64 type then bitcast to v4i32. This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step. Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast. There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it. This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues. llvm-svn: 344071
2018-10-10 03:05:50 +08:00
; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pxor %xmm2, %xmm1
; SSE41-NEXT: pxor %xmm2, %xmm0
; SSE41-NEXT: movdqa %xmm0, %xmm2
; SSE41-NEXT: pcmpgtd %xmm1, %xmm2
; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
; SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSE41-NEXT: pand %xmm3, %xmm0
; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; SSE41-NEXT: por %xmm0, %xmm1
; SSE41-NEXT: pcmpeqd %xmm0, %xmm0
; SSE41-NEXT: pxor %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: le_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; SSE42-NEXT: pxor %xmm2, %xmm1
; SSE42-NEXT: pxor %xmm2, %xmm0
; SSE42-NEXT: pcmpgtq %xmm1, %xmm0
; SSE42-NEXT: pcmpeqd %xmm1, %xmm1
; SSE42-NEXT: pxor %xmm1, %xmm0
; SSE42-NEXT: retq
;
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
; AVX1-LABEL: le_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
; AVX1: # %bb.0:
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX1-NEXT: vpxor %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: le_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
; AVX2: # %bb.0:
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX2-NEXT: vpxor %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vpcmpgtq %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: le_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomleuq %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
;
; AVX512-LABEL: le_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
; AVX512: # %bb.0:
Followup on Proposal to move MIR physical register namespace to '$' sigil. Discussed here: http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html In preparation for adding support for named vregs we are changing the sigil for physical registers in MIR to '$' from '%'. This will prevent name clashes of named physical register with named vregs. llvm-svn: 323922
2018-02-01 06:04:26 +08:00
; AVX512-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
[X86] Allow VPMAXUQ/VPMAXSQ/VPMINUQ/VPMINSQ to be used with 128/256 bit vectors when AVX512 is enabled. These instructions can be used by widening to 512-bits and extracting back to 128/256. We do similar to several other instructions already. llvm-svn: 319641
2017-12-04 15:21:01 +08:00
; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm1
; AVX512-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ule <2 x i64> %a, %b
%2 = sext <2 x i1> %1 to <2 x i64>
ret <2 x i64> %2
}
define <4 x i32> @le_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
; SSE2-LABEL: le_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
; SSE2: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
; SSE2-NEXT: pxor %xmm2, %xmm1
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: pcmpgtd %xmm1, %xmm0
; SSE2-NEXT: pcmpeqd %xmm1, %xmm1
; SSE2-NEXT: pxor %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: le_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
; SSE41: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pminud %xmm0, %xmm1
; SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: le_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: pminud %xmm0, %xmm1
; SSE42-NEXT: pcmpeqd %xmm1, %xmm0
; SSE42-NEXT: retq
;
; AVX-LABEL: le_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpminud %xmm1, %xmm0, %xmm1
; AVX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: le_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomleud %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp ule <4 x i32> %a, %b
%2 = sext <4 x i1> %1 to <4 x i32>
ret <4 x i32> %2
}
define <8 x i16> @le_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
; SSE2-LABEL: le_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
; SSE2: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: psubusw %xmm1, %xmm0
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqw %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: le_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
; SSE41: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pminuw %xmm0, %xmm1
; SSE41-NEXT: pcmpeqw %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: le_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: pminuw %xmm0, %xmm1
; SSE42-NEXT: pcmpeqw %xmm1, %xmm0
; SSE42-NEXT: retq
;
; AVX-LABEL: le_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpminuw %xmm1, %xmm0, %xmm1
; AVX-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: le_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomleuw %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp ule <8 x i16> %a, %b
%2 = sext <8 x i1> %1 to <8 x i16>
ret <8 x i16> %2
}
define <16 x i8> @le_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
; SSE-LABEL: le_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
; SSE: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: pminub %xmm0, %xmm1
; SSE-NEXT: pcmpeqb %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: le_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
; AVX: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX-NEXT: vpminub %xmm1, %xmm0, %xmm1
; AVX-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; XOP-LABEL: le_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomleub %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
%1 = icmp ule <16 x i8> %a, %b
%2 = sext <16 x i1> %1 to <16 x i8>
ret <16 x i8> %2
}
;
; Less Than
;
define <2 x i64> @lt_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
; SSE2-LABEL: lt_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
; SSE2: # %bb.0:
[X86] When lowering unsigned v2i64 setcc without SSE42, flip the sign bits in the v2i64 type then bitcast to v4i32. This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step. Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast. There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it. This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues. llvm-svn: 344071
2018-10-10 03:05:50 +08:00
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: pxor %xmm2, %xmm1
; SSE2-NEXT: movdqa %xmm1, %xmm2
; SSE2-NEXT: pcmpgtd %xmm0, %xmm2
; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
; SSE2-NEXT: pcmpeqd %xmm0, %xmm1
; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; SSE2-NEXT: pand %xmm3, %xmm1
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
; SSE2-NEXT: por %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: lt_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
; SSE41: # %bb.0:
[X86] When lowering unsigned v2i64 setcc without SSE42, flip the sign bits in the v2i64 type then bitcast to v4i32. This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step. Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast. There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it. This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues. llvm-svn: 344071
2018-10-10 03:05:50 +08:00
; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [9223372039002259456,9223372039002259456]
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE41-NEXT: pxor %xmm2, %xmm0
; SSE41-NEXT: pxor %xmm2, %xmm1
; SSE41-NEXT: movdqa %xmm1, %xmm2
; SSE41-NEXT: pcmpgtd %xmm0, %xmm2
; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[0,0,2,2]
; SSE41-NEXT: pcmpeqd %xmm0, %xmm1
; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; SSE41-NEXT: pand %xmm3, %xmm1
; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,1,3,3]
; SSE41-NEXT: por %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: lt_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
; SSE42: # %bb.0:
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE42-NEXT: movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; SSE42-NEXT: pxor %xmm2, %xmm0
; SSE42-NEXT: pxor %xmm1, %xmm2
; SSE42-NEXT: pcmpgtq %xmm0, %xmm2
; SSE42-NEXT: movdqa %xmm2, %xmm0
; SSE42-NEXT: retq
;
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX1-LABEL: lt_v2i64:
; AVX1: # %bb.0:
; AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX1-NEXT: vpxor %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpxor %xmm2, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: lt_v2i64:
; AVX2: # %bb.0:
; AVX2-NEXT: vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
; AVX2-NEXT: vpxor %xmm2, %xmm0, %xmm0
; AVX2-NEXT: vpxor %xmm2, %xmm1, %xmm1
; AVX2-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: lt_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomltuq %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
;
; AVX512-LABEL: lt_v2i64:
; AVX512: # %bb.0:
; AVX512-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0
; AVX512-NEXT: vpmaxuq %zmm1, %zmm0, %zmm1
; AVX512-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ult <2 x i64> %a, %b
%2 = sext <2 x i1> %1 to <2 x i64>
ret <2 x i64> %2
}
define <4 x i32> @lt_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; SSE2-LABEL: lt_v4i32:
; SSE2: # %bb.0:
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: pxor %xmm1, %xmm2
; SSE2-NEXT: pcmpgtd %xmm0, %xmm2
; SSE2-NEXT: movdqa %xmm2, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: lt_v4i32:
; SSE41: # %bb.0:
; SSE41-NEXT: pmaxud %xmm0, %xmm1
; SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
; SSE41-NEXT: pxor %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: lt_v4i32:
; SSE42: # %bb.0:
; SSE42-NEXT: pmaxud %xmm0, %xmm1
; SSE42-NEXT: pcmpeqd %xmm1, %xmm0
; SSE42-NEXT: pcmpeqd %xmm1, %xmm1
; SSE42-NEXT: pxor %xmm1, %xmm0
; SSE42-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; AVX1-LABEL: lt_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
; AVX1: # %bb.0:
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX1-NEXT: vpmaxud %xmm1, %xmm0, %xmm1
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX1-NEXT: retq
;
; AVX2-LABEL: lt_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
; AVX2: # %bb.0:
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX2-NEXT: vpmaxud %xmm1, %xmm0, %xmm1
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX2-NEXT: retq
;
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-LABEL: lt_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomltud %xmm1, %xmm0, %xmm0
; XOP-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; AVX512-LABEL: lt_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
; AVX512: # %bb.0:
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX512-NEXT: vpmaxud %xmm1, %xmm0, %xmm1
; AVX512-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT: vzeroupper
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; AVX512-NEXT: retq
%1 = icmp ult <4 x i32> %a, %b
%2 = sext <4 x i1> %1 to <4 x i32>
ret <4 x i32> %2
}
define <8 x i16> @lt_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; SSE2-LABEL: lt_v8i16:
; SSE2: # %bb.0:
; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
; SSE2-NEXT: pxor %xmm2, %xmm0
; SSE2-NEXT: pxor %xmm1, %xmm2
; SSE2-NEXT: pcmpgtw %xmm0, %xmm2
; SSE2-NEXT: movdqa %xmm2, %xmm0
; SSE2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; SSE41-LABEL: lt_v8i16:
; SSE41: # %bb.0:
; SSE41-NEXT: pmaxuw %xmm0, %xmm1
; SSE41-NEXT: pcmpeqw %xmm1, %xmm0
; SSE41-NEXT: pcmpeqd %xmm1, %xmm1
; SSE41-NEXT: pxor %xmm1, %xmm0
; SSE41-NEXT: retq
;
; SSE42-LABEL: lt_v8i16:
; SSE42: # %bb.0:
; SSE42-NEXT: pmaxuw %xmm0, %xmm1
; SSE42-NEXT: pcmpeqw %xmm1, %xmm0
; SSE42-NEXT: pcmpeqd %xmm1, %xmm1
; SSE42-NEXT: pxor %xmm1, %xmm0
; SSE42-NEXT: retq
;
; AVX1-LABEL: lt_v8i16:
; AVX1: # %bb.0:
; AVX1-NEXT: vpmaxuw %xmm1, %xmm0, %xmm1
; AVX1-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: lt_v8i16:
; AVX2: # %bb.0:
; AVX2-NEXT: vpmaxuw %xmm1, %xmm0, %xmm1
; AVX2-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: lt_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomltuw %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
;
; AVX512-LABEL: lt_v8i16:
; AVX512: # %bb.0:
; AVX512-NEXT: vpmaxuw %xmm1, %xmm0, %xmm1
; AVX512-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ult <8 x i16> %a, %b
%2 = sext <8 x i1> %1 to <8 x i16>
ret <8 x i16> %2
}
define <16 x i8> @lt_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
; SSE-LABEL: lt_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
; SSE: # %bb.0:
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; SSE-NEXT: pmaxub %xmm0, %xmm1
; SSE-NEXT: pcmpeqb %xmm1, %xmm0
; SSE-NEXT: pcmpeqd %xmm1, %xmm1
; SSE-NEXT: pxor %xmm1, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; SSE-NEXT: retq
;
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
; AVX1-LABEL: lt_v16i8:
; AVX1: # %bb.0:
; AVX1-NEXT: vpmaxub %xmm1, %xmm0, %xmm1
; AVX1-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: lt_v16i8:
; AVX2: # %bb.0:
; AVX2-NEXT: vpmaxub %xmm1, %xmm0, %xmm1
; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX2-NEXT: vpxor %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
;
; XOP-LABEL: lt_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
; XOP: # %bb.0:
[X86][XOP] Added support for the lowering of 128-bit vector integer comparisons to XOP PCOM/PCOMU instructions. The XOP vector integer comparisons can deal with all signed/unsigned comparison cases directly and can be easily commuted as well (D7646). llvm-svn: 249976
2015-10-11 22:15:17 +08:00
; XOP-NEXT: vpcomltub %xmm1, %xmm0, %xmm0
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
; XOP-NEXT: retq
[X86] Use min/max for vector ult/ugt compares if avoids a sign flip. Summary: Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare. I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior. Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into. I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare. Reviewers: spatel, RKSimon Reviewed By: RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D42935 llvm-svn: 324842
2018-02-12 01:11:40 +08:00
;
; AVX512-LABEL: lt_v16i8:
; AVX512: # %bb.0:
; AVX512-NEXT: vpmaxub %xmm1, %xmm0, %xmm1
; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX512-NEXT: vpternlogq $15, %zmm0, %zmm0, %zmm0
; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
[X86][SSE] Vector signed/unsigned integer compare tests. llvm-svn: 249954
2015-10-11 06:21:05 +08:00
%1 = icmp ult <16 x i8> %a, %b
%2 = sext <16 x i1> %1 to <16 x i8>
ret <16 x i8> %2
}