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/merge-consecutive-loads-256.ll

683 lines
28 KiB
LLVM
Raw Normal View History

[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
[X86][SSE] Add AVX512 merge consecutive load tests Add AVX512F/AVX512BW 512-bit tests. Add AVX512F tests to existing 128/256-bit tests. llvm-svn: 259410
2016-02-02 05:30:50 +08:00
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX --check-prefix=AVX512F
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; Just one 32-bit run to make sure we do reasonable things.
; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32-AVX
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
define <4 x double> @merge_4f64_2f64_23(<2 x double>* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_4f64_2f64_23:
[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] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
; AVX-NEXT: vmovups 32(%rdi), %ymm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4f64_2f64_23:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vmovups 32(%eax), %ymm0
; X32-AVX-NEXT: retl
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
%ptr0 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 2
%ptr1 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 3
%val0 = load <2 x double>, <2 x double>* %ptr0
%val1 = load <2 x double>, <2 x double>* %ptr1
%res = shufflevector <2 x double> %val0, <2 x double> %val1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
ret <4 x double> %res
}
define <4 x double> @merge_4f64_2f64_2z(<2 x double>* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_4f64_2f64_2z:
[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] Remove patterns for inserting a load into a zero vector. We can instead block the load folding isProfitableToFold. Then isel will emit a register->register move for the zeroing part and a separate load. The PostProcessISelDAG should be able to remove the register->register move. This saves us patterns and fixes the fact that we only had unaligned load patterns. The test changes show places where we should have been using an aligned load. llvm-svn: 336828
2018-07-12 02:09:04 +08:00
; AVX-NEXT: vmovaps 32(%rdi), %xmm0
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4f64_2f64_2z:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86] Remove patterns for inserting a load into a zero vector. We can instead block the load folding isProfitableToFold. Then isel will emit a register->register move for the zeroing part and a separate load. The PostProcessISelDAG should be able to remove the register->register move. This saves us patterns and fixes the fact that we only had unaligned load patterns. The test changes show places where we should have been using an aligned load. llvm-svn: 336828
2018-07-12 02:09:04 +08:00
; X32-AVX-NEXT: vmovaps 32(%eax), %xmm0
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
%ptr0 = getelementptr inbounds <2 x double>, <2 x double>* %ptr, i64 2
%val0 = load <2 x double>, <2 x double>* %ptr0
%res = shufflevector <2 x double> %val0, <2 x double> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
ret <4 x double> %res
}
define <4 x double> @merge_4f64_f64_2345(double* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_4f64_f64_2345:
[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] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
; AVX-NEXT: vmovups 16(%rdi), %ymm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4f64_f64_2345:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vmovups 16(%eax), %ymm0
; X32-AVX-NEXT: retl
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
%ptr0 = getelementptr inbounds double, double* %ptr, i64 2
%ptr1 = getelementptr inbounds double, double* %ptr, i64 3
%ptr2 = getelementptr inbounds double, double* %ptr, i64 4
%ptr3 = getelementptr inbounds double, double* %ptr, i64 5
%val0 = load double, double* %ptr0
%val1 = load double, double* %ptr1
%val2 = load double, double* %ptr2
%val3 = load double, double* %ptr3
%res0 = insertelement <4 x double> undef, double %val0, i32 0
%res1 = insertelement <4 x double> %res0, double %val1, i32 1
%res2 = insertelement <4 x double> %res1, double %val2, i32 2
%res3 = insertelement <4 x double> %res2, double %val3, i32 3
ret <4 x double> %res3
}
define <4 x double> @merge_4f64_f64_3zuu(double* %ptr) nounwind uwtable noinline ssp {
[X86][AVX512] Add support for AVX512 VMOVSD/VMOVSS shuffle decoding llvm-svn: 259427
2016-02-02 06:26:28 +08:00
; AVX-LABEL: merge_4f64_f64_3zuu:
[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][AVX512] Add support for AVX512 VMOVSD/VMOVSS shuffle decoding llvm-svn: 259427
2016-02-02 06:26:28 +08:00
; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4f64_f64_3zuu:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; X32-AVX-NEXT: retl
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
%ptr0 = getelementptr inbounds double, double* %ptr, i64 3
%val0 = load double, double* %ptr0
%res0 = insertelement <4 x double> undef, double %val0, i32 0
%res1 = insertelement <4 x double> %res0, double 0.0, i32 1
ret <4 x double> %res1
}
define <4 x double> @merge_4f64_f64_34uu(double* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_4f64_f64_34uu:
[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] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
; AVX-NEXT: vmovups 24(%rdi), %xmm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4f64_f64_34uu:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vmovups 24(%eax), %xmm0
; X32-AVX-NEXT: retl
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
%ptr0 = getelementptr inbounds double, double* %ptr, i64 3
%ptr1 = getelementptr inbounds double, double* %ptr, i64 4
%val0 = load double, double* %ptr0
%val1 = load double, double* %ptr1
%res0 = insertelement <4 x double> undef, double %val0, i32 0
%res1 = insertelement <4 x double> %res0, double %val1, i32 1
ret <4 x double> %res1
}
define <4 x double> @merge_4f64_f64_45zz(double* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_4f64_f64_45zz:
[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] Fix some isel patterns that used aligned vector load instructions with unaligned predicates. These patterns weren't checking the alignment of the load, but were using the aligned instructions. This will cause a GP fault if the data isn't aligned. I believe these were introduced in r312450. llvm-svn: 326967
2018-03-08 08:21:17 +08:00
; AVX-NEXT: vmovups 32(%rdi), %xmm0
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4f64_f64_45zz:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86] Fix some isel patterns that used aligned vector load instructions with unaligned predicates. These patterns weren't checking the alignment of the load, but were using the aligned instructions. This will cause a GP fault if the data isn't aligned. I believe these were introduced in r312450. llvm-svn: 326967
2018-03-08 08:21:17 +08:00
; X32-AVX-NEXT: vmovups 32(%eax), %xmm0
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][SSE] Added more exhaustive merge consecutive load tests llvm-svn: 257876
2016-01-15 17:52:50 +08:00
%ptr0 = getelementptr inbounds double, double* %ptr, i64 4
%ptr1 = getelementptr inbounds double, double* %ptr, i64 5
%val0 = load double, double* %ptr0
%val1 = load double, double* %ptr1
%res0 = insertelement <4 x double> zeroinitializer, double %val0, i32 0
%res1 = insertelement <4 x double> %res0, double %val1, i32 1
ret <4 x double> %res1
}
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
define <4 x double> @merge_4f64_f64_34z6(double* %ptr) nounwind uwtable noinline ssp {
[X86][AVX] Add shuffle masking support for EltsFromConsecutiveLoads Add support for the case where we have a consecutive load (which must include the first + last elements) with a mixture of undef/zero elements. We load the vector and then apply a shuffle to clear the zero'd elements. Differential Revision: http://reviews.llvm.org/D17297 llvm-svn: 261490
2016-02-22 03:15:48 +08:00
; AVX-LABEL: merge_4f64_f64_34z6:
[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] Add custom execution domain fixing for BLENDPD/BLENDPS/PBLENDD/PBLENDW (PR34873) Add support for custom execution domain fixing and implement support for BLENDPD/BLENDPS/PBLENDD/PBLENDW. Differential Revision: https://reviews.llvm.org/D42042 llvm-svn: 322524
2018-01-16 06:18:45 +08:00
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3],ymm0[4,5],mem[6,7]
[X86][AVX] Add shuffle masking support for EltsFromConsecutiveLoads Add support for the case where we have a consecutive load (which must include the first + last elements) with a mixture of undef/zero elements. We load the vector and then apply a shuffle to clear the zero'd elements. Differential Revision: http://reviews.llvm.org/D17297 llvm-svn: 261490
2016-02-22 03:15:48 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4f64_f64_34z6:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Add custom execution domain fixing for BLENDPD/BLENDPS/PBLENDD/PBLENDW (PR34873) Add support for custom execution domain fixing and implement support for BLENDPD/BLENDPS/PBLENDD/PBLENDW. Differential Revision: https://reviews.llvm.org/D42042 llvm-svn: 322524
2018-01-16 06:18:45 +08:00
; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3],ymm0[4,5],mem[6,7]
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds double, double* %ptr, i64 3
%ptr1 = getelementptr inbounds double, double* %ptr, i64 4
%ptr3 = getelementptr inbounds double, double* %ptr, i64 6
%val0 = load double, double* %ptr0
%val1 = load double, double* %ptr1
%val3 = load double, double* %ptr3
%res0 = insertelement <4 x double> undef, double %val0, i32 0
%res1 = insertelement <4 x double> %res0, double %val1, i32 1
%res2 = insertelement <4 x double> %res1, double 0.0, i32 2
%res3 = insertelement <4 x double> %res2, double %val3, i32 3
ret <4 x double> %res3
}
define <4 x i64> @merge_4i64_2i64_3z(<2 x i64>* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_4i64_2i64_3z:
[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] Remove patterns for inserting a load into a zero vector. We can instead block the load folding isProfitableToFold. Then isel will emit a register->register move for the zeroing part and a separate load. The PostProcessISelDAG should be able to remove the register->register move. This saves us patterns and fixes the fact that we only had unaligned load patterns. The test changes show places where we should have been using an aligned load. llvm-svn: 336828
2018-07-12 02:09:04 +08:00
; AVX-NEXT: vmovaps 48(%rdi), %xmm0
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4i64_2i64_3z:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86] Remove patterns for inserting a load into a zero vector. We can instead block the load folding isProfitableToFold. Then isel will emit a register->register move for the zeroing part and a separate load. The PostProcessISelDAG should be able to remove the register->register move. This saves us patterns and fixes the fact that we only had unaligned load patterns. The test changes show places where we should have been using an aligned load. llvm-svn: 336828
2018-07-12 02:09:04 +08:00
; X32-AVX-NEXT: vmovaps 48(%eax), %xmm0
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds <2 x i64>, <2 x i64>* %ptr, i64 3
%val0 = load <2 x i64>, <2 x i64>* %ptr0
%res = shufflevector <2 x i64> %val0, <2 x i64> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
ret <4 x i64> %res
}
define <4 x i64> @merge_4i64_i64_1234(i64* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_4i64_i64_1234:
[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][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
; AVX-NEXT: vmovups 8(%rdi), %ymm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4i64_i64_1234:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vmovups 8(%eax), %ymm0
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i64, i64* %ptr, i64 1
%ptr1 = getelementptr inbounds i64, i64* %ptr, i64 2
%ptr2 = getelementptr inbounds i64, i64* %ptr, i64 3
%ptr3 = getelementptr inbounds i64, i64* %ptr, i64 4
%val0 = load i64, i64* %ptr0
%val1 = load i64, i64* %ptr1
%val2 = load i64, i64* %ptr2
%val3 = load i64, i64* %ptr3
%res0 = insertelement <4 x i64> undef, i64 %val0, i32 0
%res1 = insertelement <4 x i64> %res0, i64 %val1, i32 1
%res2 = insertelement <4 x i64> %res1, i64 %val2, i32 2
%res3 = insertelement <4 x i64> %res2, i64 %val3, i32 3
ret <4 x i64> %res3
}
define <4 x i64> @merge_4i64_i64_1zzu(i64* %ptr) nounwind uwtable noinline ssp {
[X86][AVX512] Add support for AVX512 VMOVQ (load) shuffle decoding llvm-svn: 259496
2016-02-02 21:32:56 +08:00
; AVX-LABEL: merge_4i64_i64_1zzu:
[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] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
[X86][AVX512] Add support for AVX512 VMOVQ (load) shuffle decoding llvm-svn: 259496
2016-02-02 21:32:56 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4i64_i64_1zzu:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i64, i64* %ptr, i64 1
%val0 = load i64, i64* %ptr0
%res0 = insertelement <4 x i64> undef, i64 %val0, i32 0
%res1 = insertelement <4 x i64> %res0, i64 0, i32 1
[X86][AVX] Fixed typo in zero element insertion llvm-svn: 278798
2016-08-16 21:33:33 +08:00
%res2 = insertelement <4 x i64> %res1, i64 0, i32 2
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
ret <4 x i64> %res2
}
define <4 x i64> @merge_4i64_i64_23zz(i64* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_4i64_i64_23zz:
[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] Fix some isel patterns that used aligned vector load instructions with unaligned predicates. These patterns weren't checking the alignment of the load, but were using the aligned instructions. This will cause a GP fault if the data isn't aligned. I believe these were introduced in r312450. llvm-svn: 326967
2018-03-08 08:21:17 +08:00
; AVX-NEXT: vmovups 16(%rdi), %xmm0
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_4i64_i64_23zz:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86] Fix some isel patterns that used aligned vector load instructions with unaligned predicates. These patterns weren't checking the alignment of the load, but were using the aligned instructions. This will cause a GP fault if the data isn't aligned. I believe these were introduced in r312450. llvm-svn: 326967
2018-03-08 08:21:17 +08:00
; X32-AVX-NEXT: vmovups 16(%eax), %xmm0
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i64, i64* %ptr, i64 2
%ptr1 = getelementptr inbounds i64, i64* %ptr, i64 3
%val0 = load i64, i64* %ptr0
%val1 = load i64, i64* %ptr1
%res0 = insertelement <4 x i64> zeroinitializer, i64 %val0, i32 0
%res1 = insertelement <4 x i64> %res0, i64 %val1, i32 1
ret <4 x i64> %res1
}
define <8 x float> @merge_8f32_2f32_23z5(<2 x float>* %ptr) nounwind uwtable noinline ssp {
[X86][SSE] Add AVX512 merge consecutive load tests Add AVX512F/AVX512BW 512-bit tests. Add AVX512F tests to existing 128/256-bit tests. llvm-svn: 259410
2016-02-02 05:30:50 +08:00
; AVX1-LABEL: merge_8f32_2f32_23z5:
[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][SSE] Add support for combining target shuffles to PSLLDQ/PSRLDQ byte shifts llvm-svn: 278502
2016-08-12 19:24:34 +08:00
; AVX1-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX1-NEXT: vmovups 16(%rdi), %xmm1
; AVX1-NEXT: vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7]
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
[X86][SSE] Add AVX512 merge consecutive load tests Add AVX512F/AVX512BW 512-bit tests. Add AVX512F tests to existing 128/256-bit tests. llvm-svn: 259410
2016-02-02 05:30:50 +08:00
; AVX1-NEXT: retq
;
; AVX2-LABEL: merge_8f32_2f32_23z5:
[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][SSE] Add support for combining target shuffles to PSLLDQ/PSRLDQ byte shifts llvm-svn: 278502
2016-08-12 19:24:34 +08:00
; AVX2-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX2-NEXT: vmovdqu 16(%rdi), %xmm1
; AVX2-NEXT: vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7]
; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
[X86][SSE] Add AVX512 merge consecutive load tests Add AVX512F/AVX512BW 512-bit tests. Add AVX512F tests to existing 128/256-bit tests. llvm-svn: 259410
2016-02-02 05:30:50 +08:00
; AVX2-NEXT: retq
;
; AVX512F-LABEL: merge_8f32_2f32_23z5:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512F: # %bb.0:
[X86][SSE] Add support for combining target shuffles to PSLLDQ/PSRLDQ byte shifts llvm-svn: 278502
2016-08-12 19:24:34 +08:00
; AVX512F-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX512F-NEXT: vmovdqu 16(%rdi), %xmm1
; AVX512F-NEXT: vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7]
; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
[X86][SSE] Add AVX512 merge consecutive load tests Add AVX512F/AVX512BW 512-bit tests. Add AVX512F tests to existing 128/256-bit tests. llvm-svn: 259410
2016-02-02 05:30:50 +08:00
; AVX512F-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_8f32_2f32_23z5:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Add custom execution domain fixing for BLENDPD/BLENDPS/PBLENDD/PBLENDW (PR34873) Add support for custom execution domain fixing and implement support for BLENDPD/BLENDPS/PBLENDD/PBLENDW. Differential Revision: https://reviews.llvm.org/D42042 llvm-svn: 322524
2018-01-16 06:18:45 +08:00
; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3],ymm0[4,5],mem[6,7]
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 2
%ptr1 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 3
%ptr3 = getelementptr inbounds <2 x float>, <2 x float>* %ptr, i64 5
%val0 = load <2 x float>, <2 x float>* %ptr0
%val1 = load <2 x float>, <2 x float>* %ptr1
%val3 = load <2 x float>, <2 x float>* %ptr3
%res01 = shufflevector <2 x float> %val0, <2 x float> %val1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%res23 = shufflevector <2 x float> zeroinitializer, <2 x float> %val3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%res = shufflevector <4 x float> %res01, <4 x float> %res23, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x float> %res
}
define <8 x float> @merge_8f32_4f32_z2(<4 x float>* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_8f32_4f32_z2:
[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][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; AVX-NEXT: vinsertf128 $1, 32(%rdi), %ymm0, %ymm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_8f32_4f32_z2:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; X32-AVX-NEXT: vinsertf128 $1, 32(%eax), %ymm0, %ymm0
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr1 = getelementptr inbounds <4 x float>, <4 x float>* %ptr, i64 2
%val1 = load <4 x float>, <4 x float>* %ptr1
%res = shufflevector <4 x float> zeroinitializer, <4 x float> %val1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x float> %res
}
define <8 x float> @merge_8f32_f32_12zzuuzz(float* %ptr) nounwind uwtable noinline ssp {
[X86][AVX] Add support for 64-bit VZEXT_LOAD of 256/512-bit vectors to EltsFromConsecutiveLoads Follow up to D16217 and D16729 This change uncovered an odd pattern where VZEXT_LOAD v4i64 was being lowered to a load of the lower v2i64 (so the 2nd i64 destination element wasn't being zeroed), I can't find any use/reason for this and have removed the pattern and replaced it so only the 1st i64 element is loaded and the upper bits all zeroed. This matches the description for X86ISD::VZEXT_LOAD Differential Revision: http://reviews.llvm.org/D16768 llvm-svn: 259635
2016-02-03 17:41:59 +08:00
; AVX-LABEL: merge_8f32_f32_12zzuuzz:
[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] Select domain for 32/64-bit partial loads for EltsFromConsecutiveLoads Choose between MOVD/MOVSS and MOVQ/MOVSD depending on the target vector type. This has a lot fewer test changes than trying to add this to X86InstrInfo::setExecutionDomain..... llvm-svn: 259816
2016-02-05 03:27:51 +08:00
; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
[X86][AVX] Add support for 64-bit VZEXT_LOAD of 256/512-bit vectors to EltsFromConsecutiveLoads Follow up to D16217 and D16729 This change uncovered an odd pattern where VZEXT_LOAD v4i64 was being lowered to a load of the lower v2i64 (so the 2nd i64 destination element wasn't being zeroed), I can't find any use/reason for this and have removed the pattern and replaced it so only the 1st i64 element is loaded and the upper bits all zeroed. This matches the description for X86ISD::VZEXT_LOAD Differential Revision: http://reviews.llvm.org/D16768 llvm-svn: 259635
2016-02-03 17:41:59 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_8f32_f32_12zzuuzz:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Select domain for 32/64-bit partial loads for EltsFromConsecutiveLoads Choose between MOVD/MOVSS and MOVQ/MOVSD depending on the target vector type. This has a lot fewer test changes than trying to add this to X86InstrInfo::setExecutionDomain..... llvm-svn: 259816
2016-02-05 03:27:51 +08:00
; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds float, float* %ptr, i64 1
%ptr1 = getelementptr inbounds float, float* %ptr, i64 2
%val0 = load float, float* %ptr0
%val1 = load float, float* %ptr1
%res0 = insertelement <8 x float> undef, float %val0, i32 0
%res1 = insertelement <8 x float> %res0, float %val1, i32 1
%res2 = insertelement <8 x float> %res1, float 0.0, i32 2
%res3 = insertelement <8 x float> %res2, float 0.0, i32 3
%res6 = insertelement <8 x float> %res3, float 0.0, i32 6
%res7 = insertelement <8 x float> %res6, float 0.0, i32 7
ret <8 x float> %res7
}
define <8 x float> @merge_8f32_f32_1u3u5zu8(float* %ptr) nounwind uwtable noinline ssp {
[X86][AVX] Add shuffle masking support for EltsFromConsecutiveLoads Add support for the case where we have a consecutive load (which must include the first + last elements) with a mixture of undef/zero elements. We load the vector and then apply a shuffle to clear the zero'd elements. Differential Revision: http://reviews.llvm.org/D17297 llvm-svn: 261490
2016-02-22 03:15:48 +08:00
; AVX-LABEL: merge_8f32_f32_1u3u5zu8:
[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] SET0 to use XMM registers where possible PR26018 PR32862 Differential Revision: https://reviews.llvm.org/D35839 llvm-svn: 309298
2017-07-28 01:47:01 +08:00
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
[X86][AVX] Add shuffle masking support for EltsFromConsecutiveLoads Add support for the case where we have a consecutive load (which must include the first + last elements) with a mixture of undef/zero elements. We load the vector and then apply a shuffle to clear the zero'd elements. Differential Revision: http://reviews.llvm.org/D17297 llvm-svn: 261490
2016-02-22 03:15:48 +08:00
; AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7]
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_8f32_f32_1u3u5zu8:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86] SET0 to use XMM registers where possible PR26018 PR32862 Differential Revision: https://reviews.llvm.org/D35839 llvm-svn: 309298
2017-07-28 01:47:01 +08:00
; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
[X86][AVX] Add shuffle masking support for EltsFromConsecutiveLoads Add support for the case where we have a consecutive load (which must include the first + last elements) with a mixture of undef/zero elements. We load the vector and then apply a shuffle to clear the zero'd elements. Differential Revision: http://reviews.llvm.org/D17297 llvm-svn: 261490
2016-02-22 03:15:48 +08:00
; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7]
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds float, float* %ptr, i64 1
%ptr2 = getelementptr inbounds float, float* %ptr, i64 3
%ptr4 = getelementptr inbounds float, float* %ptr, i64 5
%ptr7 = getelementptr inbounds float, float* %ptr, i64 8
%val0 = load float, float* %ptr0
%val2 = load float, float* %ptr2
%val4 = load float, float* %ptr4
%val7 = load float, float* %ptr7
%res0 = insertelement <8 x float> undef, float %val0, i32 0
%res2 = insertelement <8 x float> %res0, float %val2, i32 2
%res4 = insertelement <8 x float> %res2, float %val4, i32 4
%res5 = insertelement <8 x float> %res4, float 0.0, i32 5
%res7 = insertelement <8 x float> %res5, float %val7, i32 7
ret <8 x float> %res7
}
define <8 x i32> @merge_8i32_4i32_z3(<4 x i32>* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_8i32_4i32_z3:
[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][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; AVX-NEXT: vinsertf128 $1, 48(%rdi), %ymm0, %ymm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_8i32_4i32_z3:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; X32-AVX-NEXT: vinsertf128 $1, 48(%eax), %ymm0, %ymm0
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr1 = getelementptr inbounds <4 x i32>, <4 x i32>* %ptr, i64 3
%val1 = load <4 x i32>, <4 x i32>* %ptr1
%res = shufflevector <4 x i32> zeroinitializer, <4 x i32> %val1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <8 x i32> %res
}
define <8 x i32> @merge_8i32_i32_56zz9uzz(i32* %ptr) nounwind uwtable noinline ssp {
[X86][SSE] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; AVX-LABEL: merge_8i32_i32_56zz9uzz:
[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] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_8i32_i32_56zz9uzz:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; X32-AVX-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i32, i32* %ptr, i64 5
%ptr1 = getelementptr inbounds i32, i32* %ptr, i64 6
%ptr4 = getelementptr inbounds i32, i32* %ptr, i64 9
%val0 = load i32, i32* %ptr0
%val1 = load i32, i32* %ptr1
%val4 = load i32, i32* %ptr4
%res0 = insertelement <8 x i32> undef, i32 %val0, i32 0
%res1 = insertelement <8 x i32> %res0, i32 %val1, i32 1
%res2 = insertelement <8 x i32> %res1, i32 0, i32 2
%res3 = insertelement <8 x i32> %res2, i32 0, i32 3
%res4 = insertelement <8 x i32> %res3, i32 %val4, i32 4
%res6 = insertelement <8 x i32> %res4, i32 0, i32 6
%res7 = insertelement <8 x i32> %res6, i32 0, i32 7
ret <8 x i32> %res7
}
define <8 x i32> @merge_8i32_i32_1u3u5zu8(i32* %ptr) nounwind uwtable noinline ssp {
[X86] Add VBLENDPS/VPBLENDD to the execution domain fixing tables. llvm-svn: 312449
2017-09-04 01:52:23 +08:00
; AVX-LABEL: merge_8i32_i32_1u3u5zu8:
[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] Add VBLENDPS/VPBLENDD to the execution domain fixing tables. llvm-svn: 312449
2017-09-04 01:52:23 +08:00
; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
; AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7]
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_8i32_i32_1u3u5zu8:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86] SET0 to use XMM registers where possible PR26018 PR32862 Differential Revision: https://reviews.llvm.org/D35839 llvm-svn: 309298
2017-07-28 01:47:01 +08:00
; X32-AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0
[X86][AVX] Add shuffle masking support for EltsFromConsecutiveLoads Add support for the case where we have a consecutive load (which must include the first + last elements) with a mixture of undef/zero elements. We load the vector and then apply a shuffle to clear the zero'd elements. Differential Revision: http://reviews.llvm.org/D17297 llvm-svn: 261490
2016-02-22 03:15:48 +08:00
; X32-AVX-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3,4],ymm0[5],mem[6,7]
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i32, i32* %ptr, i64 1
%ptr2 = getelementptr inbounds i32, i32* %ptr, i64 3
%ptr4 = getelementptr inbounds i32, i32* %ptr, i64 5
%ptr7 = getelementptr inbounds i32, i32* %ptr, i64 8
%val0 = load i32, i32* %ptr0
%val2 = load i32, i32* %ptr2
%val4 = load i32, i32* %ptr4
%val7 = load i32, i32* %ptr7
%res0 = insertelement <8 x i32> undef, i32 %val0, i32 0
%res2 = insertelement <8 x i32> %res0, i32 %val2, i32 2
%res4 = insertelement <8 x i32> %res2, i32 %val4, i32 4
%res5 = insertelement <8 x i32> %res4, i32 0, i32 5
%res7 = insertelement <8 x i32> %res5, i32 %val7, i32 7
ret <8 x i32> %res7
}
define <16 x i16> @merge_16i16_i16_89zzzuuuuuuuuuuuz(i16* %ptr) nounwind uwtable noinline ssp {
[X86][SSE] Add general 32-bit LOAD + VZEXT_MOVL support to EltsFromConsecutiveLoads This patch adds support for consecutive (load/undef elements) 32-bit loads, followed by trailing undef/zero elements to be combined to a single MOVD load. Differential Revision: http://reviews.llvm.org/D16729 llvm-svn: 259796
2016-02-05 00:12:56 +08:00
; AVX-LABEL: merge_16i16_i16_89zzzuuuuuuuuuuuz:
[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] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
[X86][SSE] Add general 32-bit LOAD + VZEXT_MOVL support to EltsFromConsecutiveLoads This patch adds support for consecutive (load/undef elements) 32-bit loads, followed by trailing undef/zero elements to be combined to a single MOVD load. Differential Revision: http://reviews.llvm.org/D16729 llvm-svn: 259796
2016-02-05 00:12:56 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_16i16_i16_89zzzuuuuuuuuuuuz:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; X32-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i16, i16* %ptr, i64 8
%ptr1 = getelementptr inbounds i16, i16* %ptr, i64 9
%val0 = load i16, i16* %ptr0
%val1 = load i16, i16* %ptr1
%res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
%res1 = insertelement <16 x i16> %res0, i16 %val1, i16 1
%res2 = insertelement <16 x i16> %res1, i16 0, i16 2
%res3 = insertelement <16 x i16> %res2, i16 0, i16 3
%res4 = insertelement <16 x i16> %res3, i16 0, i16 4
%resF = insertelement <16 x i16> %res4, i16 0, i16 15
ret <16 x i16> %resF
}
define <16 x i16> @merge_16i16_i16_45u7uuuuuuuuuuuu(i16* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_16i16_i16_45u7uuuuuuuuuuuu:
[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] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_16i16_i16_45u7uuuuuuuuuuuu:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i16, i16* %ptr, i64 4
%ptr1 = getelementptr inbounds i16, i16* %ptr, i64 5
%ptr3 = getelementptr inbounds i16, i16* %ptr, i64 7
%val0 = load i16, i16* %ptr0
%val1 = load i16, i16* %ptr1
%val3 = load i16, i16* %ptr3
%res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
%res1 = insertelement <16 x i16> %res0, i16 %val1, i16 1
%res3 = insertelement <16 x i16> %res1, i16 %val3, i16 3
ret <16 x i16> %res3
}
define <16 x i16> @merge_16i16_i16_0uu3uuuuuuuuCuEF(i16* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_16i16_i16_0uu3uuuuuuuuCuEF:
[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][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
; AVX-NEXT: vmovups (%rdi), %ymm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_16i16_i16_0uu3uuuuuuuuCuEF:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vmovups (%eax), %ymm0
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0
%ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3
%ptrC = getelementptr inbounds i16, i16* %ptr, i64 12
%ptrE = getelementptr inbounds i16, i16* %ptr, i64 14
%ptrF = getelementptr inbounds i16, i16* %ptr, i64 15
%val0 = load i16, i16* %ptr0
%val3 = load i16, i16* %ptr3
%valC = load i16, i16* %ptrC
%valE = load i16, i16* %ptrE
%valF = load i16, i16* %ptrF
%res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
%res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3
%resC = insertelement <16 x i16> %res3, i16 %valC, i16 12
%resE = insertelement <16 x i16> %resC, i16 %valE, i16 14
%resF = insertelement <16 x i16> %resE, i16 %valF, i16 15
ret <16 x i16> %resF
}
define <16 x i16> @merge_16i16_i16_0uu3zzuuuuuzCuEF(i16* %ptr) nounwind uwtable noinline ssp {
[X86] Remove unnecessary bitcast from the front of AVX1Only 256-bit logical operation patterns. llvm-svn: 278088
2016-08-09 11:06:26 +08:00
; AVX-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF:
[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] Remove unnecessary bitcast from the front of AVX1Only 256-bit logical operation patterns. llvm-svn: 278088
2016-08-09 11:06:26 +08:00
; AVX-NEXT: vmovups (%rdi), %ymm0
; AVX-NEXT: vandps {{.*}}(%rip), %ymm0, %ymm0
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86] Remove unnecessary bitcast from the front of AVX1Only 256-bit logical operation patterns. llvm-svn: 278088
2016-08-09 11:06:26 +08:00
; X32-AVX-NEXT: vmovups (%eax), %ymm0
; X32-AVX-NEXT: vandps {{\.LCPI.*}}, %ymm0, %ymm0
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0
%ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3
%ptrC = getelementptr inbounds i16, i16* %ptr, i64 12
%ptrE = getelementptr inbounds i16, i16* %ptr, i64 14
%ptrF = getelementptr inbounds i16, i16* %ptr, i64 15
%val0 = load i16, i16* %ptr0
%val3 = load i16, i16* %ptr3
%valC = load i16, i16* %ptrC
%valE = load i16, i16* %ptrE
%valF = load i16, i16* %ptrF
%res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
%res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3
%res4 = insertelement <16 x i16> %res3, i16 0, i16 4
%res5 = insertelement <16 x i16> %res4, i16 0, i16 5
%resC = insertelement <16 x i16> %res5, i16 %valC, i16 12
%resD = insertelement <16 x i16> %resC, i16 0, i16 13
%resE = insertelement <16 x i16> %resD, i16 %valE, i16 14
%resF = insertelement <16 x i16> %resE, i16 %valF, i16 15
ret <16 x i16> %resF
}
define <32 x i8> @merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu(i8* %ptr) nounwind uwtable noinline ssp {
; AVX-LABEL: merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu:
[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] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_32i8_i8_45u7uuuuuuuuuuuuuuuuuuuuuuuuuuuu:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; X32-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%ptr0 = getelementptr inbounds i8, i8* %ptr, i64 4
%ptr1 = getelementptr inbounds i8, i8* %ptr, i64 5
%ptr3 = getelementptr inbounds i8, i8* %ptr, i64 7
%val0 = load i8, i8* %ptr0
%val1 = load i8, i8* %ptr1
%val3 = load i8, i8* %ptr3
%res0 = insertelement <32 x i8> undef, i8 %val0, i8 0
%res1 = insertelement <32 x i8> %res0, i8 %val1, i8 1
%res3 = insertelement <32 x i8> %res1, i8 %val3, i8 3
ret <32 x i8> %res3
}
define <32 x i8> @merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu(i8* %ptr) nounwind uwtable noinline ssp {
[X86][SSE] Add general 32-bit LOAD + VZEXT_MOVL support to EltsFromConsecutiveLoads This patch adds support for consecutive (load/undef elements) 32-bit loads, followed by trailing undef/zero elements to be combined to a single MOVD load. Differential Revision: http://reviews.llvm.org/D16729 llvm-svn: 259796
2016-02-05 00:12:56 +08:00
; AVX-LABEL: merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu:
[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] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
[X86][SSE] Add general 32-bit LOAD + VZEXT_MOVL support to EltsFromConsecutiveLoads This patch adds support for consecutive (load/undef elements) 32-bit loads, followed by trailing undef/zero elements to be combined to a single MOVD load. Differential Revision: http://reviews.llvm.org/D16729 llvm-svn: 259796
2016-02-05 00:12:56 +08:00
; AVX-NEXT: retq
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
;
; X32-AVX-LABEL: merge_32i8_i8_23u5uuuuuuuuuuzzzzuuuuuuuuuuuuuu:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
[X86][SSE] Fix domains for VZEXT_LOAD type instructions Add the missing domain equivalences for movss, movsd, movd and movq zero extending loading instructions. Differential Revision: https://reviews.llvm.org/D27684 llvm-svn: 289825
2016-12-16 00:05:29 +08:00
; X32-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
[X86][SSE] Added i686 target tests to make sure we are correctly loading consecutive entries as 64-bit integers llvm-svn: 259794
2016-02-04 23:51:55 +08:00
; X32-AVX-NEXT: retl
[X86][SSE] Add AVX512 merge consecutive load tests Add AVX512F/AVX512BW 512-bit tests. Add AVX512F tests to existing 128/256-bit tests. llvm-svn: 259410
2016-02-02 05:30:50 +08:00
%ptr0 = getelementptr inbounds i8, i8* %ptr, i64 2
%ptr1 = getelementptr inbounds i8, i8* %ptr, i64 3
%ptr3 = getelementptr inbounds i8, i8* %ptr, i64 5
[X86][AVX] Added more thorough 256-bit vector consecutive load tests. llvm-svn: 259100
2016-01-29 06:29:51 +08:00
%val0 = load i8, i8* %ptr0
%val1 = load i8, i8* %ptr1
%val3 = load i8, i8* %ptr3
%res0 = insertelement <32 x i8> undef, i8 %val0, i8 0
%res1 = insertelement <32 x i8> %res0, i8 %val1, i8 1
%res3 = insertelement <32 x i8> %res1, i8 %val3, i8 3
%resE = insertelement <32 x i8> %res3, i8 0, i8 14
%resF = insertelement <32 x i8> %resE, i8 0, i8 15
%resG = insertelement <32 x i8> %resF, i8 0, i8 16
%resH = insertelement <32 x i8> %resG, i8 0, i8 17
ret <32 x i8> %resH
}
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
;
; consecutive loads including any/all volatiles may not be combined
;
define <4 x double> @merge_4f64_f64_34uz_volatile(double* %ptr) nounwind uwtable noinline ssp {
[X86] Canonicalize (concat_vectors X, zero) -> (insert_subvector zero, X, 0). In a future patch, I plan to teach isel to use a small vector move with implicit zeroing of the upper elements when it sees the (insert_subvector zero, X, 0) pattern. llvm-svn: 312448
2017-09-04 01:52:19 +08:00
; AVX-LABEL: merge_4f64_f64_34uz_volatile:
[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] Canonicalize (concat_vectors X, zero) -> (insert_subvector zero, X, 0). In a future patch, I plan to teach isel to use a small vector move with implicit zeroing of the upper elements when it sees the (insert_subvector zero, X, 0) pattern. llvm-svn: 312448
2017-09-04 01:52:19 +08:00
; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; AVX-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
; AVX-NEXT: retq
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
;
; X32-AVX-LABEL: merge_4f64_f64_34uz_volatile:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
; X32-AVX-NEXT: vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
; X32-AVX-NEXT: retl
%ptr0 = getelementptr inbounds double, double* %ptr, i64 3
%ptr1 = getelementptr inbounds double, double* %ptr, i64 4
%val0 = load volatile double, double* %ptr0
%val1 = load volatile double, double* %ptr1
%res0 = insertelement <4 x double> undef, double %val0, i32 0
%res1 = insertelement <4 x double> %res0, double %val1, i32 1
%res3 = insertelement <4 x double> %res1, double 0.0, i32 3
ret <4 x double> %res3
}
define <16 x i16> @merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile(i16* %ptr) nounwind uwtable noinline ssp {
; AVX1-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile:
[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][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; AVX1-NEXT: vpxor %xmm0, %xmm0, %xmm0
; AVX1-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1
; AVX1-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0
; AVX1-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0
; AVX1-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled. Recommiting with compiler time improvements Recommitting after fixup of 32-bit aliasing sign offset bug in DAGCombiner. * Simplify Consecutive Merge Store Candidate Search Now that address aliasing is much less conservative, push through simplified store merging search and chain alias analysis which only checks for parallel stores through the chain subgraph. This is cleaner as the separation of non-interfering loads/stores from the store-merging logic. When merging stores search up the chain through a single load, and finds all possible stores by looking down from through a load and a TokenFactor to all stores visited. This improves the quality of the output SelectionDAG and the output Codegen (save perhaps for some ARM cases where we correctly constructs wider loads, but then promotes them to float operations which appear but requires more expensive constant generation). Some minor peephole optimizations to deal with improved SubDAG shapes (listed below) Additional Minor Changes: 1. Finishes removing unused AliasLoad code 2. Unifies the chain aggregation in the merged stores across code paths 3. Re-add the Store node to the worklist after calling SimplifyDemandedBits. 4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is arbitrary, but seems sufficient to not cause regressions in tests. 5. Remove Chain dependencies of Memory operations on CopyfromReg nodes as these are captured by data dependence 6. Forward loads-store values through tokenfactors containing {CopyToReg,CopyFromReg} Values. 7. Peephole to convert buildvector of extract_vector_elt to extract_subvector if possible (see CodeGen/AArch64/store-merge.ll) 8. Store merging for the ARM target is restricted to 32-bit as some in some contexts invalid 64-bit operations are being generated. This can be removed once appropriate checks are added. This finishes the change Matt Arsenault started in r246307 and jyknight's original patch. Many tests required some changes as memory operations are now reorderable, improving load-store forwarding. One test in particular is worth noting: CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store forwarding converts a load-store pair into a parallel store and a memory-realized bitcast of the same value. However, because we lose the sharing of the explicit and implicit store values we must create another local store. A similar transformation happens before SelectionDAG as well. Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle llvm-svn: 297695
2017-03-14 08:34:14 +08:00
; AVX1-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile:
[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][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; AVX2-NEXT: vpxor %xmm0, %xmm0, %xmm0
; AVX2-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1
; AVX2-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0
; AVX2-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0
; AVX2-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled. Recommiting with compiler time improvements Recommitting after fixup of 32-bit aliasing sign offset bug in DAGCombiner. * Simplify Consecutive Merge Store Candidate Search Now that address aliasing is much less conservative, push through simplified store merging search and chain alias analysis which only checks for parallel stores through the chain subgraph. This is cleaner as the separation of non-interfering loads/stores from the store-merging logic. When merging stores search up the chain through a single load, and finds all possible stores by looking down from through a load and a TokenFactor to all stores visited. This improves the quality of the output SelectionDAG and the output Codegen (save perhaps for some ARM cases where we correctly constructs wider loads, but then promotes them to float operations which appear but requires more expensive constant generation). Some minor peephole optimizations to deal with improved SubDAG shapes (listed below) Additional Minor Changes: 1. Finishes removing unused AliasLoad code 2. Unifies the chain aggregation in the merged stores across code paths 3. Re-add the Store node to the worklist after calling SimplifyDemandedBits. 4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is arbitrary, but seems sufficient to not cause regressions in tests. 5. Remove Chain dependencies of Memory operations on CopyfromReg nodes as these are captured by data dependence 6. Forward loads-store values through tokenfactors containing {CopyToReg,CopyFromReg} Values. 7. Peephole to convert buildvector of extract_vector_elt to extract_subvector if possible (see CodeGen/AArch64/store-merge.ll) 8. Store merging for the ARM target is restricted to 32-bit as some in some contexts invalid 64-bit operations are being generated. This can be removed once appropriate checks are added. This finishes the change Matt Arsenault started in r246307 and jyknight's original patch. Many tests required some changes as memory operations are now reorderable, improving load-store forwarding. One test in particular is worth noting: CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store forwarding converts a load-store pair into a parallel store and a memory-realized bitcast of the same value. However, because we lose the sharing of the explicit and implicit store values we must create another local store. A similar transformation happens before SelectionDAG as well. Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle llvm-svn: 297695
2017-03-14 08:34:14 +08:00
; AVX2-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
; AVX512F-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512F: # %bb.0:
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; AVX512F-NEXT: vpxor %xmm0, %xmm0, %xmm0
; AVX512F-NEXT: vpinsrw $0, (%rdi), %xmm0, %xmm1
; AVX512F-NEXT: vpinsrw $4, 24(%rdi), %xmm0, %xmm0
; AVX512F-NEXT: vpinsrw $6, 28(%rdi), %xmm0, %xmm0
; AVX512F-NEXT: vpinsrw $7, 30(%rdi), %xmm0, %xmm0
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled. Recommiting with compiler time improvements Recommitting after fixup of 32-bit aliasing sign offset bug in DAGCombiner. * Simplify Consecutive Merge Store Candidate Search Now that address aliasing is much less conservative, push through simplified store merging search and chain alias analysis which only checks for parallel stores through the chain subgraph. This is cleaner as the separation of non-interfering loads/stores from the store-merging logic. When merging stores search up the chain through a single load, and finds all possible stores by looking down from through a load and a TokenFactor to all stores visited. This improves the quality of the output SelectionDAG and the output Codegen (save perhaps for some ARM cases where we correctly constructs wider loads, but then promotes them to float operations which appear but requires more expensive constant generation). Some minor peephole optimizations to deal with improved SubDAG shapes (listed below) Additional Minor Changes: 1. Finishes removing unused AliasLoad code 2. Unifies the chain aggregation in the merged stores across code paths 3. Re-add the Store node to the worklist after calling SimplifyDemandedBits. 4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is arbitrary, but seems sufficient to not cause regressions in tests. 5. Remove Chain dependencies of Memory operations on CopyfromReg nodes as these are captured by data dependence 6. Forward loads-store values through tokenfactors containing {CopyToReg,CopyFromReg} Values. 7. Peephole to convert buildvector of extract_vector_elt to extract_subvector if possible (see CodeGen/AArch64/store-merge.ll) 8. Store merging for the ARM target is restricted to 32-bit as some in some contexts invalid 64-bit operations are being generated. This can be removed once appropriate checks are added. This finishes the change Matt Arsenault started in r246307 and jyknight's original patch. Many tests required some changes as memory operations are now reorderable, improving load-store forwarding. One test in particular is worth noting: CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store forwarding converts a load-store pair into a parallel store and a memory-realized bitcast of the same value. However, because we lose the sharing of the explicit and implicit store values we must create another local store. A similar transformation happens before SelectionDAG as well. Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle llvm-svn: 297695
2017-03-14 08:34:14 +08:00
; AVX512F-NEXT: vpinsrw $3, 6(%rdi), %xmm1, %xmm1
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; AVX512F-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0
; AVX512F-NEXT: retq
;
; X32-AVX-LABEL: merge_16i16_i16_0uu3zzuuuuuzCuEF_volatile:
[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
; X32-AVX: # %bb.0:
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: vpxor %xmm0, %xmm0, %xmm0
; X32-AVX-NEXT: vpinsrw $0, (%eax), %xmm0, %xmm1
; X32-AVX-NEXT: vpinsrw $4, 24(%eax), %xmm0, %xmm0
; X32-AVX-NEXT: vpinsrw $6, 28(%eax), %xmm0, %xmm0
; X32-AVX-NEXT: vpinsrw $7, 30(%eax), %xmm0, %xmm0
In visitSTORE, always use FindBetterChain, rather than only when UseAA is enabled. Recommiting with compiler time improvements Recommitting after fixup of 32-bit aliasing sign offset bug in DAGCombiner. * Simplify Consecutive Merge Store Candidate Search Now that address aliasing is much less conservative, push through simplified store merging search and chain alias analysis which only checks for parallel stores through the chain subgraph. This is cleaner as the separation of non-interfering loads/stores from the store-merging logic. When merging stores search up the chain through a single load, and finds all possible stores by looking down from through a load and a TokenFactor to all stores visited. This improves the quality of the output SelectionDAG and the output Codegen (save perhaps for some ARM cases where we correctly constructs wider loads, but then promotes them to float operations which appear but requires more expensive constant generation). Some minor peephole optimizations to deal with improved SubDAG shapes (listed below) Additional Minor Changes: 1. Finishes removing unused AliasLoad code 2. Unifies the chain aggregation in the merged stores across code paths 3. Re-add the Store node to the worklist after calling SimplifyDemandedBits. 4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is arbitrary, but seems sufficient to not cause regressions in tests. 5. Remove Chain dependencies of Memory operations on CopyfromReg nodes as these are captured by data dependence 6. Forward loads-store values through tokenfactors containing {CopyToReg,CopyFromReg} Values. 7. Peephole to convert buildvector of extract_vector_elt to extract_subvector if possible (see CodeGen/AArch64/store-merge.ll) 8. Store merging for the ARM target is restricted to 32-bit as some in some contexts invalid 64-bit operations are being generated. This can be removed once appropriate checks are added. This finishes the change Matt Arsenault started in r246307 and jyknight's original patch. Many tests required some changes as memory operations are now reorderable, improving load-store forwarding. One test in particular is worth noting: CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store forwarding converts a load-store pair into a parallel store and a memory-realized bitcast of the same value. However, because we lose the sharing of the explicit and implicit store values we must create another local store. A similar transformation happens before SelectionDAG as well. Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle llvm-svn: 297695
2017-03-14 08:34:14 +08:00
; X32-AVX-NEXT: vpinsrw $3, 6(%eax), %xmm1, %xmm1
[X86][SSE] Added tests to ensure that consecutive loads including any/all volatiles are not combined llvm-svn: 264225
2016-03-24 08:14:37 +08:00
; X32-AVX-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
; X32-AVX-NEXT: retl
%ptr0 = getelementptr inbounds i16, i16* %ptr, i64 0
%ptr3 = getelementptr inbounds i16, i16* %ptr, i64 3
%ptrC = getelementptr inbounds i16, i16* %ptr, i64 12
%ptrE = getelementptr inbounds i16, i16* %ptr, i64 14
%ptrF = getelementptr inbounds i16, i16* %ptr, i64 15
%val0 = load volatile i16, i16* %ptr0
%val3 = load i16, i16* %ptr3
%valC = load i16, i16* %ptrC
%valE = load i16, i16* %ptrE
%valF = load volatile i16, i16* %ptrF
%res0 = insertelement <16 x i16> undef, i16 %val0, i16 0
%res3 = insertelement <16 x i16> %res0, i16 %val3, i16 3
%res4 = insertelement <16 x i16> %res3, i16 0, i16 4
%res5 = insertelement <16 x i16> %res4, i16 0, i16 5
%resC = insertelement <16 x i16> %res5, i16 %valC, i16 12
%resD = insertelement <16 x i16> %resC, i16 0, i16 13
%resE = insertelement <16 x i16> %resD, i16 %valE, i16 14
%resF = insertelement <16 x i16> %resE, i16 %valF, i16 15
ret <16 x i16> %resF
}