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/avx512-shuffles/broadcast-vector-int.ll

1431 lines
71 KiB
LLVM
Raw Normal View History

[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
[X86] Replace -mcpu option with -mattr in LIT tests added in https://reviews.llvm.org/rL312442 llvm-svn: 312474
2017-09-04 17:31:32 +08:00
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512dq %s -o - | FileCheck %s
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; FIXME: fixing PR34394 should fix the i32x2 memory cases resulting in a simple vbroadcasti32x2 instruction.
define <4 x i32> @test_2xi32_to_4xi32(<4 x i32> %vec) {
; CHECK-LABEL: test_2xi32_to_4xi32:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vpbroadcastq %xmm0, %xmm0
; CHECK-NEXT: retq
%res = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_2xi32_to_4xi32_mask0(<4 x i32> %vec, <4 x i32> %default, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %xmm3, %xmm2, %k1
[X86] Add 128-bit version of vbroadcasti32x2 to shuffle comment decoding. llvm-svn: 315395
2017-10-11 08:11:53 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm1 {%k1} = xmm0[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vmovdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mask0(<4 x i32> %vec, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm1, %k1
[X86] Add 128-bit version of vbroadcasti32x2 to shuffle comment decoding. llvm-svn: 315395
2017-10-11 08:11:53 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = xmm0[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_2xi32_to_4xi32_mask1(<4 x i32> %vec, <4 x i32> %default, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %xmm3, %xmm2, %k1
[X86] Add 128-bit version of vbroadcasti32x2 to shuffle comment decoding. llvm-svn: 315395
2017-10-11 08:11:53 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm1 {%k1} = xmm0[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vmovdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mask1(<4 x i32> %vec, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm1, %k1
[X86] Add 128-bit version of vbroadcasti32x2 to shuffle comment decoding. llvm-svn: 315395
2017-10-11 08:11:53 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = xmm0[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_2xi32_to_4xi32_mask2(<4 x i32> %vec, <4 x i32> %default, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %xmm3, %xmm2, %k1
[X86] Add 128-bit version of vbroadcasti32x2 to shuffle comment decoding. llvm-svn: 315395
2017-10-11 08:11:53 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm1 {%k1} = xmm0[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vmovdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mask2(<4 x i32> %vec, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm1, %k1
[X86] Add 128-bit version of vbroadcasti32x2 to shuffle comment decoding. llvm-svn: 315395
2017-10-11 08:11:53 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = xmm0[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_2xi32_to_4xi32_mask3(<4 x i32> %vec, <4 x i32> %default, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %xmm3, %xmm2, %k1
[X86] Add 128-bit version of vbroadcasti32x2 to shuffle comment decoding. llvm-svn: 315395
2017-10-11 08:11:53 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm1 {%k1} = xmm0[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vmovdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mask3(<4 x i32> %vec, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm1, %k1
[X86] Add 128-bit version of vbroadcasti32x2 to shuffle comment decoding. llvm-svn: 315395
2017-10-11 08:11:53 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = xmm0[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
define <8 x i32> @test_2xi32_to_8xi32(<8 x i32> %vec) {
; CHECK-LABEL: test_2xi32_to_8xi32:
[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
; CHECK: # %bb.0:
[X86] Remove patterns that select unmasked vbroadcastf2x32/vbroadcasti2x32. Prefer vbroadcastsd/vpbroadcastq instead. There's no advantage to using these instructions when they aren't masked. This enables some additional execution domain switching without needing to update the table. llvm-svn: 315674
2017-10-13 14:07:10 +08:00
; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%res = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_2xi32_to_8xi32_mask0(<8 x i32> %vec, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %ymm3, %ymm2, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm1 {%k1} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mask0(<8 x i32> %vec, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_2xi32_to_8xi32_mask1(<8 x i32> %vec, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %ymm3, %ymm2, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm1 {%k1} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mask1(<8 x i32> %vec, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_2xi32_to_8xi32_mask2(<8 x i32> %vec, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %ymm3, %ymm2, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm1 {%k1} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mask2(<8 x i32> %vec, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_2xi32_to_8xi32_mask3(<8 x i32> %vec, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %ymm3, %ymm2, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm1 {%k1} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mask3(<8 x i32> %vec, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
define <16 x i32> @test_2xi32_to_16xi32(<16 x i32> %vec) {
; CHECK-LABEL: test_2xi32_to_16xi32:
[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
; CHECK: # %bb.0:
[X86] Remove patterns that select unmasked vbroadcastf2x32/vbroadcasti2x32. Prefer vbroadcastsd/vpbroadcastq instead. There's no advantage to using these instructions when they aren't masked. This enables some additional execution domain switching without needing to update the table. llvm-svn: 315674
2017-10-13 14:07:10 +08:00
; CHECK-NEXT: vbroadcastsd %xmm0, %zmm0
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%res = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_2xi32_to_16xi32_mask0(<16 x i32> %vec, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %zmm3, %zmm2, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm1 {%k1} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mask0(<16 x i32> %vec, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_2xi32_to_16xi32_mask1(<16 x i32> %vec, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %zmm3, %zmm2, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm1 {%k1} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mask1(<16 x i32> %vec, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_2xi32_to_16xi32_mask2(<16 x i32> %vec, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %zmm3, %zmm2, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm1 {%k1} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mask2(<16 x i32> %vec, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_2xi32_to_16xi32_mask3(<16 x i32> %vec, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %zmm3, %zmm2, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm1 {%k1} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mask3(<16 x i32> %vec, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
define <4 x i32> @test_2xi32_to_4xi32_mem(<2 x i32>* %vp) {
; CHECK-LABEL: test_2xi32_to_4xi32_mem:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpbroadcastq (%rdi), %xmm0
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%res = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_2xi32_to_4xi32_mem_mask0(<2 x i32>* %vp, <4 x i32> %default, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} = mem[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mem_mask0(<2 x i32>* %vp, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = mem[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_2xi32_to_4xi32_mem_mask1(<2 x i32>* %vp, <4 x i32> %default, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} = mem[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mem_mask1(<2 x i32>* %vp, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = mem[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_2xi32_to_4xi32_mem_mask2(<2 x i32>* %vp, <4 x i32> %default, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} = mem[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mem_mask2(<2 x i32>* %vp, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = mem[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_2xi32_to_4xi32_mem_mask3(<2 x i32>* %vp, <4 x i32> %default, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %xmm2, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} = mem[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mem_mask3(<2 x i32>* %vp, <4 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[AVX512] Don't mark EXTLOAD as legal with AVX512. Continue using custom lowering. Summary: This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes. There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8. Reviewers: RKSimon, zvi, delena Reviewed By: delena Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D38714 llvm-svn: 315860
2017-10-16 00:41:17 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = mem[0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i32> %res
}
define <8 x i32> @test_2xi32_to_8xi32_mem(<2 x i32>* %vp) {
; CHECK-LABEL: test_2xi32_to_8xi32_mem:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm0 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; CHECK-NEXT: vpbroadcastq %xmm0, %ymm0
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%res = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_2xi32_to_8xi32_mem_mask0(<2 x i32>* %vp, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm2 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %ymm3, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} = xmm2[0,1,0,1,0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask0(<2 x i32>* %vp, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm1[0,1,0,1,0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_2xi32_to_8xi32_mem_mask1(<2 x i32>* %vp, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm2 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %ymm3, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} = xmm2[0,1,0,1,0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask1(<2 x i32>* %vp, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm1[0,1,0,1,0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_2xi32_to_8xi32_mem_mask2(<2 x i32>* %vp, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm2 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %ymm3, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} = xmm2[0,1,0,1,0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask2(<2 x i32>* %vp, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm1[0,1,0,1,0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_2xi32_to_8xi32_mem_mask3(<2 x i32>* %vp, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm2 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %ymm3, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} = xmm2[0,1,0,1,0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask3(<2 x i32>* %vp, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm1[0,1,0,1,0,1,0,1]
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
define <16 x i32> @test_2xi32_to_16xi32_mem(<2 x i32>* %vp) {
; CHECK-LABEL: test_2xi32_to_16xi32_mem:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm0 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm1 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpermd %zmm0, %zmm1, %zmm0
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%res = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask0(<2 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm2 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm3 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpxor %xmm4, %xmm4, %xmm4
; CHECK-NEXT: vpcmpeqd %zmm4, %zmm1, %k1
; CHECK-NEXT: vpermd %zmm2, %zmm3, %zmm0 {%k1}
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask0(<2 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm2 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %zmm3, %zmm0, %k1
; CHECK-NEXT: vpermd %zmm1, %zmm2, %zmm0 {%k1} {z}
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask1(<2 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm2 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm3 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpxor %xmm4, %xmm4, %xmm4
; CHECK-NEXT: vpcmpeqd %zmm4, %zmm1, %k1
; CHECK-NEXT: vpermd %zmm2, %zmm3, %zmm0 {%k1}
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask1(<2 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm2 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %zmm3, %zmm0, %k1
; CHECK-NEXT: vpermd %zmm1, %zmm2, %zmm0 {%k1} {z}
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask2(<2 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm2 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm3 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpxor %xmm4, %xmm4, %xmm4
; CHECK-NEXT: vpcmpeqd %zmm4, %zmm1, %k1
; CHECK-NEXT: vpermd %zmm2, %zmm3, %zmm0 {%k1}
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask2(<2 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm2 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %zmm3, %zmm0, %k1
; CHECK-NEXT: vpermd %zmm1, %zmm2, %zmm0 {%k1} {z}
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask3(<2 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm2 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm3 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpxor %xmm4, %xmm4, %xmm4
; CHECK-NEXT: vpcmpeqd %zmm4, %zmm1, %k1
; CHECK-NEXT: vpermd %zmm2, %zmm3, %zmm0 {%k1}
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask3(<2 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpmovzxdq {{.*#+}} xmm1 = mem[0],zero,mem[1],zero
; CHECK-NEXT: vmovdqa32 {{.*#+}} zmm2 = [0,2,0,2,0,2,0,2,0,2,0,2,0,2,0,2]
; CHECK-NEXT: vpxor %xmm3, %xmm3, %xmm3
; CHECK-NEXT: vpcmpeqd %zmm3, %zmm0, %k1
; CHECK-NEXT: vpermd %zmm1, %zmm2, %zmm0 {%k1} {z}
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: retq
%vec = load <2 x i32>, <2 x i32>* %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
define <8 x i32> @test_4xi32_to_8xi32_mem(<4 x i32>* %vp) {
; CHECK-LABEL: test_4xi32_to_8xi32_mem:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%res = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_4xi32_to_8xi32_mem_mask0(<4 x i32>* %vp, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi32_to_8xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_4xi32_to_8xi32_mem_mask0(<4 x i32>* %vp, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi32_to_8xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %ymm1, %ymm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_4xi32_to_8xi32_mem_mask1(<4 x i32>* %vp, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi32_to_8xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_4xi32_to_8xi32_mem_mask1(<4 x i32>* %vp, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi32_to_8xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %ymm1, %ymm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_4xi32_to_8xi32_mem_mask2(<4 x i32>* %vp, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi32_to_8xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_4xi32_to_8xi32_mem_mask2(<4 x i32>* %vp, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi32_to_8xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %ymm1, %ymm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_4xi32_to_8xi32_mem_mask3(<4 x i32>* %vp, <8 x i32> %default, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi32_to_8xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i32> @test_masked_z_4xi32_to_8xi32_mem_mask3(<4 x i32>* %vp, <8 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi32_to_8xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %ymm1, %ymm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i32> %res
}
define <16 x i32> @test_4xi32_to_16xi32_mem(<4 x i32>* %vp) {
; CHECK-LABEL: test_4xi32_to_16xi32_mem:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%res = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_4xi32_to_16xi32_mem_mask0(<4 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi32_to_16xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_4xi32_to_16xi32_mem_mask0(<4 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi32_to_16xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_4xi32_to_16xi32_mem_mask1(<4 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi32_to_16xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_4xi32_to_16xi32_mem_mask1(<4 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi32_to_16xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_4xi32_to_16xi32_mem_mask2(<4 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi32_to_16xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_4xi32_to_16xi32_mem_mask2(<4 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi32_to_16xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_4xi32_to_16xi32_mem_mask3(<4 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi32_to_16xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_4xi32_to_16xi32_mem_mask3(<4 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi32_to_16xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, <4 x i32>* %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
define <4 x i64> @test_2xi64_to_4xi64_mem(<2 x i64>* %vp) {
; CHECK-LABEL: test_2xi64_to_4xi64_mem:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%res = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
ret <4 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i64> @test_masked_2xi64_to_4xi64_mem_mask0(<2 x i64>* %vp, <4 x i64> %default, <4 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi64_to_4xi64_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i64> @test_masked_z_2xi64_to_4xi64_mem_mask0(<2 x i64>* %vp, <4 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi64_to_4xi64_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %ymm1, %ymm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i64> @test_masked_2xi64_to_4xi64_mem_mask1(<2 x i64>* %vp, <4 x i64> %default, <4 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi64_to_4xi64_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i64> @test_masked_z_2xi64_to_4xi64_mem_mask1(<2 x i64>* %vp, <4 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi64_to_4xi64_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %ymm1, %ymm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i64> @test_masked_2xi64_to_4xi64_mem_mask2(<2 x i64>* %vp, <4 x i64> %default, <4 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi64_to_4xi64_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i64> @test_masked_z_2xi64_to_4xi64_mem_mask2(<2 x i64>* %vp, <4 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi64_to_4xi64_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %ymm1, %ymm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i64> @test_masked_2xi64_to_4xi64_mem_mask3(<2 x i64>* %vp, <4 x i64> %default, <4 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi64_to_4xi64_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %ymm2, %ymm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <4 x i64> @test_masked_z_2xi64_to_4xi64_mem_mask3(<2 x i64>* %vp, <4 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi64_to_4xi64_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %ymm1, %ymm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <4 x i64> %res
}
define <8 x i64> @test_2xi64_to_8xi64_mem(<2 x i64>* %vp) {
; CHECK-LABEL: test_2xi64_to_8xi64_mem:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%res = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_2xi64_to_8xi64_mem_mask0(<2 x i64>* %vp, <8 x i64> %default, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi64_to_8xi64_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_z_2xi64_to_8xi64_mem_mask0(<2 x i64>* %vp, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi64_to_8xi64_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_2xi64_to_8xi64_mem_mask1(<2 x i64>* %vp, <8 x i64> %default, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi64_to_8xi64_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_z_2xi64_to_8xi64_mem_mask1(<2 x i64>* %vp, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi64_to_8xi64_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_2xi64_to_8xi64_mem_mask2(<2 x i64>* %vp, <8 x i64> %default, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi64_to_8xi64_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_z_2xi64_to_8xi64_mem_mask2(<2 x i64>* %vp, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi64_to_8xi64_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_2xi64_to_8xi64_mem_mask3(<2 x i64>* %vp, <8 x i64> %default, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_2xi64_to_8xi64_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_z_2xi64_to_8xi64_mem_mask3(<2 x i64>* %vp, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_2xi64_to_8xi64_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, <2 x i64>* %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
define <16 x i32> @test_8xi32_to_16xi32_mem(<8 x i32>* %vp) {
; CHECK-LABEL: test_8xi32_to_16xi32_mem:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%res = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_8xi32_to_16xi32_mem_mask0(<8 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_8xi32_to_16xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_8xi32_to_16xi32_mem_mask0(<8 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_8xi32_to_16xi32_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_8xi32_to_16xi32_mem_mask1(<8 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_8xi32_to_16xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_8xi32_to_16xi32_mem_mask1(<8 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_8xi32_to_16xi32_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_8xi32_to_16xi32_mem_mask2(<8 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_8xi32_to_16xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_8xi32_to_16xi32_mem_mask2(<8 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_8xi32_to_16xi32_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_8xi32_to_16xi32_mem_mask3(<8 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_8xi32_to_16xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqd %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <16 x i32> @test_masked_z_8xi32_to_16xi32_mem_mask3(<8 x i32>* %vp, <16 x i32> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_8xi32_to_16xi32_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, <8 x i32>* %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <16 x i32> %res
}
define <8 x i64> @test_4xi64_to_8xi64_mem(<4 x i64>* %vp) {
; CHECK-LABEL: test_4xi64_to_8xi64_mem:
[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
; CHECK: # %bb.0:
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%res = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_4xi64_to_8xi64_mem_mask0(<4 x i64>* %vp, <8 x i64> %default, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi64_to_8xi64_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_z_4xi64_to_8xi64_mem_mask0(<4 x i64>* %vp, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi64_to_8xi64_mem_mask0:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_4xi64_to_8xi64_mem_mask1(<4 x i64>* %vp, <8 x i64> %default, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi64_to_8xi64_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_z_4xi64_to_8xi64_mem_mask1(<4 x i64>* %vp, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi64_to_8xi64_mem_mask1:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_4xi64_to_8xi64_mem_mask2(<4 x i64>* %vp, <8 x i64> %default, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi64_to_8xi64_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_z_4xi64_to_8xi64_mem_mask2(<4 x i64>* %vp, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi64_to_8xi64_mem_mask2:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_4xi64_to_8xi64_mem_mask3(<4 x i64>* %vp, <8 x i64> %default, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_4xi64_to_8xi64_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm2, %xmm2, %xmm2
; CHECK-NEXT: vpcmpeqq %zmm2, %zmm1, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
define <8 x i64> @test_masked_z_4xi64_to_8xi64_mem_mask3(<4 x i64>* %vp, <8 x i64> %mask) {
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-LABEL: test_masked_z_4xi64_to_8xi64_mem_mask3:
[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
; CHECK: # %bb.0:
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
; CHECK-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, <4 x i64>* %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
[X86] Don't use constant condition for select instruction when testing masking ops. We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects. llvm-svn: 315848
2017-10-15 14:05:50 +08:00
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
[X86][AVX512] Add simple tests for all AVX512 shuffle instructions. Throughout an effort to strongly check the behavior of CodeGen with the IR shufflevector instruction we generated many tests while predicting the best X86 sequence that may be generated. This is a subset of the generated tests that we think may add value to our X86 set of tests. Some of the checks are not optimal and will be changed after fixing: 1. PR34394 2. PR34382 3. PR34380 4. PR34359 Differential Revision: https://reviews.llvm.org/D37329 llvm-svn: 312442
2017-09-03 21:53:44 +08:00
ret <8 x i64> %res
}