diff options
Diffstat (limited to 'test/CodeGen/X86/GlobalISel/select-memop-v256.mir')
-rw-r--r-- | test/CodeGen/X86/GlobalISel/select-memop-v256.mir | 188 |
1 files changed, 188 insertions, 0 deletions
diff --git a/test/CodeGen/X86/GlobalISel/select-memop-v256.mir b/test/CodeGen/X86/GlobalISel/select-memop-v256.mir new file mode 100644 index 000000000000..b9a7e4a8cc4a --- /dev/null +++ b/test/CodeGen/X86/GlobalISel/select-memop-v256.mir @@ -0,0 +1,188 @@ +# RUN: llc -mtriple=x86_64-linux-gnu -mattr=+avx -global-isel -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=ALL --check-prefix=NO_AVX512VL --check-prefix=NO_AVX512F --check-prefix=AVX +# RUN: llc -mtriple=x86_64-linux-gnu -mattr=+avx512f -global-isel -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=ALL --check-prefix=NO_AVX512VL --check-prefix=AVX512ALL --check-prefix=AVX512F +# RUN: llc -mtriple=x86_64-linux-gnu -mattr=+avx512f -mattr=+avx512vl -global-isel -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=ALL --check-prefix=AVX512ALL --check-prefix=AVX512VL + + +--- | + define <8 x i32> @test_load_v8i32_noalign(<8 x i32>* %p1) { + %r = load <8 x i32>, <8 x i32>* %p1, align 1 + ret <8 x i32> %r + } + + define <8 x i32> @test_load_v8i32_align(<8 x i32>* %p1) { + %r = load <8 x i32>, <8 x i32>* %p1, align 32 + ret <8 x i32> %r + } + + define void @test_store_v8i32_noalign(<8 x i32> %val, <8 x i32>* %p1) { + store <8 x i32> %val, <8 x i32>* %p1, align 1 + ret void + } + + define void @test_store_v8i32_align(<8 x i32> %val, <8 x i32>* %p1) { + store <8 x i32> %val, <8 x i32>* %p1, align 32 + ret void + } + + +... +--- +name: test_load_v8i32_noalign +# ALL-LABEL: name: test_load_v8i32_noalign +alignment: 4 +legalized: true +regBankSelected: true +# NO_AVX512F: registers: +# NO_AVX512F-NEXT: - { id: 0, class: gr64 } +# NO_AVX512F-NEXT: - { id: 1, class: vr256 } +# +# AVX512ALL: registers: +# AVX512ALL-NEXT: - { id: 0, class: gr64 } +# AVX512ALL-NEXT: - { id: 1, class: vr256x } +registers: + - { id: 0, class: gpr } + - { id: 1, class: vecr } +# NO_AVX512F: %0 = COPY %rdi +# NO_AVX512F-NEXT: %1 = VMOVUPSYrm %0, 1, _, 0, _ :: (load 32 from %ir.p1, align 1) +# NO_AVX512F-NEXT: %ymm0 = COPY %1 +# NO_AVX512F-NEXT: RET 0, implicit %ymm0 +# +# AVX512F: %0 = COPY %rdi +# AVX512F-NEXT: %1 = VMOVUPSZ256rm_NOVLX %0, 1, _, 0, _ :: (load 32 from %ir.p1, align 1) +# AVX512F-NEXT: %ymm0 = COPY %1 +# AVX512F-NEXT: RET 0, implicit %ymm0 +# +# AVX512VL: %0 = COPY %rdi +# AVX512VL-NEXT: %1 = VMOVUPSZ256rm %0, 1, _, 0, _ :: (load 32 from %ir.p1, align 1) +# AVX512VL-NEXT: %ymm0 = COPY %1 +# AVX512VL-NEXT: RET 0, implicit %ymm0 +body: | + bb.1 (%ir-block.0): + liveins: %rdi + + %0(p0) = COPY %rdi + %1(<8 x s32>) = G_LOAD %0(p0) :: (load 32 from %ir.p1, align 1) + %ymm0 = COPY %1(<8 x s32>) + RET 0, implicit %ymm0 + +... +--- +name: test_load_v8i32_align +# ALL-LABEL: name: test_load_v8i32_align +alignment: 4 +legalized: true +regBankSelected: true +# NO_AVX512F: registers: +# NO_AVX512F-NEXT: - { id: 0, class: gr64 } +# NO_AVX512F-NEXT: - { id: 1, class: vr256 } +# +# AVX512ALL: registers: +# AVX512ALL-NEXT: - { id: 0, class: gr64 } +# AVX512ALL-NEXT: - { id: 1, class: vr256x } +registers: + - { id: 0, class: gpr } + - { id: 1, class: vecr } +# NO_AVX512F: %0 = COPY %rdi +# NO_AVX512F-NEXT: %1 = VMOVAPSYrm %0, 1, _, 0, _ :: (load 32 from %ir.p1) +# NO_AVX512F-NEXT: %ymm0 = COPY %1 +# NO_AVX512F-NEXT: RET 0, implicit %ymm0 +# +# AVX512F: %0 = COPY %rdi +# AVX512F-NEXT: %1 = VMOVAPSZ256rm_NOVLX %0, 1, _, 0, _ :: (load 32 from %ir.p1) +# AVX512F-NEXT: %ymm0 = COPY %1 +# AVX512F-NEXT: RET 0, implicit %ymm0 +# +# AVX512VL: %0 = COPY %rdi +# AVX512VL-NEXT: %1 = VMOVAPSZ256rm %0, 1, _, 0, _ :: (load 32 from %ir.p1) +# AVX512VL-NEXT: %ymm0 = COPY %1 +# AVX512VL-NEXT: RET 0, implicit %ymm0 +body: | + bb.1 (%ir-block.0): + liveins: %rdi + + %0(p0) = COPY %rdi + %1(<8 x s32>) = G_LOAD %0(p0) :: (load 32 from %ir.p1) + %ymm0 = COPY %1(<8 x s32>) + RET 0, implicit %ymm0 + +... +--- +name: test_store_v8i32_noalign +# ALL-LABEL: name: test_store_v8i32_noalign +alignment: 4 +legalized: true +regBankSelected: true +# NO_AVX512F: registers: +# NO_AVX512F-NEXT: - { id: 0, class: vr256 } +# NO_AVX512F-NEXT: - { id: 1, class: gr64 } +# +# AVX512ALL: registers: +# AVX512ALL-NEXT: - { id: 0, class: vr256x } +# AVX512ALL-NEXT: - { id: 1, class: gr64 } +registers: + - { id: 0, class: vecr } + - { id: 1, class: gpr } +# NO_AVX512F: %0 = COPY %ymm0 +# NO_AVX512F-NEXT: %1 = COPY %rdi +# NO_AVX512F-NEXT: VMOVUPSYmr %1, 1, _, 0, _, %0 :: (store 32 into %ir.p1, align 1) +# NO_AVX512F-NEXT: RET 0 +# +# AVX512F: %0 = COPY %ymm0 +# AVX512F-NEXT: %1 = COPY %rdi +# AVX512F-NEXT: VMOVUPSZ256mr_NOVLX %1, 1, _, 0, _, %0 :: (store 32 into %ir.p1, align 1) +# AVX512F-NEXT: RET 0 +# +# AVX512VL: %0 = COPY %ymm0 +# AVX512VL-NEXT: %1 = COPY %rdi +# AVX512VL-NEXT: VMOVUPSZ256mr %1, 1, _, 0, _, %0 :: (store 32 into %ir.p1, align 1) +# AVX512VL-NEXT: RET 0 +body: | + bb.1 (%ir-block.0): + liveins: %rdi, %ymm0 + + %0(<8 x s32>) = COPY %ymm0 + %1(p0) = COPY %rdi + G_STORE %0(<8 x s32>), %1(p0) :: (store 32 into %ir.p1, align 1) + RET 0 + +... +--- +name: test_store_v8i32_align +# ALL-LABEL: name: test_store_v8i32_align +alignment: 4 +legalized: true +regBankSelected: true +# NO_AVX512F: registers: +# NO_AVX512F-NEXT: - { id: 0, class: vr256 } +# NO_AVX512F-NEXT: - { id: 1, class: gr64 } +# +# AVX512ALL: registers: +# AVX512ALL-NEXT: - { id: 0, class: vr256x } +# AVX512ALL-NEXT: - { id: 1, class: gr64 } +registers: + - { id: 0, class: vecr } + - { id: 1, class: gpr } +# NO_AVX512F: %0 = COPY %ymm0 +# NO_AVX512F-NEXT: %1 = COPY %rdi +# NO_AVX512F-NEXT: VMOVAPSYmr %1, 1, _, 0, _, %0 :: (store 32 into %ir.p1) +# NO_AVX512F-NEXT: RET 0 +# +# AVX512F: %0 = COPY %ymm0 +# AVX512F-NEXT: %1 = COPY %rdi +# AVX512F-NEXT: VMOVAPSZ256mr_NOVLX %1, 1, _, 0, _, %0 :: (store 32 into %ir.p1) +# AVX512F-NEXT: RET 0 +# +# AVX512VL: %0 = COPY %ymm0 +# AVX512VL-NEXT: %1 = COPY %rdi +# AVX512VL-NEXT: VMOVAPSZ256mr %1, 1, _, 0, _, %0 :: (store 32 into %ir.p1) +# AVX512VL-NEXT: RET 0 +body: | + bb.1 (%ir-block.0): + liveins: %rdi, %ymm0 + + %0(<8 x s32>) = COPY %ymm0 + %1(p0) = COPY %rdi + G_STORE %0(<8 x s32>), %1(p0) :: (store 32 into %ir.p1) + RET 0 + +... |