# RUN: llc -mtriple=x86_64-linux-gnu -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=SSE # 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 i8 @test_load_i8(i8* %p1) { %r = load i8, i8* %p1 ret i8 %r } define i16 @test_load_i16(i16* %p1) { %r = load i16, i16* %p1 ret i16 %r } define i32 @test_load_i32(i32* %p1) { %r = load i32, i32* %p1 ret i32 %r } define i64 @test_load_i64(i64* %p1) { %r = load i64, i64* %p1 ret i64 %r } define float @test_load_float(float* %p1) { %r = load float, float* %p1 ret float %r } define float @test_load_float_vecreg(float* %p1) { %r = load float, float* %p1 ret float %r } define double @test_load_double(double* %p1) { %r = load double, double* %p1 ret double %r } define double @test_load_double_vecreg(double* %p1) { %r = load double, double* %p1 ret double %r } define i32* @test_store_i32(i32 %val, i32* %p1) { store i32 %val, i32* %p1 ret i32* %p1 } define i64* @test_store_i64(i64 %val, i64* %p1) { store i64 %val, i64* %p1 ret i64* %p1 } define float* @test_store_float(float %val, float* %p1) { store float %val, float* %p1 ret float* %p1 } define float* @test_store_float_vec(float %val, float* %p1) { store float %val, float* %p1 ret float* %p1 } define double* @test_store_double(double %val, double* %p1) { store double %val, double* %p1 ret double* %p1 } define double* @test_store_double_vec(double %val, double* %p1) { store double %val, double* %p1 ret double* %p1 } define i32* @test_load_ptr(i32** %ptr1) { %p = load i32*, i32** %ptr1 ret i32* %p } define void @test_store_ptr(i32** %ptr1, i32* %a) { store i32* %a, i32** %ptr1 ret void } define i32 @test_gep_folding(i32* %arr, i32 %val) { %arrayidx = getelementptr i32, i32* %arr, i32 5 store i32 %val, i32* %arrayidx %r = load i32, i32* %arrayidx ret i32 %r } define i32 @test_gep_folding_largeGepIndex(i32* %arr, i32 %val) #0 { %arrayidx = getelementptr i32, i32* %arr, i64 57179869180 store i32 %val, i32* %arrayidx %r = load i32, i32* %arrayidx ret i32 %r } ... --- # ALL-LABEL: name: test_load_i8 name: test_load_i8 alignment: 4 legalized: true regBankSelected: true registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr8, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL: %1:gr8 = MOV8rm %0, 1, %noreg, 0, %noreg :: (load 1 from %ir.p1) # ALL: %al = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(s8) = G_LOAD %0(p0) :: (load 1 from %ir.p1) %al = COPY %1(s8) RET 0, implicit %al ... --- # ALL-LABEL: name: test_load_i16 name: test_load_i16 alignment: 4 legalized: true regBankSelected: true registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr16, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL: %1:gr16 = MOV16rm %0, 1, %noreg, 0, %noreg :: (load 2 from %ir.p1) # ALL: %ax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(s16) = G_LOAD %0(p0) :: (load 2 from %ir.p1) %ax = COPY %1(s16) RET 0, implicit %ax ... --- # ALL-LABEL: name: test_load_i32 name: test_load_i32 alignment: 4 legalized: true regBankSelected: true registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr32, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL: %1:gr32 = MOV32rm %0, 1, %noreg, 0, %noreg :: (load 4 from %ir.p1) # ALL: %eax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(s32) = G_LOAD %0(p0) :: (load 4 from %ir.p1) %eax = COPY %1(s32) RET 0, implicit %eax ... --- # ALL-LABEL: name: test_load_i64 name: test_load_i64 alignment: 4 legalized: true regBankSelected: true registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr64, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL: %1:gr64 = MOV64rm %0, 1, %noreg, 0, %noreg :: (load 8 from %ir.p1) # ALL: %rax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(s64) = G_LOAD %0(p0) :: (load 8 from %ir.p1) %rax = COPY %1(s64) RET 0, implicit %rax ... --- # ALL-LABEL: name: test_load_float name: test_load_float alignment: 4 legalized: true regBankSelected: true registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr32, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL: %1:gr32 = MOV32rm %0, 1, %noreg, 0, %noreg :: (load 4 from %ir.p1) # ALL: %xmm0 = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(s32) = G_LOAD %0(p0) :: (load 4 from %ir.p1) %xmm0 = COPY %1(s32) RET 0, implicit %xmm0 ... --- # ALL-LABEL: name: test_load_float_vecreg name: test_load_float_vecreg alignment: 4 legalized: true regBankSelected: true registers: - { id: 0, class: gpr } - { id: 1, class: vecr } # ALL: %0:gr64 = COPY %rdi # SSE: %1:fr32 = MOVSSrm %0, 1, %noreg, 0, %noreg :: (load 4 from %ir.p1) # AVX: %1:fr32 = VMOVSSrm %0, 1, %noreg, 0, %noreg :: (load 4 from %ir.p1) # AVX512ALL: %1:fr32x = VMOVSSZrm %0, 1, %noreg, 0, %noreg :: (load 4 from %ir.p1) # ALL: %xmm0 = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(s32) = G_LOAD %0(p0) :: (load 4 from %ir.p1) %xmm0 = COPY %1(s32) RET 0, implicit %xmm0 ... --- # ALL-LABEL: name: test_load_double name: test_load_double alignment: 4 legalized: true regBankSelected: true registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr64, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL: %1:gr64 = MOV64rm %0, 1, %noreg, 0, %noreg :: (load 8 from %ir.p1) # ALL: %xmm0 = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(s64) = G_LOAD %0(p0) :: (load 8 from %ir.p1) %xmm0 = COPY %1(s64) RET 0, implicit %xmm0 ... --- # ALL-LABEL: name: test_load_double_vecreg name: test_load_double_vecreg alignment: 4 legalized: true regBankSelected: true registers: - { id: 0, class: gpr } - { id: 1, class: vecr } # ALL: %0:gr64 = COPY %rdi # SSE: %1:fr64 = MOVSDrm %0, 1, %noreg, 0, %noreg :: (load 8 from %ir.p1) # AVX: %1:fr64 = VMOVSDrm %0, 1, %noreg, 0, %noreg :: (load 8 from %ir.p1) # AVX512ALL: %1:fr64x = VMOVSDZrm %0, 1, %noreg, 0, %noreg :: (load 8 from %ir.p1) # ALL: %xmm0 = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(s64) = G_LOAD %0(p0) :: (load 8 from %ir.p1) %xmm0 = COPY %1(s64) RET 0, implicit %xmm0 ... --- # ALL-LABEL: name: test_store_i32 name: test_store_i32 alignment: 4 legalized: true regBankSelected: true registers: # ALL: - { id: 0, class: gr32, preferred-register: '' } # ALL: - { id: 1, class: gr64, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %0:gr32 = COPY %edi # ALL: %1:gr64 = COPY %rsi # ALL: MOV32mr %1, 1, %noreg, 0, %noreg, %0 :: (store 4 into %ir.p1) # ALL: %rax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %edi, %rsi %0(s32) = COPY %edi %1(p0) = COPY %rsi G_STORE %0(s32), %1(p0) :: (store 4 into %ir.p1) %rax = COPY %1(p0) RET 0, implicit %rax ... --- # ALL-LABEL: name: test_store_i64 name: test_store_i64 alignment: 4 legalized: true regBankSelected: true registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr64, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL: %1:gr64 = COPY %rsi # ALL: MOV64mr %1, 1, %noreg, 0, %noreg, %0 :: (store 8 into %ir.p1) # ALL: %rax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi, %rsi %0(s64) = COPY %rdi %1(p0) = COPY %rsi G_STORE %0(s64), %1(p0) :: (store 8 into %ir.p1) %rax = COPY %1(p0) RET 0, implicit %rax ... --- # ALL-LABEL: name: test_store_float name: test_store_float alignment: 4 legalized: true regBankSelected: true registers: - { id: 0, class: vecr } - { id: 1, class: gpr } - { id: 2, class: gpr } # NO_AVX512F: %0:fr32 = COPY %xmm0 # AVX512ALL: %0:fr32x = COPY %xmm0 # ALL: %1:gr64 = COPY %rdi # ALL: %2:gr32 = COPY %0 # ALL: MOV32mr %1, 1, %noreg, 0, %noreg, %2 :: (store 4 into %ir.p1) # ALL: %rax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi, %xmm0 %0(s32) = COPY %xmm0 %1(p0) = COPY %rdi %2(s32) = COPY %0(s32) G_STORE %2(s32), %1(p0) :: (store 4 into %ir.p1) %rax = COPY %1(p0) RET 0, implicit %rax ... --- # ALL-LABEL: name: test_store_float_vec name: test_store_float_vec alignment: 4 legalized: true regBankSelected: true registers: - { id: 0, class: vecr } - { id: 1, class: gpr } # NO_AVX512F: %0:fr32 = COPY %xmm0 # AVX512ALL: %0:fr32x = COPY %xmm0 # ALL: %1:gr64 = COPY %rdi # SSE: MOVSSmr %1, 1, %noreg, 0, %noreg, %0 :: (store 4 into %ir.p1) # AVX: VMOVSSmr %1, 1, %noreg, 0, %noreg, %0 :: (store 4 into %ir.p1) # AVX512ALL: VMOVSSZmr %1, 1, %noreg, 0, %noreg, %0 :: (store 4 into %ir.p1) # ALL: %rax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi, %xmm0 %0(s32) = COPY %xmm0 %1(p0) = COPY %rdi G_STORE %0(s32), %1(p0) :: (store 4 into %ir.p1) %rax = COPY %1(p0) RET 0, implicit %rax ... --- # ALL-LABEL: name: test_store_double name: test_store_double alignment: 4 legalized: true regBankSelected: true registers: - { id: 0, class: vecr } - { id: 1, class: gpr } - { id: 2, class: gpr } # NO_AVX512X: %0:fr64 = COPY %xmm0 # AVX512ALL: %0:fr64x = COPY %xmm0 # ALL: %1:gr64 = COPY %rdi # ALL: %2:gr64 = COPY %0 # ALL: MOV64mr %1, 1, %noreg, 0, %noreg, %2 :: (store 8 into %ir.p1) # ALL: %rax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi, %xmm0 %0(s64) = COPY %xmm0 %1(p0) = COPY %rdi %2(s64) = COPY %0(s64) G_STORE %2(s64), %1(p0) :: (store 8 into %ir.p1) %rax = COPY %1(p0) RET 0, implicit %rax ... --- # ALL-LABEL: name: test_store_double_vec name: test_store_double_vec alignment: 4 legalized: true regBankSelected: true registers: - { id: 0, class: vecr } - { id: 1, class: gpr } # NO_AVX512F: %0:fr64 = COPY %xmm0 # AVX512ALL: %0:fr64x = COPY %xmm0 # ALL: %1:gr64 = COPY %rdi # SSE: MOVSDmr %1, 1, %noreg, 0, %noreg, %0 :: (store 8 into %ir.p1) # AVX: VMOVSDmr %1, 1, %noreg, 0, %noreg, %0 :: (store 8 into %ir.p1) # AVX512ALL: VMOVSDZmr %1, 1, %noreg, 0, %noreg, %0 :: (store 8 into %ir.p1) # ALL: %rax = COPY %1 body: | bb.1 (%ir-block.0): liveins: %rdi, %xmm0 %0(s64) = COPY %xmm0 %1(p0) = COPY %rdi G_STORE %0(s64), %1(p0) :: (store 8 into %ir.p1) %rax = COPY %1(p0) RET 0, implicit %rax ... --- # ALL-LABEL: name: test_load_ptr name: test_load_ptr alignment: 4 legalized: true regBankSelected: true selected: false registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr64, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: %1:gr64 = MOV64rm %0, 1, %noreg, 0, %noreg :: (load 8 from %ir.ptr1) body: | bb.1 (%ir-block.0): liveins: %rdi %0(p0) = COPY %rdi %1(p0) = G_LOAD %0(p0) :: (load 8 from %ir.ptr1) %rax = COPY %1(p0) RET 0, implicit %rax ... --- # ALL-LABEL: name: test_store_ptr name: test_store_ptr alignment: 4 legalized: true regBankSelected: true selected: false registers: # ALL: - { id: 0, class: gr64, preferred-register: '' } # ALL: - { id: 1, class: gr64, preferred-register: '' } - { id: 0, class: gpr } - { id: 1, class: gpr } # ALL: MOV64mr %0, 1, %noreg, 0, %noreg, %1 :: (store 8 into %ir.ptr1) body: | bb.1 (%ir-block.0): liveins: %rdi, %rsi %0(p0) = COPY %rdi %1(p0) = COPY %rsi G_STORE %1(p0), %0(p0) :: (store 8 into %ir.ptr1) RET 0 ... --- name: test_gep_folding # ALL-LABEL: name: test_gep_folding alignment: 4 legalized: true regBankSelected: true # ALL: registers: # ALL-NEXT: - { id: 0, class: gr64, preferred-register: '' } # ALL-NEXT: - { id: 1, class: gr32, preferred-register: '' } # ALL-NEXT: - { id: 2, class: gpr, preferred-register: '' } # ALL-NEXT: - { id: 3, class: gpr, preferred-register: '' } # ALL-NEXT: - { id: 4, class: gr32, preferred-register: '' } registers: - { id: 0, class: gpr } - { id: 1, class: gpr } - { id: 2, class: gpr } - { id: 3, class: gpr } - { id: 4, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL-NEXT: %1:gr32 = COPY %esi # ALL-NEXT: MOV32mr %0, 1, %noreg, 20, %noreg, %1 :: (store 4 into %ir.arrayidx) # ALL-NEXT: %4:gr32 = MOV32rm %0, 1, %noreg, 20, %noreg :: (load 4 from %ir.arrayidx) # ALL-NEXT: %eax = COPY %4 # ALL-NEXT: RET 0, implicit %eax body: | bb.1 (%ir-block.0): liveins: %esi, %rdi %0(p0) = COPY %rdi %1(s32) = COPY %esi %2(s64) = G_CONSTANT i64 20 %3(p0) = G_GEP %0, %2(s64) G_STORE %1(s32), %3(p0) :: (store 4 into %ir.arrayidx) %4(s32) = G_LOAD %3(p0) :: (load 4 from %ir.arrayidx) %eax = COPY %4(s32) RET 0, implicit %eax ... --- name: test_gep_folding_largeGepIndex # ALL-LABEL: name: test_gep_folding_largeGepIndex alignment: 4 legalized: true regBankSelected: true # ALL: registers: # ALL-NEXT: - { id: 0, class: gr64, preferred-register: '' } # ALL-NEXT: - { id: 1, class: gr32, preferred-register: '' } # ALL-NEXT: - { id: 2, class: gr64_nosp, preferred-register: '' } # ALL-NEXT: - { id: 3, class: gr64, preferred-register: '' } # ALL-NEXT: - { id: 4, class: gr32, preferred-register: '' } registers: - { id: 0, class: gpr } - { id: 1, class: gpr } - { id: 2, class: gpr } - { id: 3, class: gpr } - { id: 4, class: gpr } # ALL: %0:gr64 = COPY %rdi # ALL-NEXT: %1:gr32 = COPY %esi # ALL-NEXT: %2:gr64_nosp = MOV64ri 228719476720 # ALL-NEXT: %3:gr64 = LEA64r %0, 1, %2, 0, %noreg # ALL-NEXT: MOV32mr %3, 1, %noreg, 0, %noreg, %1 :: (store 4 into %ir.arrayidx) # ALL-NEXT: %4:gr32 = MOV32rm %3, 1, %noreg, 0, %noreg :: (load 4 from %ir.arrayidx) # ALL-NEXT: %eax = COPY %4 # ALL-NEXT: RET 0, implicit %eax body: | bb.1 (%ir-block.0): liveins: %esi, %rdi %0(p0) = COPY %rdi %1(s32) = COPY %esi %2(s64) = G_CONSTANT i64 228719476720 %3(p0) = G_GEP %0, %2(s64) G_STORE %1(s32), %3(p0) :: (store 4 into %ir.arrayidx) %4(s32) = G_LOAD %3(p0) :: (load 4 from %ir.arrayidx) %eax = COPY %4(s32) RET 0, implicit %eax ...