--- /dev/null
+# 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 i64 @test_add_i64(i64 %arg1, i64 %arg2) {
+ %ret = add i64 %arg1, %arg2
+ ret i64 %ret
+ }
+
+ define i32 @test_add_i32(i32 %arg1, i32 %arg2) {
+ %ret = add i32 %arg1, %arg2
+ ret i32 %ret
+ }
+
+ define float @test_add_float(float %arg1, float %arg2) {
+ %ret = fadd float %arg1, %arg2
+ ret float %ret
+ }
+
+ define double @test_add_double(double %arg1, double %arg2) {
+ %ret = fadd double %arg1, %arg2
+ ret double %ret
+ }
+
+ define <4 x i32> @test_add_v4i32(<4 x i32> %arg1, <4 x i32> %arg2) {
+ %ret = add <4 x i32> %arg1, %arg2
+ ret <4 x i32> %ret
+ }
+
+ define <4 x float> @test_add_v4f32(<4 x float> %arg1, <4 x float> %arg2) {
+ %ret = fadd <4 x float> %arg1, %arg2
+ ret <4 x float> %ret
+ }
+...
+
+---
+name: test_add_i64
+legalized: true
+regBankSelected: true
+# ALL: registers:
+# ALL-NEXT: - { id: 0, class: gr64 }
+# ALL-NEXT: - { id: 1, class: gr64 }
+# ALL-NEXT: - { id: 2, class: gr64 }
+registers:
+ - { id: 0, class: gpr }
+ - { id: 1, class: gpr }
+ - { id: 2, class: gpr }
+# ALL: %0 = COPY %rdi
+# ALL-NEXT: %1 = COPY %rsi
+# ALL-NEXT: %2 = ADD64rr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %edi, %esi
+
+ %0(s64) = COPY %rdi
+ %1(s64) = COPY %rsi
+ %2(s64) = G_ADD %0, %1
+ %rax = COPY %2(s64)
+
+...
+
+---
+name: test_add_i32
+legalized: true
+regBankSelected: true
+# ALL: registers:
+# ALL-NEXT: - { id: 0, class: gr32 }
+# ALL-NEXT: - { id: 1, class: gr32 }
+# ALL-NEXT: - { id: 2, class: gr32 }
+registers:
+ - { id: 0, class: gpr }
+ - { id: 1, class: gpr }
+ - { id: 2, class: gpr }
+# ALL: %0 = COPY %edi
+# ALL-NEXT: %1 = COPY %esi
+# ALL-NEXT: %2 = ADD32rr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %edi, %esi
+
+ %0(s32) = COPY %edi
+ %1(s32) = COPY %esi
+ %2(s32) = G_ADD %0, %1
+ %rax = COPY %2(s32)
+
+...
+---
+name: test_add_float
+alignment: 4
+legalized: true
+regBankSelected: true
+selected: false
+tracksRegLiveness: true
+# ALL: registers:
+# NO_AVX512F-NEXT: - { id: 0, class: fr32 }
+# NO_AVX512F-NEXT: - { id: 1, class: fr32 }
+# NO_AVX512F-NEXT: - { id: 2, class: fr32 }
+# AVX512ALL-NEXT: - { id: 0, class: fr32x }
+# AVX512ALL-NEXT: - { id: 1, class: fr32x }
+# AVX512ALL-NEXT: - { id: 2, class: fr32x }
+registers:
+ - { id: 0, class: vecr }
+ - { id: 1, class: vecr }
+ - { id: 2, class: vecr }
+# ALL: %0 = COPY %xmm0
+# ALL-NEXT: %1 = COPY %xmm1
+# SSE-NEXT: %2 = ADDSSrr %0, %1
+# AVX-NEXT: %2 = VADDSSrr %0, %1
+# AVX512F-NEXT: %2 = VADDSSZrr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %xmm0, %xmm1
+
+ %0(s32) = COPY %xmm0
+ %1(s32) = COPY %xmm1
+ %2(s32) = G_FADD %0, %1
+ %xmm0 = COPY %2(s32)
+ RET 0, implicit %xmm0
+
+...
+---
+name: test_add_double
+alignment: 4
+legalized: true
+regBankSelected: true
+selected: false
+tracksRegLiveness: true
+# ALL: registers:
+# NO_AVX512F-NEXT: - { id: 0, class: fr64 }
+# NO_AVX512F-NEXT: - { id: 1, class: fr64 }
+# NO_AVX512F-NEXT: - { id: 2, class: fr64 }
+# AVX512ALL-NEXT: - { id: 0, class: fr64x }
+# AVX512ALL-NEXT: - { id: 1, class: fr64x }
+# AVX512ALL-NEXT: - { id: 2, class: fr64x }
+registers:
+ - { id: 0, class: vecr }
+ - { id: 1, class: vecr }
+ - { id: 2, class: vecr }
+# ALL: %0 = COPY %xmm0
+# ALL-NEXT: %1 = COPY %xmm1
+# SSE-NEXT: %2 = ADDSDrr %0, %1
+# AVX-NEXT: %2 = VADDSDrr %0, %1
+# AVX512F-NEXT: %2 = VADDSDZrr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %xmm0, %xmm1
+
+ %0(s64) = COPY %xmm0
+ %1(s64) = COPY %xmm1
+ %2(s64) = G_FADD %0, %1
+ %xmm0 = COPY %2(s64)
+ RET 0, implicit %xmm0
+
+...
+---
+name: test_add_v4i32
+alignment: 4
+legalized: true
+regBankSelected: true
+selected: false
+tracksRegLiveness: true
+# ALL: registers:
+# NO_AVX512VL-NEXT: - { id: 0, class: vr128 }
+# NO_AVX512VL-NEXT: - { id: 1, class: vr128 }
+# NO_AVX512VL-NEXT: - { id: 2, class: vr128 }
+# AVX512VL-NEXT: - { id: 0, class: vr128x }
+# AVX512VL-NEXT: - { id: 1, class: vr128x }
+# AVX512VL-NEXT: - { id: 2, class: vr128x }
+registers:
+ - { id: 0, class: vecr }
+ - { id: 1, class: vecr }
+ - { id: 2, class: vecr }
+# ALL: %0 = COPY %xmm0
+# ALL-NEXT: %1 = COPY %xmm1
+# SSE-NEXT: %2 = PADDDrr %0, %1
+# AVX-NEXT: %2 = VPADDDrr %0, %1
+# AVX512F-NEXT: %2 = VPADDDrr %0, %1
+# AVX512VL-NEXT: %2 = VPADDDZ128rr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %xmm0, %xmm1
+
+ %0(<4 x s32>) = COPY %xmm0
+ %1(<4 x s32>) = COPY %xmm1
+ %2(<4 x s32>) = G_ADD %0, %1
+ %xmm0 = COPY %2(<4 x s32>)
+ RET 0, implicit %xmm0
+
+...
+---
+name: test_add_v4f32
+alignment: 4
+legalized: true
+regBankSelected: true
+selected: false
+tracksRegLiveness: true
+# ALL: registers:
+# NO_AVX512VL-NEXT: - { id: 0, class: vr128 }
+# NO_AVX512VL-NEXT: - { id: 1, class: vr128 }
+# NO_AVX512VL-NEXT: - { id: 2, class: vr128 }
+# AVX512VL-NEXT: - { id: 0, class: vr128x }
+# AVX512VL-NEXT: - { id: 1, class: vr128x }
+# AVX512VL-NEXT: - { id: 2, class: vr128x }
+registers:
+ - { id: 0, class: vecr }
+ - { id: 1, class: vecr }
+ - { id: 2, class: vecr }
+# ALL: %0 = COPY %xmm0
+# ALL-NEXT: %1 = COPY %xmm1
+# SSE-NEXT: %2 = ADDPSrr %0, %1
+# AVX-NEXT: %2 = VADDPSrr %0, %1
+# AVX512F-NEXT: %2 = VADDPSrr %0, %1
+# AVX512VL-NEXT: %2 = VADDPSZ128rr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %xmm0, %xmm1
+
+ %0(<4 x s32>) = COPY %xmm0
+ %1(<4 x s32>) = COPY %xmm1
+ %2(<4 x s32>) = G_FADD %0, %1
+ %xmm0 = COPY %2(<4 x s32>)
+ RET 0, implicit %xmm0
+
+...
# 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 i64 @test_add_i64(i64 %arg1, i64 %arg2) {
- %ret = add i64 %arg1, %arg2
- ret i64 %ret
- }
-
- define i32 @test_add_i32(i32 %arg1, i32 %arg2) {
- %ret = add i32 %arg1, %arg2
- ret i32 %ret
- }
-
- define i64 @test_sub_i64(i64 %arg1, i64 %arg2) {
- %ret = sub i64 %arg1, %arg2
- ret i64 %ret
- }
-
- define i32 @test_sub_i32(i32 %arg1, i32 %arg2) {
- %ret = sub i32 %arg1, %arg2
- ret i32 %ret
- }
-
- define float @test_add_float(float %arg1, float %arg2) {
- %ret = fadd float %arg1, %arg2
- ret float %ret
- }
-
- define double @test_add_double(double %arg1, double %arg2) {
- %ret = fadd double %arg1, %arg2
- ret double %ret
- }
-
- define float @test_sub_float(float %arg1, float %arg2) {
- %ret = fsub float %arg1, %arg2
- ret float %ret
- }
-
- define double @test_sub_double(double %arg1, double %arg2) {
- %ret = fsub double %arg1, %arg2
- ret double %ret
- }
-
- define <4 x i32> @test_add_v4i32(<4 x i32> %arg1, <4 x i32> %arg2) {
- %ret = add <4 x i32> %arg1, %arg2
- ret <4 x i32> %ret
- }
-
- define <4 x i32> @test_sub_v4i32(<4 x i32> %arg1, <4 x i32> %arg2) {
- %ret = sub <4 x i32> %arg1, %arg2
- ret <4 x i32> %ret
- }
-
- define <4 x float> @test_add_v4f32(<4 x float> %arg1, <4 x float> %arg2) {
- %ret = fadd <4 x float> %arg1, %arg2
- ret <4 x float> %ret
- }
-
- define <4 x float> @test_sub_v4f32(<4 x float> %arg1, <4 x float> %arg2) {
- %ret = fsub <4 x float> %arg1, %arg2
- ret <4 x float> %ret
- }
-
- define i8 @test_load_i8(i8* %p1) {
+ define i8 @test_load_i8(i8* %p1) {
%r = load i8, i8* %p1
ret i8 %r
}
%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
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
ret <4 x i32>* %p1
}
-...
-
----
-name: test_add_i64
-legalized: true
-regBankSelected: true
-# ALL: registers:
-# ALL-NEXT: - { id: 0, class: gr64 }
-# ALL-NEXT: - { id: 1, class: gr64 }
-# ALL-NEXT: - { id: 2, class: gr64 }
-registers:
- - { id: 0, class: gpr }
- - { id: 1, class: gpr }
- - { id: 2, class: gpr }
-# ALL: %0 = COPY %rdi
-# ALL-NEXT: %1 = COPY %rsi
-# ALL-NEXT: %2 = ADD64rr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %edi, %esi
-
- %0(s64) = COPY %rdi
- %1(s64) = COPY %rsi
- %2(s64) = G_ADD %0, %1
- %rax = COPY %2(s64)
-
-...
-
----
-name: test_add_i32
-legalized: true
-regBankSelected: true
-# ALL: registers:
-# ALL-NEXT: - { id: 0, class: gr32 }
-# ALL-NEXT: - { id: 1, class: gr32 }
-# ALL-NEXT: - { id: 2, class: gr32 }
-registers:
- - { id: 0, class: gpr }
- - { id: 1, class: gpr }
- - { id: 2, class: gpr }
-# ALL: %0 = COPY %edi
-# ALL-NEXT: %1 = COPY %esi
-# ALL-NEXT: %2 = ADD32rr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %edi, %esi
-
- %0(s32) = COPY %edi
- %1(s32) = COPY %esi
- %2(s32) = G_ADD %0, %1
- %rax = COPY %2(s32)
-
-...
-
----
-name: test_sub_i64
-legalized: true
-regBankSelected: true
-# ALL: registers:
-# ALL-NEXT: - { id: 0, class: gr64 }
-# ALL-NEXT: - { id: 1, class: gr64 }
-# ALL-NEXT: - { id: 2, class: gr64 }
-registers:
- - { id: 0, class: gpr }
- - { id: 1, class: gpr }
- - { id: 2, class: gpr }
-# ALL: %0 = COPY %rdi
-# ALL-NEXT: %1 = COPY %rsi
-# ALL-NEXT: %2 = SUB64rr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %edi, %esi
-
- %0(s64) = COPY %rdi
- %1(s64) = COPY %rsi
- %2(s64) = G_SUB %0, %1
- %rax = COPY %2(s64)
-
-...
-
----
-name: test_sub_i32
-legalized: true
-regBankSelected: true
-# ALL: registers:
-# ALL-NEXT: - { id: 0, class: gr32 }
-# ALL-NEXT: - { id: 1, class: gr32 }
-# ALL-NEXT: - { id: 2, class: gr32 }
-registers:
- - { id: 0, class: gpr }
- - { id: 1, class: gpr }
- - { id: 2, class: gpr }
-# ALL: %0 = COPY %edi
-# ALL-NEXT: %1 = COPY %esi
-# ALL-NEXT: %2 = SUB32rr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %edi, %esi
-
- %0(s32) = COPY %edi
- %1(s32) = COPY %esi
- %2(s32) = G_SUB %0, %1
- %rax = COPY %2(s32)
-
-...
-
----
-name: test_add_float
-alignment: 4
-legalized: true
-regBankSelected: true
-selected: false
-tracksRegLiveness: true
-# ALL: registers:
-# NO_AVX512F-NEXT: - { id: 0, class: fr32 }
-# NO_AVX512F-NEXT: - { id: 1, class: fr32 }
-# NO_AVX512F-NEXT: - { id: 2, class: fr32 }
-# AVX512ALL-NEXT: - { id: 0, class: fr32x }
-# AVX512ALL-NEXT: - { id: 1, class: fr32x }
-# AVX512ALL-NEXT: - { id: 2, class: fr32x }
-registers:
- - { id: 0, class: vecr }
- - { id: 1, class: vecr }
- - { id: 2, class: vecr }
-# ALL: %0 = COPY %xmm0
-# ALL-NEXT: %1 = COPY %xmm1
-# SSE-NEXT: %2 = ADDSSrr %0, %1
-# AVX-NEXT: %2 = VADDSSrr %0, %1
-# AVX512F-NEXT: %2 = VADDSSZrr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %xmm0, %xmm1
-
- %0(s32) = COPY %xmm0
- %1(s32) = COPY %xmm1
- %2(s32) = G_FADD %0, %1
- %xmm0 = COPY %2(s32)
- RET 0, implicit %xmm0
-
-...
----
-name: test_add_double
-alignment: 4
-legalized: true
-regBankSelected: true
-selected: false
-tracksRegLiveness: true
-# ALL: registers:
-# NO_AVX512F-NEXT: - { id: 0, class: fr64 }
-# NO_AVX512F-NEXT: - { id: 1, class: fr64 }
-# NO_AVX512F-NEXT: - { id: 2, class: fr64 }
-# AVX512ALL-NEXT: - { id: 0, class: fr64x }
-# AVX512ALL-NEXT: - { id: 1, class: fr64x }
-# AVX512ALL-NEXT: - { id: 2, class: fr64x }
-registers:
- - { id: 0, class: vecr }
- - { id: 1, class: vecr }
- - { id: 2, class: vecr }
-# ALL: %0 = COPY %xmm0
-# ALL-NEXT: %1 = COPY %xmm1
-# SSE-NEXT: %2 = ADDSDrr %0, %1
-# AVX-NEXT: %2 = VADDSDrr %0, %1
-# AVX512F-NEXT: %2 = VADDSDZrr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %xmm0, %xmm1
-
- %0(s64) = COPY %xmm0
- %1(s64) = COPY %xmm1
- %2(s64) = G_FADD %0, %1
- %xmm0 = COPY %2(s64)
- RET 0, implicit %xmm0
-
-...
----
-name: test_sub_float
-alignment: 4
-legalized: true
-regBankSelected: true
-selected: false
-tracksRegLiveness: true
-# ALL: registers:
-# NO_AVX512F-NEXT: - { id: 0, class: fr32 }
-# NO_AVX512F-NEXT: - { id: 1, class: fr32 }
-# NO_AVX512F-NEXT: - { id: 2, class: fr32 }
-# AVX512ALL-NEXT: - { id: 0, class: fr32x }
-# AVX512ALL-NEXT: - { id: 1, class: fr32x }
-# AVX512ALL-NEXT: - { id: 2, class: fr32x }
-registers:
- - { id: 0, class: vecr }
- - { id: 1, class: vecr }
- - { id: 2, class: vecr }
-# ALL: %0 = COPY %xmm0
-# ALL-NEXT: %1 = COPY %xmm1
-# SSE-NEXT: %2 = SUBSSrr %0, %1
-# AVX-NEXT: %2 = VSUBSSrr %0, %1
-# AVX512F-NEXT: %2 = VSUBSSZrr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %xmm0, %xmm1
-
- %0(s32) = COPY %xmm0
- %1(s32) = COPY %xmm1
- %2(s32) = G_FSUB %0, %1
- %xmm0 = COPY %2(s32)
- RET 0, implicit %xmm0
-
-...
----
-name: test_sub_double
-alignment: 4
-legalized: true
-regBankSelected: true
-selected: false
-tracksRegLiveness: true
-# ALL: registers:
-# NO_AVX512F-NEXT: - { id: 0, class: fr64 }
-# NO_AVX512F-NEXT: - { id: 1, class: fr64 }
-# NO_AVX512F-NEXT: - { id: 2, class: fr64 }
-# AVX512ALL-NEXT: - { id: 0, class: fr64x }
-# AVX512ALL-NEXT: - { id: 1, class: fr64x }
-# AVX512ALL-NEXT: - { id: 2, class: fr64x }
-registers:
- - { id: 0, class: vecr }
- - { id: 1, class: vecr }
- - { id: 2, class: vecr }
-# ALL: %0 = COPY %xmm0
-# ALL-NEXT: %1 = COPY %xmm1
-# SSE-NEXT: %2 = SUBSDrr %0, %1
-# AVX-NEXT: %2 = VSUBSDrr %0, %1
-# AVX512F-NEXT: %2 = VSUBSDZrr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %xmm0, %xmm1
-
- %0(s64) = COPY %xmm0
- %1(s64) = COPY %xmm1
- %2(s64) = G_FSUB %0, %1
- %xmm0 = COPY %2(s64)
- RET 0, implicit %xmm0
-...
----
-name: test_add_v4i32
-alignment: 4
-legalized: true
-regBankSelected: true
-selected: false
-tracksRegLiveness: true
-# ALL: registers:
-# NO_AVX512VL-NEXT: - { id: 0, class: vr128 }
-# NO_AVX512VL-NEXT: - { id: 1, class: vr128 }
-# NO_AVX512VL-NEXT: - { id: 2, class: vr128 }
-# AVX512VL-NEXT: - { id: 0, class: vr128x }
-# AVX512VL-NEXT: - { id: 1, class: vr128x }
-# AVX512VL-NEXT: - { id: 2, class: vr128x }
-registers:
- - { id: 0, class: vecr }
- - { id: 1, class: vecr }
- - { id: 2, class: vecr }
-# ALL: %0 = COPY %xmm0
-# ALL-NEXT: %1 = COPY %xmm1
-# SSE-NEXT: %2 = PADDDrr %0, %1
-# AVX-NEXT: %2 = VPADDDrr %0, %1
-# AVX512F-NEXT: %2 = VPADDDrr %0, %1
-# AVX512VL-NEXT: %2 = VPADDDZ128rr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %xmm0, %xmm1
-
- %0(<4 x s32>) = COPY %xmm0
- %1(<4 x s32>) = COPY %xmm1
- %2(<4 x s32>) = G_ADD %0, %1
- %xmm0 = COPY %2(<4 x s32>)
- RET 0, implicit %xmm0
-
-...
----
-name: test_sub_v4i32
-alignment: 4
-legalized: true
-regBankSelected: true
-selected: false
-tracksRegLiveness: true
-# ALL: registers:
-# NO_AVX512VL-NEXT: - { id: 0, class: vr128 }
-# NO_AVX512VL-NEXT: - { id: 1, class: vr128 }
-# NO_AVX512VL-NEXT: - { id: 2, class: vr128 }
-# AVX512VL-NEXT: - { id: 0, class: vr128x }
-# AVX512VL-NEXT: - { id: 1, class: vr128x }
-# AVX512VL-NEXT: - { id: 2, class: vr128x }
-registers:
- - { id: 0, class: vecr }
- - { id: 1, class: vecr }
- - { id: 2, class: vecr }
-# ALL: %0 = COPY %xmm0
-# ALL-NEXT: %1 = COPY %xmm1
-# SSE-NEXT: %2 = PSUBDrr %0, %1
-# AVX-NEXT: %2 = VPSUBDrr %0, %1
-# AVX512F-NEXT: %2 = VPSUBDrr %0, %1
-# AVX512VL-NEXT: %2 = VPSUBDZ128rr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %xmm0, %xmm1
-
- %0(<4 x s32>) = COPY %xmm0
- %1(<4 x s32>) = COPY %xmm1
- %2(<4 x s32>) = G_SUB %0, %1
- %xmm0 = COPY %2(<4 x s32>)
- RET 0, implicit %xmm0
-
-...
----
-name: test_add_v4f32
-alignment: 4
-legalized: true
-regBankSelected: true
-selected: false
-tracksRegLiveness: true
-# ALL: registers:
-# NO_AVX512VL-NEXT: - { id: 0, class: vr128 }
-# NO_AVX512VL-NEXT: - { id: 1, class: vr128 }
-# NO_AVX512VL-NEXT: - { id: 2, class: vr128 }
-# AVX512VL-NEXT: - { id: 0, class: vr128x }
-# AVX512VL-NEXT: - { id: 1, class: vr128x }
-# AVX512VL-NEXT: - { id: 2, class: vr128x }
-registers:
- - { id: 0, class: vecr }
- - { id: 1, class: vecr }
- - { id: 2, class: vecr }
-# ALL: %0 = COPY %xmm0
-# ALL-NEXT: %1 = COPY %xmm1
-# SSE-NEXT: %2 = ADDPSrr %0, %1
-# AVX-NEXT: %2 = VADDPSrr %0, %1
-# AVX512F-NEXT: %2 = VADDPSrr %0, %1
-# AVX512VL-NEXT: %2 = VADDPSZ128rr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %xmm0, %xmm1
-
- %0(<4 x s32>) = COPY %xmm0
- %1(<4 x s32>) = COPY %xmm1
- %2(<4 x s32>) = G_FADD %0, %1
- %xmm0 = COPY %2(<4 x s32>)
- RET 0, implicit %xmm0
-
-...
----
-name: test_sub_v4f32
-alignment: 4
-legalized: true
-regBankSelected: true
-selected: false
-tracksRegLiveness: true
-# ALL: registers:
-# NO_AVX512VL-NEXT: - { id: 0, class: vr128 }
-# NO_AVX512VL-NEXT: - { id: 1, class: vr128 }
-# NO_AVX512VL-NEXT: - { id: 2, class: vr128 }
-# AVX512VL-NEXT: - { id: 0, class: vr128x }
-# AVX512VL-NEXT: - { id: 1, class: vr128x }
-# AVX512VL-NEXT: - { id: 2, class: vr128x }
-registers:
- - { id: 0, class: vecr }
- - { id: 1, class: vecr }
- - { id: 2, class: vecr }
-# ALL: %0 = COPY %xmm0
-# ALL-NEXT: %1 = COPY %xmm1
-# SSE-NEXT: %2 = SUBPSrr %0, %1
-# AVX-NEXT: %2 = VSUBPSrr %0, %1
-# AVX512F-NEXT: %2 = VSUBPSrr %0, %1
-# AVX512VL-NEXT: %2 = VSUBPSZ128rr %0, %1
-body: |
- bb.1 (%ir-block.0):
- liveins: %xmm0, %xmm1
-
- %0(<4 x s32>) = COPY %xmm0
- %1(<4 x s32>) = COPY %xmm1
- %2(<4 x s32>) = G_FSUB %0, %1
- %xmm0 = COPY %2(<4 x s32>)
- RET 0, implicit %xmm0
-
...
---
# ALL-LABEL: name: test_load_i8
--- /dev/null
+# 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 i64 @test_sub_i64(i64 %arg1, i64 %arg2) {
+ %ret = sub i64 %arg1, %arg2
+ ret i64 %ret
+ }
+
+ define i32 @test_sub_i32(i32 %arg1, i32 %arg2) {
+ %ret = sub i32 %arg1, %arg2
+ ret i32 %ret
+ }
+
+ define float @test_sub_float(float %arg1, float %arg2) {
+ %ret = fsub float %arg1, %arg2
+ ret float %ret
+ }
+
+ define double @test_sub_double(double %arg1, double %arg2) {
+ %ret = fsub double %arg1, %arg2
+ ret double %ret
+ }
+
+ define <4 x i32> @test_sub_v4i32(<4 x i32> %arg1, <4 x i32> %arg2) {
+ %ret = sub <4 x i32> %arg1, %arg2
+ ret <4 x i32> %ret
+ }
+
+ define <4 x float> @test_sub_v4f32(<4 x float> %arg1, <4 x float> %arg2) {
+ %ret = fsub <4 x float> %arg1, %arg2
+ ret <4 x float> %ret
+ }
+
+...
+---
+name: test_sub_i64
+legalized: true
+regBankSelected: true
+# ALL: registers:
+# ALL-NEXT: - { id: 0, class: gr64 }
+# ALL-NEXT: - { id: 1, class: gr64 }
+# ALL-NEXT: - { id: 2, class: gr64 }
+registers:
+ - { id: 0, class: gpr }
+ - { id: 1, class: gpr }
+ - { id: 2, class: gpr }
+# ALL: %0 = COPY %rdi
+# ALL-NEXT: %1 = COPY %rsi
+# ALL-NEXT: %2 = SUB64rr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %edi, %esi
+
+ %0(s64) = COPY %rdi
+ %1(s64) = COPY %rsi
+ %2(s64) = G_SUB %0, %1
+ %rax = COPY %2(s64)
+
+...
+
+---
+name: test_sub_i32
+legalized: true
+regBankSelected: true
+# ALL: registers:
+# ALL-NEXT: - { id: 0, class: gr32 }
+# ALL-NEXT: - { id: 1, class: gr32 }
+# ALL-NEXT: - { id: 2, class: gr32 }
+registers:
+ - { id: 0, class: gpr }
+ - { id: 1, class: gpr }
+ - { id: 2, class: gpr }
+# ALL: %0 = COPY %edi
+# ALL-NEXT: %1 = COPY %esi
+# ALL-NEXT: %2 = SUB32rr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %edi, %esi
+
+ %0(s32) = COPY %edi
+ %1(s32) = COPY %esi
+ %2(s32) = G_SUB %0, %1
+ %rax = COPY %2(s32)
+
+...
+---
+name: test_sub_float
+alignment: 4
+legalized: true
+regBankSelected: true
+selected: false
+tracksRegLiveness: true
+# ALL: registers:
+# NO_AVX512F-NEXT: - { id: 0, class: fr32 }
+# NO_AVX512F-NEXT: - { id: 1, class: fr32 }
+# NO_AVX512F-NEXT: - { id: 2, class: fr32 }
+# AVX512ALL-NEXT: - { id: 0, class: fr32x }
+# AVX512ALL-NEXT: - { id: 1, class: fr32x }
+# AVX512ALL-NEXT: - { id: 2, class: fr32x }
+registers:
+ - { id: 0, class: vecr }
+ - { id: 1, class: vecr }
+ - { id: 2, class: vecr }
+# ALL: %0 = COPY %xmm0
+# ALL-NEXT: %1 = COPY %xmm1
+# SSE-NEXT: %2 = SUBSSrr %0, %1
+# AVX-NEXT: %2 = VSUBSSrr %0, %1
+# AVX512F-NEXT: %2 = VSUBSSZrr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %xmm0, %xmm1
+
+ %0(s32) = COPY %xmm0
+ %1(s32) = COPY %xmm1
+ %2(s32) = G_FSUB %0, %1
+ %xmm0 = COPY %2(s32)
+ RET 0, implicit %xmm0
+
+...
+---
+name: test_sub_double
+alignment: 4
+legalized: true
+regBankSelected: true
+selected: false
+tracksRegLiveness: true
+# ALL: registers:
+# NO_AVX512F-NEXT: - { id: 0, class: fr64 }
+# NO_AVX512F-NEXT: - { id: 1, class: fr64 }
+# NO_AVX512F-NEXT: - { id: 2, class: fr64 }
+# AVX512ALL-NEXT: - { id: 0, class: fr64x }
+# AVX512ALL-NEXT: - { id: 1, class: fr64x }
+# AVX512ALL-NEXT: - { id: 2, class: fr64x }
+registers:
+ - { id: 0, class: vecr }
+ - { id: 1, class: vecr }
+ - { id: 2, class: vecr }
+# ALL: %0 = COPY %xmm0
+# ALL-NEXT: %1 = COPY %xmm1
+# SSE-NEXT: %2 = SUBSDrr %0, %1
+# AVX-NEXT: %2 = VSUBSDrr %0, %1
+# AVX512F-NEXT: %2 = VSUBSDZrr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %xmm0, %xmm1
+
+ %0(s64) = COPY %xmm0
+ %1(s64) = COPY %xmm1
+ %2(s64) = G_FSUB %0, %1
+ %xmm0 = COPY %2(s64)
+ RET 0, implicit %xmm0
+...
+---
+name: test_sub_v4i32
+alignment: 4
+legalized: true
+regBankSelected: true
+selected: false
+tracksRegLiveness: true
+# ALL: registers:
+# NO_AVX512VL-NEXT: - { id: 0, class: vr128 }
+# NO_AVX512VL-NEXT: - { id: 1, class: vr128 }
+# NO_AVX512VL-NEXT: - { id: 2, class: vr128 }
+# AVX512VL-NEXT: - { id: 0, class: vr128x }
+# AVX512VL-NEXT: - { id: 1, class: vr128x }
+# AVX512VL-NEXT: - { id: 2, class: vr128x }
+registers:
+ - { id: 0, class: vecr }
+ - { id: 1, class: vecr }
+ - { id: 2, class: vecr }
+# ALL: %0 = COPY %xmm0
+# ALL-NEXT: %1 = COPY %xmm1
+# SSE-NEXT: %2 = PSUBDrr %0, %1
+# AVX-NEXT: %2 = VPSUBDrr %0, %1
+# AVX512F-NEXT: %2 = VPSUBDrr %0, %1
+# AVX512VL-NEXT: %2 = VPSUBDZ128rr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %xmm0, %xmm1
+
+ %0(<4 x s32>) = COPY %xmm0
+ %1(<4 x s32>) = COPY %xmm1
+ %2(<4 x s32>) = G_SUB %0, %1
+ %xmm0 = COPY %2(<4 x s32>)
+ RET 0, implicit %xmm0
+
+...
+---
+name: test_sub_v4f32
+alignment: 4
+legalized: true
+regBankSelected: true
+selected: false
+tracksRegLiveness: true
+# ALL: registers:
+# NO_AVX512VL-NEXT: - { id: 0, class: vr128 }
+# NO_AVX512VL-NEXT: - { id: 1, class: vr128 }
+# NO_AVX512VL-NEXT: - { id: 2, class: vr128 }
+# AVX512VL-NEXT: - { id: 0, class: vr128x }
+# AVX512VL-NEXT: - { id: 1, class: vr128x }
+# AVX512VL-NEXT: - { id: 2, class: vr128x }
+registers:
+ - { id: 0, class: vecr }
+ - { id: 1, class: vecr }
+ - { id: 2, class: vecr }
+# ALL: %0 = COPY %xmm0
+# ALL-NEXT: %1 = COPY %xmm1
+# SSE-NEXT: %2 = SUBPSrr %0, %1
+# AVX-NEXT: %2 = VSUBPSrr %0, %1
+# AVX512F-NEXT: %2 = VSUBPSrr %0, %1
+# AVX512VL-NEXT: %2 = VSUBPSZ128rr %0, %1
+body: |
+ bb.1 (%ir-block.0):
+ liveins: %xmm0, %xmm1
+
+ %0(<4 x s32>) = COPY %xmm0
+ %1(<4 x s32>) = COPY %xmm1
+ %2(<4 x s32>) = G_FSUB %0, %1
+ %xmm0 = COPY %2(<4 x s32>)
+ RET 0, implicit %xmm0
+
+...