derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
437 lines
16 KiB
LLVM
437 lines
16 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -mtriple i686-pc-linux < %s | FileCheck %s --check-prefix=CHECK-32
|
|
; RUN: llc -mtriple x86_64-pc-linux < %s | FileCheck %s --check-prefix=CHECK-64
|
|
|
|
define i8 @atomicrmw_usub_cond_i8(ptr %ptr, i8 %val) {
|
|
; CHECK-32-LABEL: atomicrmw_usub_cond_i8:
|
|
; CHECK-32: # %bb.0:
|
|
; CHECK-32-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; CHECK-32-NEXT: movzbl (%edx), %eax
|
|
; CHECK-32-NEXT: jmp .LBB0_1
|
|
; CHECK-32-NEXT: .p2align 4
|
|
; CHECK-32-NEXT: .LBB0_3: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB0_1 Depth=1
|
|
; CHECK-32-NEXT: lock cmpxchgb %ah, (%edx)
|
|
; CHECK-32-NEXT: je .LBB0_4
|
|
; CHECK-32-NEXT: .LBB0_1: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-32-NEXT: movb %al, %ah
|
|
; CHECK-32-NEXT: subb %cl, %ah
|
|
; CHECK-32-NEXT: jae .LBB0_3
|
|
; CHECK-32-NEXT: # %bb.2: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB0_1 Depth=1
|
|
; CHECK-32-NEXT: movb %al, %ah
|
|
; CHECK-32-NEXT: jmp .LBB0_3
|
|
; CHECK-32-NEXT: .LBB0_4: # %atomicrmw.end
|
|
; CHECK-32-NEXT: retl
|
|
;
|
|
; CHECK-64-LABEL: atomicrmw_usub_cond_i8:
|
|
; CHECK-64: # %bb.0:
|
|
; CHECK-64-NEXT: movzbl (%rdi), %eax
|
|
; CHECK-64-NEXT: .p2align 4
|
|
; CHECK-64-NEXT: .LBB0_1: # %atomicrmw.start
|
|
; CHECK-64-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-64-NEXT: movzbl %al, %ecx
|
|
; CHECK-64-NEXT: subb %sil, %al
|
|
; CHECK-64-NEXT: movzbl %al, %edx
|
|
; CHECK-64-NEXT: cmovbl %ecx, %edx
|
|
; CHECK-64-NEXT: movl %ecx, %eax
|
|
; CHECK-64-NEXT: lock cmpxchgb %dl, (%rdi)
|
|
; CHECK-64-NEXT: jne .LBB0_1
|
|
; CHECK-64-NEXT: # %bb.2: # %atomicrmw.end
|
|
; CHECK-64-NEXT: retq
|
|
%result = atomicrmw usub_cond ptr %ptr, i8 %val seq_cst
|
|
ret i8 %result
|
|
}
|
|
|
|
define i16 @atomicrmw_usub_cond_i16(ptr %ptr, i16 %val) {
|
|
; CHECK-32-LABEL: atomicrmw_usub_cond_i16:
|
|
; CHECK-32: # %bb.0:
|
|
; CHECK-32-NEXT: pushl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: .cfi_offset %esi, -8
|
|
; CHECK-32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; CHECK-32-NEXT: movzwl (%edx), %eax
|
|
; CHECK-32-NEXT: jmp .LBB1_1
|
|
; CHECK-32-NEXT: .p2align 4
|
|
; CHECK-32-NEXT: .LBB1_3: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB1_1 Depth=1
|
|
; CHECK-32-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; CHECK-32-NEXT: lock cmpxchgw %si, (%edx)
|
|
; CHECK-32-NEXT: # kill: def $ax killed $ax def $eax
|
|
; CHECK-32-NEXT: je .LBB1_4
|
|
; CHECK-32-NEXT: .LBB1_1: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-32-NEXT: movl %eax, %esi
|
|
; CHECK-32-NEXT: subw %cx, %si
|
|
; CHECK-32-NEXT: jae .LBB1_3
|
|
; CHECK-32-NEXT: # %bb.2: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB1_1 Depth=1
|
|
; CHECK-32-NEXT: movl %eax, %esi
|
|
; CHECK-32-NEXT: jmp .LBB1_3
|
|
; CHECK-32-NEXT: .LBB1_4: # %atomicrmw.end
|
|
; CHECK-32-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; CHECK-32-NEXT: popl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 4
|
|
; CHECK-32-NEXT: retl
|
|
;
|
|
; CHECK-64-LABEL: atomicrmw_usub_cond_i16:
|
|
; CHECK-64: # %bb.0:
|
|
; CHECK-64-NEXT: movzwl (%rdi), %eax
|
|
; CHECK-64-NEXT: .p2align 4
|
|
; CHECK-64-NEXT: .LBB1_1: # %atomicrmw.start
|
|
; CHECK-64-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-64-NEXT: movl %eax, %ecx
|
|
; CHECK-64-NEXT: subw %si, %cx
|
|
; CHECK-64-NEXT: cmovbl %eax, %ecx
|
|
; CHECK-64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; CHECK-64-NEXT: lock cmpxchgw %cx, (%rdi)
|
|
; CHECK-64-NEXT: # kill: def $ax killed $ax def $eax
|
|
; CHECK-64-NEXT: jne .LBB1_1
|
|
; CHECK-64-NEXT: # %bb.2: # %atomicrmw.end
|
|
; CHECK-64-NEXT: # kill: def $ax killed $ax killed $eax
|
|
; CHECK-64-NEXT: retq
|
|
%result = atomicrmw usub_cond ptr %ptr, i16 %val seq_cst
|
|
ret i16 %result
|
|
}
|
|
|
|
define i32 @atomicrmw_usub_cond_i32(ptr %ptr, i32 %val) {
|
|
; CHECK-32-LABEL: atomicrmw_usub_cond_i32:
|
|
; CHECK-32: # %bb.0:
|
|
; CHECK-32-NEXT: pushl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: .cfi_offset %esi, -8
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; CHECK-32-NEXT: movl (%edx), %eax
|
|
; CHECK-32-NEXT: jmp .LBB2_1
|
|
; CHECK-32-NEXT: .p2align 4
|
|
; CHECK-32-NEXT: .LBB2_3: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB2_1 Depth=1
|
|
; CHECK-32-NEXT: lock cmpxchgl %esi, (%edx)
|
|
; CHECK-32-NEXT: je .LBB2_4
|
|
; CHECK-32-NEXT: .LBB2_1: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-32-NEXT: movl %eax, %esi
|
|
; CHECK-32-NEXT: subl %ecx, %esi
|
|
; CHECK-32-NEXT: jae .LBB2_3
|
|
; CHECK-32-NEXT: # %bb.2: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB2_1 Depth=1
|
|
; CHECK-32-NEXT: movl %eax, %esi
|
|
; CHECK-32-NEXT: jmp .LBB2_3
|
|
; CHECK-32-NEXT: .LBB2_4: # %atomicrmw.end
|
|
; CHECK-32-NEXT: popl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 4
|
|
; CHECK-32-NEXT: retl
|
|
;
|
|
; CHECK-64-LABEL: atomicrmw_usub_cond_i32:
|
|
; CHECK-64: # %bb.0:
|
|
; CHECK-64-NEXT: movl (%rdi), %eax
|
|
; CHECK-64-NEXT: .p2align 4
|
|
; CHECK-64-NEXT: .LBB2_1: # %atomicrmw.start
|
|
; CHECK-64-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-64-NEXT: movl %eax, %ecx
|
|
; CHECK-64-NEXT: subl %esi, %ecx
|
|
; CHECK-64-NEXT: cmovbl %eax, %ecx
|
|
; CHECK-64-NEXT: lock cmpxchgl %ecx, (%rdi)
|
|
; CHECK-64-NEXT: jne .LBB2_1
|
|
; CHECK-64-NEXT: # %bb.2: # %atomicrmw.end
|
|
; CHECK-64-NEXT: retq
|
|
%result = atomicrmw usub_cond ptr %ptr, i32 %val seq_cst
|
|
ret i32 %result
|
|
}
|
|
|
|
define i64 @atomicrmw_usub_cond_i64(ptr %ptr, i64 %val) {
|
|
; CHECK-32-LABEL: atomicrmw_usub_cond_i64:
|
|
; CHECK-32: # %bb.0:
|
|
; CHECK-32-NEXT: pushl %ebp
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: pushl %ebx
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 12
|
|
; CHECK-32-NEXT: pushl %edi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 16
|
|
; CHECK-32-NEXT: pushl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 20
|
|
; CHECK-32-NEXT: .cfi_offset %esi, -20
|
|
; CHECK-32-NEXT: .cfi_offset %edi, -16
|
|
; CHECK-32-NEXT: .cfi_offset %ebx, -12
|
|
; CHECK-32-NEXT: .cfi_offset %ebp, -8
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %esi
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %edi
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %ebp
|
|
; CHECK-32-NEXT: movl (%ebp), %eax
|
|
; CHECK-32-NEXT: movl 4(%ebp), %edx
|
|
; CHECK-32-NEXT: jmp .LBB3_1
|
|
; CHECK-32-NEXT: .p2align 4
|
|
; CHECK-32-NEXT: .LBB3_3: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB3_1 Depth=1
|
|
; CHECK-32-NEXT: lock cmpxchg8b (%ebp)
|
|
; CHECK-32-NEXT: je .LBB3_4
|
|
; CHECK-32-NEXT: .LBB3_1: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-32-NEXT: movl %eax, %ebx
|
|
; CHECK-32-NEXT: subl %edi, %ebx
|
|
; CHECK-32-NEXT: movl %edx, %ecx
|
|
; CHECK-32-NEXT: sbbl %esi, %ecx
|
|
; CHECK-32-NEXT: jae .LBB3_3
|
|
; CHECK-32-NEXT: # %bb.2: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB3_1 Depth=1
|
|
; CHECK-32-NEXT: movl %edx, %ecx
|
|
; CHECK-32-NEXT: movl %eax, %ebx
|
|
; CHECK-32-NEXT: jmp .LBB3_3
|
|
; CHECK-32-NEXT: .LBB3_4: # %atomicrmw.end
|
|
; CHECK-32-NEXT: popl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 16
|
|
; CHECK-32-NEXT: popl %edi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 12
|
|
; CHECK-32-NEXT: popl %ebx
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: popl %ebp
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 4
|
|
; CHECK-32-NEXT: retl
|
|
;
|
|
; CHECK-64-LABEL: atomicrmw_usub_cond_i64:
|
|
; CHECK-64: # %bb.0:
|
|
; CHECK-64-NEXT: movq (%rdi), %rax
|
|
; CHECK-64-NEXT: .p2align 4
|
|
; CHECK-64-NEXT: .LBB3_1: # %atomicrmw.start
|
|
; CHECK-64-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-64-NEXT: movq %rax, %rcx
|
|
; CHECK-64-NEXT: subq %rsi, %rcx
|
|
; CHECK-64-NEXT: cmovbq %rax, %rcx
|
|
; CHECK-64-NEXT: lock cmpxchgq %rcx, (%rdi)
|
|
; CHECK-64-NEXT: jne .LBB3_1
|
|
; CHECK-64-NEXT: # %bb.2: # %atomicrmw.end
|
|
; CHECK-64-NEXT: retq
|
|
%result = atomicrmw usub_cond ptr %ptr, i64 %val seq_cst
|
|
ret i64 %result
|
|
}
|
|
|
|
define i8 @atomicrmw_usub_sat_i8(ptr %ptr, i8 %val) {
|
|
; CHECK-32-LABEL: atomicrmw_usub_sat_i8:
|
|
; CHECK-32: # %bb.0:
|
|
; CHECK-32-NEXT: pushl %ebx
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: .cfi_offset %ebx, -8
|
|
; CHECK-32-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; CHECK-32-NEXT: movzbl (%edx), %eax
|
|
; CHECK-32-NEXT: jmp .LBB4_1
|
|
; CHECK-32-NEXT: .p2align 4
|
|
; CHECK-32-NEXT: .LBB4_3: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB4_1 Depth=1
|
|
; CHECK-32-NEXT: lock cmpxchgb %bl, (%edx)
|
|
; CHECK-32-NEXT: je .LBB4_4
|
|
; CHECK-32-NEXT: .LBB4_1: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-32-NEXT: movl %eax, %ebx
|
|
; CHECK-32-NEXT: subb %cl, %bl
|
|
; CHECK-32-NEXT: jae .LBB4_3
|
|
; CHECK-32-NEXT: # %bb.2: # in Loop: Header=BB4_1 Depth=1
|
|
; CHECK-32-NEXT: xorl %ebx, %ebx
|
|
; CHECK-32-NEXT: jmp .LBB4_3
|
|
; CHECK-32-NEXT: .LBB4_4: # %atomicrmw.end
|
|
; CHECK-32-NEXT: popl %ebx
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 4
|
|
; CHECK-32-NEXT: retl
|
|
;
|
|
; CHECK-64-LABEL: atomicrmw_usub_sat_i8:
|
|
; CHECK-64: # %bb.0:
|
|
; CHECK-64-NEXT: movzbl (%rdi), %eax
|
|
; CHECK-64-NEXT: xorl %ecx, %ecx
|
|
; CHECK-64-NEXT: .p2align 4
|
|
; CHECK-64-NEXT: .LBB4_1: # %atomicrmw.start
|
|
; CHECK-64-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-64-NEXT: movl %eax, %edx
|
|
; CHECK-64-NEXT: subb %sil, %dl
|
|
; CHECK-64-NEXT: movzbl %dl, %edx
|
|
; CHECK-64-NEXT: cmovbl %ecx, %edx
|
|
; CHECK-64-NEXT: lock cmpxchgb %dl, (%rdi)
|
|
; CHECK-64-NEXT: jne .LBB4_1
|
|
; CHECK-64-NEXT: # %bb.2: # %atomicrmw.end
|
|
; CHECK-64-NEXT: retq
|
|
%result = atomicrmw usub_sat ptr %ptr, i8 %val seq_cst
|
|
ret i8 %result
|
|
}
|
|
|
|
define i16 @atomicrmw_usub_sat_i16(ptr %ptr, i16 %val) {
|
|
; CHECK-32-LABEL: atomicrmw_usub_sat_i16:
|
|
; CHECK-32: # %bb.0:
|
|
; CHECK-32-NEXT: pushl %edi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: pushl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 12
|
|
; CHECK-32-NEXT: .cfi_offset %esi, -12
|
|
; CHECK-32-NEXT: .cfi_offset %edi, -8
|
|
; CHECK-32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; CHECK-32-NEXT: movzwl (%edx), %eax
|
|
; CHECK-32-NEXT: jmp .LBB5_1
|
|
; CHECK-32-NEXT: .p2align 4
|
|
; CHECK-32-NEXT: .LBB5_3: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB5_1 Depth=1
|
|
; CHECK-32-NEXT: lock cmpxchgw %si, (%edx)
|
|
; CHECK-32-NEXT: je .LBB5_4
|
|
; CHECK-32-NEXT: .LBB5_1: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-32-NEXT: xorl %esi, %esi
|
|
; CHECK-32-NEXT: movl %eax, %edi
|
|
; CHECK-32-NEXT: subw %cx, %di
|
|
; CHECK-32-NEXT: jb .LBB5_3
|
|
; CHECK-32-NEXT: # %bb.2: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB5_1 Depth=1
|
|
; CHECK-32-NEXT: movl %edi, %esi
|
|
; CHECK-32-NEXT: jmp .LBB5_3
|
|
; CHECK-32-NEXT: .LBB5_4: # %atomicrmw.end
|
|
; CHECK-32-NEXT: popl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: popl %edi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 4
|
|
; CHECK-32-NEXT: retl
|
|
;
|
|
; CHECK-64-LABEL: atomicrmw_usub_sat_i16:
|
|
; CHECK-64: # %bb.0:
|
|
; CHECK-64-NEXT: movzwl (%rdi), %eax
|
|
; CHECK-64-NEXT: xorl %ecx, %ecx
|
|
; CHECK-64-NEXT: .p2align 4
|
|
; CHECK-64-NEXT: .LBB5_1: # %atomicrmw.start
|
|
; CHECK-64-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-64-NEXT: movl %eax, %edx
|
|
; CHECK-64-NEXT: subw %si, %dx
|
|
; CHECK-64-NEXT: cmovbl %ecx, %edx
|
|
; CHECK-64-NEXT: lock cmpxchgw %dx, (%rdi)
|
|
; CHECK-64-NEXT: jne .LBB5_1
|
|
; CHECK-64-NEXT: # %bb.2: # %atomicrmw.end
|
|
; CHECK-64-NEXT: retq
|
|
%result = atomicrmw usub_sat ptr %ptr, i16 %val seq_cst
|
|
ret i16 %result
|
|
}
|
|
|
|
define i32 @atomicrmw_usub_sat_i32(ptr %ptr, i32 %val) {
|
|
; CHECK-32-LABEL: atomicrmw_usub_sat_i32:
|
|
; CHECK-32: # %bb.0:
|
|
; CHECK-32-NEXT: pushl %edi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: pushl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 12
|
|
; CHECK-32-NEXT: .cfi_offset %esi, -12
|
|
; CHECK-32-NEXT: .cfi_offset %edi, -8
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %ecx
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %edx
|
|
; CHECK-32-NEXT: movl (%edx), %eax
|
|
; CHECK-32-NEXT: jmp .LBB6_1
|
|
; CHECK-32-NEXT: .p2align 4
|
|
; CHECK-32-NEXT: .LBB6_3: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB6_1 Depth=1
|
|
; CHECK-32-NEXT: lock cmpxchgl %esi, (%edx)
|
|
; CHECK-32-NEXT: je .LBB6_4
|
|
; CHECK-32-NEXT: .LBB6_1: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-32-NEXT: xorl %esi, %esi
|
|
; CHECK-32-NEXT: movl %eax, %edi
|
|
; CHECK-32-NEXT: subl %ecx, %edi
|
|
; CHECK-32-NEXT: jb .LBB6_3
|
|
; CHECK-32-NEXT: # %bb.2: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB6_1 Depth=1
|
|
; CHECK-32-NEXT: movl %edi, %esi
|
|
; CHECK-32-NEXT: jmp .LBB6_3
|
|
; CHECK-32-NEXT: .LBB6_4: # %atomicrmw.end
|
|
; CHECK-32-NEXT: popl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: popl %edi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 4
|
|
; CHECK-32-NEXT: retl
|
|
;
|
|
; CHECK-64-LABEL: atomicrmw_usub_sat_i32:
|
|
; CHECK-64: # %bb.0:
|
|
; CHECK-64-NEXT: movl (%rdi), %eax
|
|
; CHECK-64-NEXT: xorl %ecx, %ecx
|
|
; CHECK-64-NEXT: .p2align 4
|
|
; CHECK-64-NEXT: .LBB6_1: # %atomicrmw.start
|
|
; CHECK-64-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-64-NEXT: movl %eax, %edx
|
|
; CHECK-64-NEXT: subl %esi, %edx
|
|
; CHECK-64-NEXT: cmovbl %ecx, %edx
|
|
; CHECK-64-NEXT: lock cmpxchgl %edx, (%rdi)
|
|
; CHECK-64-NEXT: jne .LBB6_1
|
|
; CHECK-64-NEXT: # %bb.2: # %atomicrmw.end
|
|
; CHECK-64-NEXT: retq
|
|
%result = atomicrmw usub_sat ptr %ptr, i32 %val seq_cst
|
|
ret i32 %result
|
|
}
|
|
|
|
define i64 @atomicrmw_usub_sat_i64(ptr %ptr, i64 %val) {
|
|
; CHECK-32-LABEL: atomicrmw_usub_sat_i64:
|
|
; CHECK-32: # %bb.0:
|
|
; CHECK-32-NEXT: pushl %ebp
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: pushl %ebx
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 12
|
|
; CHECK-32-NEXT: pushl %edi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 16
|
|
; CHECK-32-NEXT: pushl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 20
|
|
; CHECK-32-NEXT: .cfi_offset %esi, -20
|
|
; CHECK-32-NEXT: .cfi_offset %edi, -16
|
|
; CHECK-32-NEXT: .cfi_offset %ebx, -12
|
|
; CHECK-32-NEXT: .cfi_offset %ebp, -8
|
|
; CHECK-32-NEXT: movl {{[0-9]+}}(%esp), %ebp
|
|
; CHECK-32-NEXT: movl (%ebp), %esi
|
|
; CHECK-32-NEXT: movl 4(%ebp), %edi
|
|
; CHECK-32-NEXT: jmp .LBB7_1
|
|
; CHECK-32-NEXT: .p2align 4
|
|
; CHECK-32-NEXT: .LBB7_3: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB7_1 Depth=1
|
|
; CHECK-32-NEXT: movl %esi, %eax
|
|
; CHECK-32-NEXT: movl %edi, %edx
|
|
; CHECK-32-NEXT: lock cmpxchg8b (%ebp)
|
|
; CHECK-32-NEXT: movl %eax, %esi
|
|
; CHECK-32-NEXT: movl %edx, %edi
|
|
; CHECK-32-NEXT: je .LBB7_4
|
|
; CHECK-32-NEXT: .LBB7_1: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-32-NEXT: xorl %ecx, %ecx
|
|
; CHECK-32-NEXT: movl %esi, %eax
|
|
; CHECK-32-NEXT: subl {{[0-9]+}}(%esp), %eax
|
|
; CHECK-32-NEXT: movl %edi, %edx
|
|
; CHECK-32-NEXT: sbbl {{[0-9]+}}(%esp), %edx
|
|
; CHECK-32-NEXT: movl $0, %ebx
|
|
; CHECK-32-NEXT: jb .LBB7_3
|
|
; CHECK-32-NEXT: # %bb.2: # %atomicrmw.start
|
|
; CHECK-32-NEXT: # in Loop: Header=BB7_1 Depth=1
|
|
; CHECK-32-NEXT: movl %edx, %ecx
|
|
; CHECK-32-NEXT: movl %eax, %ebx
|
|
; CHECK-32-NEXT: jmp .LBB7_3
|
|
; CHECK-32-NEXT: .LBB7_4: # %atomicrmw.end
|
|
; CHECK-32-NEXT: movl %esi, %eax
|
|
; CHECK-32-NEXT: movl %edi, %edx
|
|
; CHECK-32-NEXT: popl %esi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 16
|
|
; CHECK-32-NEXT: popl %edi
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 12
|
|
; CHECK-32-NEXT: popl %ebx
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 8
|
|
; CHECK-32-NEXT: popl %ebp
|
|
; CHECK-32-NEXT: .cfi_def_cfa_offset 4
|
|
; CHECK-32-NEXT: retl
|
|
;
|
|
; CHECK-64-LABEL: atomicrmw_usub_sat_i64:
|
|
; CHECK-64: # %bb.0:
|
|
; CHECK-64-NEXT: movq (%rdi), %rax
|
|
; CHECK-64-NEXT: xorl %ecx, %ecx
|
|
; CHECK-64-NEXT: .p2align 4
|
|
; CHECK-64-NEXT: .LBB7_1: # %atomicrmw.start
|
|
; CHECK-64-NEXT: # =>This Inner Loop Header: Depth=1
|
|
; CHECK-64-NEXT: movq %rax, %rdx
|
|
; CHECK-64-NEXT: subq %rsi, %rdx
|
|
; CHECK-64-NEXT: cmovbq %rcx, %rdx
|
|
; CHECK-64-NEXT: lock cmpxchgq %rdx, (%rdi)
|
|
; CHECK-64-NEXT: jne .LBB7_1
|
|
; CHECK-64-NEXT: # %bb.2: # %atomicrmw.end
|
|
; CHECK-64-NEXT: retq
|
|
%result = atomicrmw usub_sat ptr %ptr, i64 %val seq_cst
|
|
ret i64 %result
|
|
}
|