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
502 lines
20 KiB
LLVM
502 lines
20 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
|
|
; Test the -max-store-memset, -max-store-memcpy, and -max-store-memmove flags,
|
|
; which control the threshold for when memset/memcpy/memmove are inlined vs
|
|
; calling the library function.
|
|
|
|
; RUN: llc %s -o - -mtriple=aarch64 | FileCheck %s --check-prefix=DEFAULT
|
|
; RUN: llc %s -o - -mtriple=aarch64 -max-store-memset=7 | FileCheck %s --check-prefix=MEMSET-LIMIT-7
|
|
; RUN: llc %s -o - -mtriple=aarch64 -max-store-memset=8 | FileCheck %s --check-prefix=MEMSET-LIMIT-8
|
|
; RUN: llc %s -o - -mtriple=aarch64 -max-store-memcpy=7 | FileCheck %s --check-prefix=MEMCPY-LIMIT-7
|
|
; RUN: llc %s -o - -mtriple=aarch64 -max-store-memcpy=8 | FileCheck %s --check-prefix=MEMCPY-LIMIT-8
|
|
; RUN: llc %s -o - -mtriple=aarch64 -max-store-memmove=7 | FileCheck %s --check-prefix=MEMMOVE-LIMIT-7
|
|
; RUN: llc %s -o - -mtriple=aarch64 -max-store-memmove=8 | FileCheck %s --check-prefix=MEMMOVE-LIMIT-8
|
|
|
|
declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg)
|
|
declare void @llvm.memcpy.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1)
|
|
declare void @llvm.memmove.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1)
|
|
|
|
; memset tests
|
|
|
|
; Test memset with 128 bytes (the AArch64 backend counts stores in 16-bit)
|
|
; This should be inlined by default and with -max-store-memset=8
|
|
define void @memset_128(ptr %dst) {
|
|
; DEFAULT-LABEL: memset_128:
|
|
; DEFAULT: // %bb.0:
|
|
; DEFAULT-NEXT: movi v0.2d, #0000000000000000
|
|
; DEFAULT-NEXT: stp q0, q0, [x0]
|
|
; DEFAULT-NEXT: stp q0, q0, [x0, #32]
|
|
; DEFAULT-NEXT: stp q0, q0, [x0, #64]
|
|
; DEFAULT-NEXT: stp q0, q0, [x0, #96]
|
|
; DEFAULT-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-7-LABEL: memset_128:
|
|
; MEMSET-LIMIT-7: // %bb.0:
|
|
; MEMSET-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMSET-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMSET-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMSET-LIMIT-7-NEXT: mov w1, wzr
|
|
; MEMSET-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMSET-LIMIT-7-NEXT: bl memset
|
|
; MEMSET-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMSET-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-8-LABEL: memset_128:
|
|
; MEMSET-LIMIT-8: // %bb.0:
|
|
; MEMSET-LIMIT-8-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q0, [x0]
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMSET-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-7-LABEL: memset_128:
|
|
; MEMCPY-LIMIT-7: // %bb.0:
|
|
; MEMCPY-LIMIT-7-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q0, [x0]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMCPY-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-8-LABEL: memset_128:
|
|
; MEMCPY-LIMIT-8: // %bb.0:
|
|
; MEMCPY-LIMIT-8-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q0, [x0]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMCPY-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-7-LABEL: memset_128:
|
|
; MEMMOVE-LIMIT-7: // %bb.0:
|
|
; MEMMOVE-LIMIT-7-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q0, q0, [x0]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMMOVE-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-8-LABEL: memset_128:
|
|
; MEMMOVE-LIMIT-8: // %bb.0:
|
|
; MEMMOVE-LIMIT-8-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q0, [x0]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMMOVE-LIMIT-8-NEXT: ret
|
|
call void @llvm.memset.p0.i64(ptr align 16 %dst, i8 0, i64 128, i1 false)
|
|
ret void
|
|
}
|
|
|
|
; Test memset in a function with optsize attribute
|
|
; With -max-store-memset=8, 128 bytes (8 stores) should still inline
|
|
define void @memset_128_optsize(ptr %dst) optsize {
|
|
; DEFAULT-LABEL: memset_128_optsize:
|
|
; DEFAULT: // %bb.0:
|
|
; DEFAULT-NEXT: movi v0.2d, #0000000000000000
|
|
; DEFAULT-NEXT: stp q0, q0, [x0]
|
|
; DEFAULT-NEXT: stp q0, q0, [x0, #32]
|
|
; DEFAULT-NEXT: stp q0, q0, [x0, #64]
|
|
; DEFAULT-NEXT: stp q0, q0, [x0, #96]
|
|
; DEFAULT-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-7-LABEL: memset_128_optsize:
|
|
; MEMSET-LIMIT-7: // %bb.0:
|
|
; MEMSET-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMSET-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMSET-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMSET-LIMIT-7-NEXT: mov w1, wzr
|
|
; MEMSET-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMSET-LIMIT-7-NEXT: bl memset
|
|
; MEMSET-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMSET-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-8-LABEL: memset_128_optsize:
|
|
; MEMSET-LIMIT-8: // %bb.0:
|
|
; MEMSET-LIMIT-8-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q0, [x0]
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMSET-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-7-LABEL: memset_128_optsize:
|
|
; MEMCPY-LIMIT-7: // %bb.0:
|
|
; MEMCPY-LIMIT-7-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q0, [x0]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMCPY-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-8-LABEL: memset_128_optsize:
|
|
; MEMCPY-LIMIT-8: // %bb.0:
|
|
; MEMCPY-LIMIT-8-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q0, [x0]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMCPY-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-7-LABEL: memset_128_optsize:
|
|
; MEMMOVE-LIMIT-7: // %bb.0:
|
|
; MEMMOVE-LIMIT-7-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q0, q0, [x0]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMMOVE-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-8-LABEL: memset_128_optsize:
|
|
; MEMMOVE-LIMIT-8: // %bb.0:
|
|
; MEMMOVE-LIMIT-8-NEXT: movi v0.2d, #0000000000000000
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q0, [x0]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q0, [x0, #32]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q0, [x0, #64]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q0, [x0, #96]
|
|
; MEMMOVE-LIMIT-8-NEXT: ret
|
|
call void @llvm.memset.p0.i64(ptr align 16 %dst, i8 0, i64 128, i1 false)
|
|
ret void
|
|
}
|
|
|
|
; memcpy tests
|
|
|
|
; Test memcpy with 128 bytes (8 stores of 16 bytes each)
|
|
; This should be inlined by default and with -max-store-memcpy=8
|
|
; but should call memcpy library with -max-store-memcpy=7
|
|
define void @memcpy_128(ptr %dst, ptr %src) {
|
|
; DEFAULT-LABEL: memcpy_128:
|
|
; DEFAULT: // %bb.0:
|
|
; DEFAULT-NEXT: ldp q1, q0, [x1, #32]
|
|
; DEFAULT-NEXT: ldp q2, q3, [x1]
|
|
; DEFAULT-NEXT: stp q1, q0, [x0, #32]
|
|
; DEFAULT-NEXT: stp q2, q3, [x0]
|
|
; DEFAULT-NEXT: ldp q1, q0, [x1, #96]
|
|
; DEFAULT-NEXT: ldp q2, q3, [x1, #64]
|
|
; DEFAULT-NEXT: stp q1, q0, [x0, #96]
|
|
; DEFAULT-NEXT: stp q2, q3, [x0, #64]
|
|
; DEFAULT-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-7-LABEL: memcpy_128:
|
|
; MEMSET-LIMIT-7: // %bb.0:
|
|
; MEMSET-LIMIT-7-NEXT: ldp q1, q0, [x1, #32]
|
|
; MEMSET-LIMIT-7-NEXT: ldp q2, q3, [x1]
|
|
; MEMSET-LIMIT-7-NEXT: stp q1, q0, [x0, #32]
|
|
; MEMSET-LIMIT-7-NEXT: stp q2, q3, [x0]
|
|
; MEMSET-LIMIT-7-NEXT: ldp q1, q0, [x1, #96]
|
|
; MEMSET-LIMIT-7-NEXT: ldp q2, q3, [x1, #64]
|
|
; MEMSET-LIMIT-7-NEXT: stp q1, q0, [x0, #96]
|
|
; MEMSET-LIMIT-7-NEXT: stp q2, q3, [x0, #64]
|
|
; MEMSET-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-8-LABEL: memcpy_128:
|
|
; MEMSET-LIMIT-8: // %bb.0:
|
|
; MEMSET-LIMIT-8-NEXT: ldp q1, q0, [x1, #32]
|
|
; MEMSET-LIMIT-8-NEXT: ldp q2, q3, [x1]
|
|
; MEMSET-LIMIT-8-NEXT: stp q1, q0, [x0, #32]
|
|
; MEMSET-LIMIT-8-NEXT: stp q2, q3, [x0]
|
|
; MEMSET-LIMIT-8-NEXT: ldp q1, q0, [x1, #96]
|
|
; MEMSET-LIMIT-8-NEXT: ldp q2, q3, [x1, #64]
|
|
; MEMSET-LIMIT-8-NEXT: stp q1, q0, [x0, #96]
|
|
; MEMSET-LIMIT-8-NEXT: stp q2, q3, [x0, #64]
|
|
; MEMSET-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-7-LABEL: memcpy_128:
|
|
; MEMCPY-LIMIT-7: // %bb.0:
|
|
; MEMCPY-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMCPY-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMCPY-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMCPY-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMCPY-LIMIT-7-NEXT: bl memcpy
|
|
; MEMCPY-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMCPY-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-8-LABEL: memcpy_128:
|
|
; MEMCPY-LIMIT-8: // %bb.0:
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q1, q0, [x1, #32]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q2, q3, [x1]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q1, q0, [x0, #32]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q2, q3, [x0]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q1, q0, [x1, #96]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q2, q3, [x1, #64]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q1, q0, [x0, #96]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q2, q3, [x0, #64]
|
|
; MEMCPY-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-7-LABEL: memcpy_128:
|
|
; MEMMOVE-LIMIT-7: // %bb.0:
|
|
; MEMMOVE-LIMIT-7-NEXT: ldp q1, q0, [x1, #32]
|
|
; MEMMOVE-LIMIT-7-NEXT: ldp q2, q3, [x1]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q1, q0, [x0, #32]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q2, q3, [x0]
|
|
; MEMMOVE-LIMIT-7-NEXT: ldp q1, q0, [x1, #96]
|
|
; MEMMOVE-LIMIT-7-NEXT: ldp q2, q3, [x1, #64]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q1, q0, [x0, #96]
|
|
; MEMMOVE-LIMIT-7-NEXT: stp q2, q3, [x0, #64]
|
|
; MEMMOVE-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-8-LABEL: memcpy_128:
|
|
; MEMMOVE-LIMIT-8: // %bb.0:
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q1, q0, [x1, #32]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q2, q3, [x1]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q1, q0, [x0, #32]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q2, q3, [x0]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q1, q0, [x1, #96]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q2, q3, [x1, #64]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q1, q0, [x0, #96]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q2, q3, [x0, #64]
|
|
; MEMMOVE-LIMIT-8-NEXT: ret
|
|
call void @llvm.memcpy.p0.p0.i64(ptr %dst, ptr %src, i64 128, i1 false)
|
|
ret void
|
|
}
|
|
|
|
; Test optsize memcpy with 128 bytes (8 stores of 16 bytes each)
|
|
; This should be inlined with -max-store-memcpy=8
|
|
; but should call memcpy library with -max-store-memcpy=7
|
|
define void @memcpy_128_optsize(ptr %dst, ptr %src) optsize {
|
|
; DEFAULT-LABEL: memcpy_128_optsize:
|
|
; DEFAULT: // %bb.0:
|
|
; DEFAULT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; DEFAULT-NEXT: .cfi_def_cfa_offset 16
|
|
; DEFAULT-NEXT: .cfi_offset w30, -16
|
|
; DEFAULT-NEXT: mov w2, #128 // =0x80
|
|
; DEFAULT-NEXT: bl memcpy
|
|
; DEFAULT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; DEFAULT-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-7-LABEL: memcpy_128_optsize:
|
|
; MEMSET-LIMIT-7: // %bb.0:
|
|
; MEMSET-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMSET-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMSET-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMSET-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMSET-LIMIT-7-NEXT: bl memcpy
|
|
; MEMSET-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMSET-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-8-LABEL: memcpy_128_optsize:
|
|
; MEMSET-LIMIT-8: // %bb.0:
|
|
; MEMSET-LIMIT-8-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMSET-LIMIT-8-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMSET-LIMIT-8-NEXT: .cfi_offset w30, -16
|
|
; MEMSET-LIMIT-8-NEXT: mov w2, #128 // =0x80
|
|
; MEMSET-LIMIT-8-NEXT: bl memcpy
|
|
; MEMSET-LIMIT-8-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMSET-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-7-LABEL: memcpy_128_optsize:
|
|
; MEMCPY-LIMIT-7: // %bb.0:
|
|
; MEMCPY-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMCPY-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMCPY-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMCPY-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMCPY-LIMIT-7-NEXT: bl memcpy
|
|
; MEMCPY-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMCPY-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-8-LABEL: memcpy_128_optsize:
|
|
; MEMCPY-LIMIT-8: // %bb.0:
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q1, q0, [x1, #32]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q2, q3, [x1]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q1, q0, [x0, #32]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q2, q3, [x0]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q1, q0, [x1, #96]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q2, q3, [x1, #64]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q1, q0, [x0, #96]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q2, q3, [x0, #64]
|
|
; MEMCPY-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-7-LABEL: memcpy_128_optsize:
|
|
; MEMMOVE-LIMIT-7: // %bb.0:
|
|
; MEMMOVE-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMMOVE-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMMOVE-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMMOVE-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMMOVE-LIMIT-7-NEXT: bl memcpy
|
|
; MEMMOVE-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMMOVE-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-8-LABEL: memcpy_128_optsize:
|
|
; MEMMOVE-LIMIT-8: // %bb.0:
|
|
; MEMMOVE-LIMIT-8-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMMOVE-LIMIT-8-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMMOVE-LIMIT-8-NEXT: .cfi_offset w30, -16
|
|
; MEMMOVE-LIMIT-8-NEXT: mov w2, #128 // =0x80
|
|
; MEMMOVE-LIMIT-8-NEXT: bl memcpy
|
|
; MEMMOVE-LIMIT-8-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMMOVE-LIMIT-8-NEXT: ret
|
|
call void @llvm.memcpy.p0.p0.i64(ptr %dst, ptr %src, i64 128, i1 false)
|
|
ret void
|
|
}
|
|
|
|
; memmove tests
|
|
|
|
; Test memmove with 128 bytes (8 stores of 16 bytes each)
|
|
; This should be inlined by default and with -max-store-memmove=8
|
|
; but should call memmove library with -max-store-memmove=7
|
|
define void @memmove_128(ptr %dst, ptr %src) {
|
|
; DEFAULT-LABEL: memmove_128:
|
|
; DEFAULT: // %bb.0:
|
|
; DEFAULT-NEXT: ldp q0, q1, [x1, #96]
|
|
; DEFAULT-NEXT: ldp q2, q3, [x1]
|
|
; DEFAULT-NEXT: ldp q4, q5, [x1, #32]
|
|
; DEFAULT-NEXT: ldp q6, q7, [x1, #64]
|
|
; DEFAULT-NEXT: stp q0, q1, [x0, #96]
|
|
; DEFAULT-NEXT: stp q2, q3, [x0]
|
|
; DEFAULT-NEXT: stp q4, q5, [x0, #32]
|
|
; DEFAULT-NEXT: stp q6, q7, [x0, #64]
|
|
; DEFAULT-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-7-LABEL: memmove_128:
|
|
; MEMSET-LIMIT-7: // %bb.0:
|
|
; MEMSET-LIMIT-7-NEXT: ldp q0, q1, [x1, #96]
|
|
; MEMSET-LIMIT-7-NEXT: ldp q2, q3, [x1]
|
|
; MEMSET-LIMIT-7-NEXT: ldp q4, q5, [x1, #32]
|
|
; MEMSET-LIMIT-7-NEXT: ldp q6, q7, [x1, #64]
|
|
; MEMSET-LIMIT-7-NEXT: stp q0, q1, [x0, #96]
|
|
; MEMSET-LIMIT-7-NEXT: stp q2, q3, [x0]
|
|
; MEMSET-LIMIT-7-NEXT: stp q4, q5, [x0, #32]
|
|
; MEMSET-LIMIT-7-NEXT: stp q6, q7, [x0, #64]
|
|
; MEMSET-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-8-LABEL: memmove_128:
|
|
; MEMSET-LIMIT-8: // %bb.0:
|
|
; MEMSET-LIMIT-8-NEXT: ldp q0, q1, [x1, #96]
|
|
; MEMSET-LIMIT-8-NEXT: ldp q2, q3, [x1]
|
|
; MEMSET-LIMIT-8-NEXT: ldp q4, q5, [x1, #32]
|
|
; MEMSET-LIMIT-8-NEXT: ldp q6, q7, [x1, #64]
|
|
; MEMSET-LIMIT-8-NEXT: stp q0, q1, [x0, #96]
|
|
; MEMSET-LIMIT-8-NEXT: stp q2, q3, [x0]
|
|
; MEMSET-LIMIT-8-NEXT: stp q4, q5, [x0, #32]
|
|
; MEMSET-LIMIT-8-NEXT: stp q6, q7, [x0, #64]
|
|
; MEMSET-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-7-LABEL: memmove_128:
|
|
; MEMCPY-LIMIT-7: // %bb.0:
|
|
; MEMCPY-LIMIT-7-NEXT: ldp q0, q1, [x1, #96]
|
|
; MEMCPY-LIMIT-7-NEXT: ldp q2, q3, [x1]
|
|
; MEMCPY-LIMIT-7-NEXT: ldp q4, q5, [x1, #32]
|
|
; MEMCPY-LIMIT-7-NEXT: ldp q6, q7, [x1, #64]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q0, q1, [x0, #96]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q2, q3, [x0]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q4, q5, [x0, #32]
|
|
; MEMCPY-LIMIT-7-NEXT: stp q6, q7, [x0, #64]
|
|
; MEMCPY-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-8-LABEL: memmove_128:
|
|
; MEMCPY-LIMIT-8: // %bb.0:
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q0, q1, [x1, #96]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q2, q3, [x1]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q4, q5, [x1, #32]
|
|
; MEMCPY-LIMIT-8-NEXT: ldp q6, q7, [x1, #64]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q0, q1, [x0, #96]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q2, q3, [x0]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q4, q5, [x0, #32]
|
|
; MEMCPY-LIMIT-8-NEXT: stp q6, q7, [x0, #64]
|
|
; MEMCPY-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-7-LABEL: memmove_128:
|
|
; MEMMOVE-LIMIT-7: // %bb.0:
|
|
; MEMMOVE-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMMOVE-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMMOVE-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMMOVE-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMMOVE-LIMIT-7-NEXT: bl memmove
|
|
; MEMMOVE-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMMOVE-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-8-LABEL: memmove_128:
|
|
; MEMMOVE-LIMIT-8: // %bb.0:
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q0, q1, [x1, #96]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q2, q3, [x1]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q4, q5, [x1, #32]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q6, q7, [x1, #64]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q1, [x0, #96]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q2, q3, [x0]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q4, q5, [x0, #32]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q6, q7, [x0, #64]
|
|
; MEMMOVE-LIMIT-8-NEXT: ret
|
|
call void @llvm.memmove.p0.p0.i64(ptr %dst, ptr %src, i64 128, i1 false)
|
|
ret void
|
|
}
|
|
|
|
; Test optsize memmove with 128 bytes (8 stores of 16 bytes each)
|
|
; This should be inlined with -max-store-memmove=8
|
|
; but should call memmove library with -max-store-memmove=7
|
|
define void @memmove_128_optsize(ptr %dst, ptr %src) optsize {
|
|
; DEFAULT-LABEL: memmove_128_optsize:
|
|
; DEFAULT: // %bb.0:
|
|
; DEFAULT-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; DEFAULT-NEXT: .cfi_def_cfa_offset 16
|
|
; DEFAULT-NEXT: .cfi_offset w30, -16
|
|
; DEFAULT-NEXT: mov w2, #128 // =0x80
|
|
; DEFAULT-NEXT: bl memmove
|
|
; DEFAULT-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; DEFAULT-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-7-LABEL: memmove_128_optsize:
|
|
; MEMSET-LIMIT-7: // %bb.0:
|
|
; MEMSET-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMSET-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMSET-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMSET-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMSET-LIMIT-7-NEXT: bl memmove
|
|
; MEMSET-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMSET-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMSET-LIMIT-8-LABEL: memmove_128_optsize:
|
|
; MEMSET-LIMIT-8: // %bb.0:
|
|
; MEMSET-LIMIT-8-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMSET-LIMIT-8-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMSET-LIMIT-8-NEXT: .cfi_offset w30, -16
|
|
; MEMSET-LIMIT-8-NEXT: mov w2, #128 // =0x80
|
|
; MEMSET-LIMIT-8-NEXT: bl memmove
|
|
; MEMSET-LIMIT-8-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMSET-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-7-LABEL: memmove_128_optsize:
|
|
; MEMCPY-LIMIT-7: // %bb.0:
|
|
; MEMCPY-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMCPY-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMCPY-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMCPY-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMCPY-LIMIT-7-NEXT: bl memmove
|
|
; MEMCPY-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMCPY-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMCPY-LIMIT-8-LABEL: memmove_128_optsize:
|
|
; MEMCPY-LIMIT-8: // %bb.0:
|
|
; MEMCPY-LIMIT-8-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMCPY-LIMIT-8-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMCPY-LIMIT-8-NEXT: .cfi_offset w30, -16
|
|
; MEMCPY-LIMIT-8-NEXT: mov w2, #128 // =0x80
|
|
; MEMCPY-LIMIT-8-NEXT: bl memmove
|
|
; MEMCPY-LIMIT-8-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMCPY-LIMIT-8-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-7-LABEL: memmove_128_optsize:
|
|
; MEMMOVE-LIMIT-7: // %bb.0:
|
|
; MEMMOVE-LIMIT-7-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill
|
|
; MEMMOVE-LIMIT-7-NEXT: .cfi_def_cfa_offset 16
|
|
; MEMMOVE-LIMIT-7-NEXT: .cfi_offset w30, -16
|
|
; MEMMOVE-LIMIT-7-NEXT: mov w2, #128 // =0x80
|
|
; MEMMOVE-LIMIT-7-NEXT: bl memmove
|
|
; MEMMOVE-LIMIT-7-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload
|
|
; MEMMOVE-LIMIT-7-NEXT: ret
|
|
;
|
|
; MEMMOVE-LIMIT-8-LABEL: memmove_128_optsize:
|
|
; MEMMOVE-LIMIT-8: // %bb.0:
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q0, q1, [x1, #96]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q2, q3, [x1]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q4, q5, [x1, #32]
|
|
; MEMMOVE-LIMIT-8-NEXT: ldp q6, q7, [x1, #64]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q0, q1, [x0, #96]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q2, q3, [x0]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q4, q5, [x0, #32]
|
|
; MEMMOVE-LIMIT-8-NEXT: stp q6, q7, [x0, #64]
|
|
; MEMMOVE-LIMIT-8-NEXT: ret
|
|
call void @llvm.memmove.p0.p0.i64(ptr %dst, ptr %src, i64 128, i1 false)
|
|
ret void
|
|
}
|