derek/TheRock
1
0 Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity
Files
1a182e2dac58879f215c0fad0f5f78c76dcafe56
TheRock/compiler/amd-llvm/llvm/test/CodeGen/AArch64/scvtf-div-mul-combine.ll
T
derek.holloway 7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
Details
CodeQL / Analyze (python) (push) Canceled after 0s
Details
pre-commit / pre-commit (push) Canceled after 0s
Details
Multi-Arch CI / setup (push) Canceled after 0s
Details
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Details
Setup (Multi-Arch) / setup (push) Canceled after 0s
Details
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Details
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
Details
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Details
Multi-Arch CI / CI Summary (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Details
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Details
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
Details
Flatten subprojects into a single base monorepo
2026-09-02 16:34:47 -07:00

475 lines
16 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-neon-syntax=apple -mattr=+fullfp16 -o - %s | FileCheck %s --check-prefixes=CHECK,CHECK-SD
; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-neon-syntax=apple -mattr=+fullfp16 -global-isel -o - %s | FileCheck %s --check-prefixes=CHECK,CHECK-GI
;; ---- sitofp + fdiv (power-of-2 divisors) ----
define <2 x float> @sitofp_v2f32_fdiv_2(<2 x i32> %in) {
; CHECK-SD-LABEL: sitofp_v2f32_fdiv_2:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: scvtf.2s v0, v0, #1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v2f32_fdiv_2:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.2s v1, #64, lsl #24
; CHECK-GI-NEXT: scvtf.2s v0, v0
; CHECK-GI-NEXT: fdiv.2s v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = sitofp <2 x i32> %in to <2 x float>
%div = fdiv <2 x float> %conv, <float 2.0, float 2.0>
ret <2 x float> %div
}
define <4 x float> @sitofp_v4f32_fdiv_4(<4 x i32> %in) {
; CHECK-SD-LABEL: sitofp_v4f32_fdiv_4:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: scvtf.4s v0, v0, #2
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v4f32_fdiv_4:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: fmov.4s v1, #4.00000000
; CHECK-GI-NEXT: scvtf.4s v0, v0
; CHECK-GI-NEXT: fdiv.4s v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = sitofp <4 x i32> %in to <4 x float>
%div = fdiv <4 x float> %conv, <float 4.0, float 4.0, float 4.0, float 4.0>
ret <4 x float> %div
}
define <2 x double> @sitofp_v2f64_fdiv_16(<2 x i64> %in) {
; CHECK-SD-LABEL: sitofp_v2f64_fdiv_16:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: scvtf.2d v0, v0, #4
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v2f64_fdiv_16:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: fmov.2d v1, #16.00000000
; CHECK-GI-NEXT: scvtf.2d v0, v0
; CHECK-GI-NEXT: fdiv.2d v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = sitofp <2 x i64> %in to <2 x double>
%div = fdiv <2 x double> %conv, <double 16.0, double 16.0>
ret <2 x double> %div
}
define <4 x half> @sitofp_v4f16_fdiv_8(<4 x i16> %in) {
; CHECK-SD-LABEL: sitofp_v4f16_fdiv_8:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: scvtf.4h v0, v0, #3
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v4f16_fdiv_8:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.4h v1, #72, lsl #8
; CHECK-GI-NEXT: scvtf.4h v0, v0
; CHECK-GI-NEXT: fdiv.4h v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = sitofp <4 x i16> %in to <4 x half>
%div = fdiv <4 x half> %conv, <half 8.0, half 8.0, half 8.0, half 8.0>
ret <4 x half> %div
}
define <8 x half> @sitofp_v8f16_fdiv_4(<8 x i16> %in) {
; CHECK-SD-LABEL: sitofp_v8f16_fdiv_4:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: scvtf.8h v0, v0, #2
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v8f16_fdiv_4:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.8h v1, #68, lsl #8
; CHECK-GI-NEXT: scvtf.8h v0, v0
; CHECK-GI-NEXT: fdiv.8h v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = sitofp <8 x i16> %in to <8 x half>
%div = fdiv <8 x half> %conv, <half 4.0, half 4.0, half 4.0, half 4.0, half 4.0, half 4.0, half 4.0, half 4.0>
ret <8 x half> %div
}
;; ---- sitofp + fmul (reciprocal power-of-2 multipliers) ----
define <2 x float> @sitofp_v2f32_fmul_half(<2 x i32> %in) {
; CHECK-LABEL: sitofp_v2f32_fmul_half:
; CHECK: // %bb.0:
; CHECK-NEXT: scvtf.2s v0, v0, #1
; CHECK-NEXT: ret
%conv = sitofp <2 x i32> %in to <2 x float>
%mul = fmul <2 x float> %conv, <float 0.5, float 0.5>
ret <2 x float> %mul
}
define <4 x float> @sitofp_v4f32_fmul_quarter(<4 x i32> %in) {
; CHECK-LABEL: sitofp_v4f32_fmul_quarter:
; CHECK: // %bb.0:
; CHECK-NEXT: scvtf.4s v0, v0, #2
; CHECK-NEXT: ret
%conv = sitofp <4 x i32> %in to <4 x float>
%mul = fmul <4 x float> %conv, <float 0.25, float 0.25, float 0.25, float 0.25>
ret <4 x float> %mul
}
define <2 x double> @sitofp_v2f64_fmul_eighth(<2 x i64> %in) {
; CHECK-LABEL: sitofp_v2f64_fmul_eighth:
; CHECK: // %bb.0:
; CHECK-NEXT: scvtf.2d v0, v0, #3
; CHECK-NEXT: ret
%conv = sitofp <2 x i64> %in to <2 x double>
%mul = fmul <2 x double> %conv, <double 0.125, double 0.125>
ret <2 x double> %mul
}
define <4 x half> @sitofp_v4f16_fmul_half(<4 x i16> %in) {
; CHECK-LABEL: sitofp_v4f16_fmul_half:
; CHECK: // %bb.0:
; CHECK-NEXT: scvtf.4h v0, v0, #1
; CHECK-NEXT: ret
%conv = sitofp <4 x i16> %in to <4 x half>
%mul = fmul <4 x half> %conv, <half 0xH3800, half 0xH3800, half 0xH3800, half 0xH3800>
ret <4 x half> %mul
}
define <8 x half> @sitofp_v8f16_fmul_quarter(<8 x i16> %in) {
; CHECK-LABEL: sitofp_v8f16_fmul_quarter:
; CHECK: // %bb.0:
; CHECK-NEXT: scvtf.8h v0, v0, #2
; CHECK-NEXT: ret
%conv = sitofp <8 x i16> %in to <8 x half>
%mul = fmul <8 x half> %conv, <half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400>
ret <8 x half> %mul
}
;; ---- uitofp + fdiv (unsigned) ----
define <2 x float> @uitofp_v2f32_fdiv_2(<2 x i32> %in) {
; CHECK-SD-LABEL: uitofp_v2f32_fdiv_2:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: ucvtf.2s v0, v0, #1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: uitofp_v2f32_fdiv_2:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.2s v1, #64, lsl #24
; CHECK-GI-NEXT: ucvtf.2s v0, v0
; CHECK-GI-NEXT: fdiv.2s v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = uitofp <2 x i32> %in to <2 x float>
%div = fdiv <2 x float> %conv, <float 2.0, float 2.0>
ret <2 x float> %div
}
define <4 x float> @uitofp_v4f32_fdiv_4(<4 x i32> %in) {
; CHECK-SD-LABEL: uitofp_v4f32_fdiv_4:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: ucvtf.4s v0, v0, #2
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: uitofp_v4f32_fdiv_4:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: fmov.4s v1, #4.00000000
; CHECK-GI-NEXT: ucvtf.4s v0, v0
; CHECK-GI-NEXT: fdiv.4s v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = uitofp <4 x i32> %in to <4 x float>
%div = fdiv <4 x float> %conv, <float 4.0, float 4.0, float 4.0, float 4.0>
ret <4 x float> %div
}
define <2 x double> @uitofp_v2f64_fdiv_16(<2 x i64> %in) {
; CHECK-SD-LABEL: uitofp_v2f64_fdiv_16:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: ucvtf.2d v0, v0, #4
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: uitofp_v2f64_fdiv_16:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: fmov.2d v1, #16.00000000
; CHECK-GI-NEXT: ucvtf.2d v0, v0
; CHECK-GI-NEXT: fdiv.2d v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = uitofp <2 x i64> %in to <2 x double>
%div = fdiv <2 x double> %conv, <double 16.0, double 16.0>
ret <2 x double> %div
}
define <4 x half> @uitofp_v4f16_fdiv_8(<4 x i16> %in) {
; CHECK-SD-LABEL: uitofp_v4f16_fdiv_8:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: ucvtf.4h v0, v0, #3
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: uitofp_v4f16_fdiv_8:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.4h v1, #72, lsl #8
; CHECK-GI-NEXT: ucvtf.4h v0, v0
; CHECK-GI-NEXT: fdiv.4h v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = uitofp <4 x i16> %in to <4 x half>
%div = fdiv <4 x half> %conv, <half 8.0, half 8.0, half 8.0, half 8.0>
ret <4 x half> %div
}
define <8 x half> @uitofp_v8f16_fdiv_4(<8 x i16> %in) {
; CHECK-SD-LABEL: uitofp_v8f16_fdiv_4:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: ucvtf.8h v0, v0, #2
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: uitofp_v8f16_fdiv_4:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.8h v1, #68, lsl #8
; CHECK-GI-NEXT: ucvtf.8h v0, v0
; CHECK-GI-NEXT: fdiv.8h v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = uitofp <8 x i16> %in to <8 x half>
%div = fdiv <8 x half> %conv, <half 4.0, half 4.0, half 4.0, half 4.0, half 4.0, half 4.0, half 4.0, half 4.0>
ret <8 x half> %div
}
;; ---- uitofp + fmul (unsigned, reciprocal) ----
define <2 x float> @uitofp_v2f32_fmul_half(<2 x i32> %in) {
; CHECK-LABEL: uitofp_v2f32_fmul_half:
; CHECK: // %bb.0:
; CHECK-NEXT: ucvtf.2s v0, v0, #1
; CHECK-NEXT: ret
%conv = uitofp <2 x i32> %in to <2 x float>
%mul = fmul <2 x float> %conv, <float 0.5, float 0.5>
ret <2 x float> %mul
}
define <4 x float> @uitofp_v4f32_fmul_quarter(<4 x i32> %in) {
; CHECK-LABEL: uitofp_v4f32_fmul_quarter:
; CHECK: // %bb.0:
; CHECK-NEXT: ucvtf.4s v0, v0, #2
; CHECK-NEXT: ret
%conv = uitofp <4 x i32> %in to <4 x float>
%mul = fmul <4 x float> %conv, <float 0.25, float 0.25, float 0.25, float 0.25>
ret <4 x float> %mul
}
define <2 x double> @uitofp_v2f64_fmul_eighth(<2 x i64> %in) {
; CHECK-LABEL: uitofp_v2f64_fmul_eighth:
; CHECK: // %bb.0:
; CHECK-NEXT: ucvtf.2d v0, v0, #3
; CHECK-NEXT: ret
%conv = uitofp <2 x i64> %in to <2 x double>
%mul = fmul <2 x double> %conv, <double 0.125, double 0.125>
ret <2 x double> %mul
}
define <4 x half> @uitofp_v4f16_fmul_half(<4 x i16> %in) {
; CHECK-LABEL: uitofp_v4f16_fmul_half:
; CHECK: // %bb.0:
; CHECK-NEXT: ucvtf.4h v0, v0, #1
; CHECK-NEXT: ret
%conv = uitofp <4 x i16> %in to <4 x half>
%mul = fmul <4 x half> %conv, <half 0xH3800, half 0xH3800, half 0xH3800, half 0xH3800>
ret <4 x half> %mul
}
define <8 x half> @uitofp_v8f16_fmul_quarter(<8 x i16> %in) {
; CHECK-LABEL: uitofp_v8f16_fmul_quarter:
; CHECK: // %bb.0:
; CHECK-NEXT: ucvtf.8h v0, v0, #2
; CHECK-NEXT: ret
%conv = uitofp <8 x i16> %in to <8 x half>
%mul = fmul <8 x half> %conv, <half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400, half 0xH3400>
ret <8 x half> %mul
}
;; ---- boundary: max valid shift per element type ----
define <2 x float> @sitofp_v2f32_fdiv_max(<2 x i32> %in) {
; CHECK-SD-LABEL: sitofp_v2f32_fdiv_max:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: scvtf.2s v0, v0, #32
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v2f32_fdiv_max:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: scvtf.2s v0, v0
; CHECK-GI-NEXT: adrp x8, .LCPI20_0
; CHECK-GI-NEXT: ldr d1, [x8, :lo12:.LCPI20_0]
; CHECK-GI-NEXT: fdiv.2s v0, v0, v1
; CHECK-GI-NEXT: ret
; 2^32 = 4294967296.0 — max valid shift for i32
%conv = sitofp <2 x i32> %in to <2 x float>
%div = fdiv <2 x float> %conv, <float 4294967296.0, float 4294967296.0>
ret <2 x float> %div
}
define <2 x double> @sitofp_v2f64_fdiv_max(<2 x i64> %in) {
; CHECK-SD-LABEL: sitofp_v2f64_fdiv_max:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: scvtf.2d v0, v0, #64
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v2f64_fdiv_max:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: scvtf.2d v0, v0
; CHECK-GI-NEXT: adrp x8, .LCPI21_0
; CHECK-GI-NEXT: ldr q1, [x8, :lo12:.LCPI21_0]
; CHECK-GI-NEXT: fdiv.2d v0, v0, v1
; CHECK-GI-NEXT: ret
; 2^64 — max valid shift for i64
%conv = sitofp <2 x i64> %in to <2 x double>
%div = fdiv <2 x double> %conv, <double 18446744073709551616.0, double 18446744073709551616.0>
ret <2 x double> %div
}
define <4 x half> @sitofp_v4f16_fdiv_max(<4 x i16> %in) {
; CHECK-SD-LABEL: sitofp_v4f16_fdiv_max:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: movi d1, #0000000000000000
; CHECK-SD-NEXT: scvtf.4h v0, v0
; CHECK-SD-NEXT: fmul.4h v0, v0, v1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v4f16_fdiv_max:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.4h v1, #124, lsl #8
; CHECK-GI-NEXT: scvtf.4h v0, v0
; CHECK-GI-NEXT: fdiv.4h v0, v0, v1
; CHECK-GI-NEXT: ret
; 2^16 = 65536.0 — max valid shift for i16 (but not representable in f16)
; f16 max is 65504, so 65536.0 overflows to inf. This should NOT match.
%conv = sitofp <4 x i16> %in to <4 x half>
%div = fdiv <4 x half> %conv, <half 0xH7C00, half 0xH7C00, half 0xH7C00, half 0xH7C00>
ret <4 x half> %div
}
define <4 x half> @sitofp_v4f16_fdiv_1024(<4 x i16> %in) {
; CHECK-SD-LABEL: sitofp_v4f16_fdiv_1024:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: scvtf.4h v0, v0, #10
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: sitofp_v4f16_fdiv_1024:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.4h v1, #100, lsl #8
; CHECK-GI-NEXT: scvtf.4h v0, v0
; CHECK-GI-NEXT: fdiv.4h v0, v0, v1
; CHECK-GI-NEXT: ret
; 2^10 = 1024.0 — shift=10, near upper end of f16 range
%conv = sitofp <4 x i16> %in to <4 x half>
%div = fdiv <4 x half> %conv, <half 1024.0, half 1024.0, half 1024.0, half 1024.0>
ret <4 x half> %div
}
;; ---- negative: out-of-range shift ----
define <2 x float> @neg_sitofp_v2f32_fdiv_too_large(<2 x i32> %in) {
; CHECK-SD-LABEL: neg_sitofp_v2f32_fdiv_too_large:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: movi.2s v1, #47, lsl #24
; CHECK-SD-NEXT: scvtf.2s v0, v0
; CHECK-SD-NEXT: fmul.2s v0, v0, v1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: neg_sitofp_v2f32_fdiv_too_large:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: movi.2s v1, #80, lsl #24
; CHECK-GI-NEXT: scvtf.2s v0, v0
; CHECK-GI-NEXT: fdiv.2s v0, v0, v1
; CHECK-GI-NEXT: ret
; 2^33 — exceeds i32 range, should NOT match
%conv = sitofp <2 x i32> %in to <2 x float>
%div = fdiv <2 x float> %conv, <float 8589934592.0, float 8589934592.0>
ret <2 x float> %div
}
define <2 x double> @neg_sitofp_v2f64_fdiv_too_large(<2 x i64> %in) {
; CHECK-SD-LABEL: neg_sitofp_v2f64_fdiv_too_large:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: mov x8, #4314448443020935168 // =0x3be0000000000000
; CHECK-SD-NEXT: scvtf.2d v0, v0
; CHECK-SD-NEXT: dup.2d v1, x8
; CHECK-SD-NEXT: fmul.2d v0, v0, v1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: neg_sitofp_v2f64_fdiv_too_large:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: scvtf.2d v0, v0
; CHECK-GI-NEXT: adrp x8, .LCPI25_0
; CHECK-GI-NEXT: ldr q1, [x8, :lo12:.LCPI25_0]
; CHECK-GI-NEXT: fdiv.2d v0, v0, v1
; CHECK-GI-NEXT: ret
; 2^65 — exceeds i64 range, should NOT match
%conv = sitofp <2 x i64> %in to <2 x double>
%div = fdiv <2 x double> %conv, <double 36893488147419103232.0, double 36893488147419103232.0>
ret <2 x double> %div
}
;; ---- negative: non-power-of-2 ----
define <2 x float> @neg_v2f32_fmul_non_pow2(<2 x i32> %in) {
; CHECK-SD-LABEL: neg_v2f32_fmul_non_pow2:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: mov w8, #36704 // =0x8f60
; CHECK-SD-NEXT: scvtf.2s v0, v0
; CHECK-SD-NEXT: movk w8, #9666, lsl #16
; CHECK-SD-NEXT: dup.2s v1, w8
; CHECK-SD-NEXT: fmul.2s v0, v0, v1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: neg_v2f32_fmul_non_pow2:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: scvtf.2s v0, v0
; CHECK-GI-NEXT: adrp x8, .LCPI26_0
; CHECK-GI-NEXT: ldr d1, [x8, :lo12:.LCPI26_0]
; CHECK-GI-NEXT: fmul.2s v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = sitofp <2 x i32> %in to <2 x float>
%mul = fmul <2 x float> %conv, <float 0x3CB851EC00000000, float 0x3CB851EC00000000>
ret <2 x float> %mul
}
define <4 x float> @neg_v4f32_fmul_non_pow2(<4 x i32> %in) {
; CHECK-SD-LABEL: neg_v4f32_fmul_non_pow2:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: mov w8, #36704 // =0x8f60
; CHECK-SD-NEXT: scvtf.4s v0, v0
; CHECK-SD-NEXT: movk w8, #9666, lsl #16
; CHECK-SD-NEXT: dup.4s v1, w8
; CHECK-SD-NEXT: fmul.4s v0, v0, v1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: neg_v4f32_fmul_non_pow2:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: scvtf.4s v0, v0
; CHECK-GI-NEXT: adrp x8, .LCPI27_0
; CHECK-GI-NEXT: ldr q1, [x8, :lo12:.LCPI27_0]
; CHECK-GI-NEXT: fmul.4s v0, v0, v1
; CHECK-GI-NEXT: ret
%conv = sitofp <4 x i32> %in to <4 x float>
%mul = fmul <4 x float> %conv, <float 0x3CB851EC00000000, float 0x3CB851EC00000000, float 0x3CB851EC00000000, float 0x3CB851EC00000000>
ret <4 x float> %mul
}
define <2 x double> @neg_v2f64_fmul_non_pow2(<2 x i64> %in) {
; CHECK-SD-LABEL: neg_v2f64_fmul_non_pow2:
; CHECK-SD: // %bb.0:
; CHECK-SD-NEXT: mov x8, #5243 // =0x147b
; CHECK-SD-NEXT: scvtf.2d v0, v0
; CHECK-SD-NEXT: movk x8, #18350, lsl #16
; CHECK-SD-NEXT: movk x8, #31457, lsl #32
; CHECK-SD-NEXT: movk x8, #16276, lsl #48
; CHECK-SD-NEXT: dup.2d v1, x8
; CHECK-SD-NEXT: fmul.2d v0, v0, v1
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: neg_v2f64_fmul_non_pow2:
; CHECK-GI: // %bb.0:
; CHECK-GI-NEXT: scvtf.2d v0, v0
; CHECK-GI-NEXT: adrp x8, .LCPI28_0
; CHECK-GI-NEXT: ldr d1, [x8, :lo12:.LCPI28_0]
; CHECK-GI-NEXT: fmul.2d v0, v0, v1[0]
; CHECK-GI-NEXT: ret
%conv = sitofp <2 x i64> %in to <2 x double>
%mul = fmul <2 x double> %conv, <double 0.02, double 0.02>
ret <2 x double> %mul
}
Reference in new issue View Git Blame Copy Permalink