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
3416 lines
124 KiB
LLVM
3416 lines
124 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -enable-subreg-liveness -mattr=+sve -force-streaming-compatible < %s | FileCheck %s
|
|
; RUN: llc -enable-subreg-liveness -mattr=+sme -force-streaming < %s | FileCheck %s
|
|
; RUN: llc -enable-subreg-liveness -force-streaming-compatible < %s | FileCheck %s --check-prefix=NONEON-NOSVE
|
|
|
|
target triple = "aarch64-unknown-linux-gnu"
|
|
|
|
;
|
|
; FCMP OEQ
|
|
;
|
|
|
|
define <2 x i16> @fcmp_oeq_v2f16(<2 x half> %op1, <2 x half> %op2) {
|
|
; CHECK-LABEL: fcmp_oeq_v2f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ptrue p0.h, vl4
|
|
; CHECK-NEXT: fcmeq p0.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: punpklo p0.h, p0.b
|
|
; CHECK-NEXT: mov z0.s, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v2f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #32
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 32
|
|
; NONEON-NOSVE-NEXT: stp d0, d1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr d0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #32
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%cmp = fcmp oeq <2 x half> %op1, %op2
|
|
%sext = sext <2 x i1> %cmp to <2 x i16>
|
|
ret <2 x i16> %sext
|
|
}
|
|
|
|
define <4 x i16> @fcmp_oeq_v4f16(<4 x half> %op1, <4 x half> %op2) {
|
|
; CHECK-LABEL: fcmp_oeq_v4f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ptrue p0.h, vl4
|
|
; CHECK-NEXT: fcmeq p0.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: mov z0.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v4f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #32
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 32
|
|
; NONEON-NOSVE-NEXT: stp d0, d1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr d0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #32
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%cmp = fcmp oeq <4 x half> %op1, %op2
|
|
%sext = sext <4 x i1> %cmp to <4 x i16>
|
|
ret <4 x i16> %sext
|
|
}
|
|
|
|
define <8 x i16> @fcmp_oeq_v8f16(<8 x half> %op1, <8 x half> %op2) {
|
|
; CHECK-LABEL: fcmp_oeq_v8f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: fcmeq p0.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: mov z0.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v8f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [sp, #-48]!
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 48
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #48
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%cmp = fcmp oeq <8 x half> %op1, %op2
|
|
%sext = sext <8 x i1> %cmp to <8 x i16>
|
|
ret <8 x i16> %sext
|
|
}
|
|
|
|
define void @fcmp_oeq_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_oeq_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmeq p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmeq p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp oeq <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
define <2 x i32> @fcmp_oeq_v2f32(<2 x float> %op1, <2 x float> %op2) {
|
|
; CHECK-LABEL: fcmp_oeq_v2f32:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ptrue p0.s, vl2
|
|
; CHECK-NEXT: fcmeq p0.s, p0/z, z0.s, z1.s
|
|
; CHECK-NEXT: mov z0.s, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v2f32:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #32
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 32
|
|
; NONEON-NOSVE-NEXT: stp d0, d1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: ldp s0, s2, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldp s1, s3, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcmp s3, s2
|
|
; NONEON-NOSVE-NEXT: csetm w9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr d0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #32
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%cmp = fcmp oeq <2 x float> %op1, %op2
|
|
%sext = sext <2 x i1> %cmp to <2 x i32>
|
|
ret <2 x i32> %sext
|
|
}
|
|
|
|
define <4 x i32> @fcmp_oeq_v4f32(<4 x float> %op1, <4 x float> %op2) {
|
|
; CHECK-LABEL: fcmp_oeq_v4f32:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ptrue p0.s, vl4
|
|
; CHECK-NEXT: fcmeq p0.s, p0/z, z0.s, z1.s
|
|
; CHECK-NEXT: mov z0.s, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v4f32:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [sp, #-48]!
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 48
|
|
; NONEON-NOSVE-NEXT: ldp s0, s2, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldp s1, s3, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcmp s3, s2
|
|
; NONEON-NOSVE-NEXT: csetm w9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldp s0, s2, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldp s1, s3, [sp]
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcmp s3, s2
|
|
; NONEON-NOSVE-NEXT: csetm w9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #48
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%cmp = fcmp oeq <4 x float> %op1, %op2
|
|
%sext = sext <4 x i1> %cmp to <4 x i32>
|
|
ret <4 x i32> %sext
|
|
}
|
|
|
|
define void @fcmp_oeq_v8f32(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_oeq_v8f32:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.s, vl4
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmeq p1.s, p0/z, z1.s, z0.s
|
|
; CHECK-NEXT: fcmeq p0.s, p0/z, z2.s, z3.s
|
|
; CHECK-NEXT: mov z0.s, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.s, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v8f32:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldp s0, s2, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldp s1, s3, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcmp s3, s2
|
|
; NONEON-NOSVE-NEXT: csetm w9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldp s0, s2, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldp s1, s3, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s3, s2
|
|
; NONEON-NOSVE-NEXT: csetm w9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldp s0, s2, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldp s1, s3, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s3, s2
|
|
; NONEON-NOSVE-NEXT: csetm w9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldp s0, s2, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldp s1, s3, [sp]
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s3, s2
|
|
; NONEON-NOSVE-NEXT: csetm w9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <8 x float>, ptr %a
|
|
%op2 = load <8 x float>, ptr %b
|
|
%cmp = fcmp oeq <8 x float> %op1, %op2
|
|
%sext = sext <8 x i1> %cmp to <8 x i32>
|
|
store <8 x i32> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
define <1 x i64> @fcmp_oeq_v1f64(<1 x double> %op1, <1 x double> %op2) {
|
|
; CHECK-LABEL: fcmp_oeq_v1f64:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: fcmp d0, d1
|
|
; CHECK-NEXT: csetm x8, eq
|
|
; CHECK-NEXT: mov z0.d, x8
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v1f64:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #16
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 16
|
|
; NONEON-NOSVE-NEXT: fcmp d0, d1
|
|
; NONEON-NOSVE-NEXT: csetm x8, eq
|
|
; NONEON-NOSVE-NEXT: fmov d0, x8
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #16
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%cmp = fcmp oeq <1 x double> %op1, %op2
|
|
%sext = sext <1 x i1> %cmp to <1 x i64>
|
|
ret <1 x i64> %sext
|
|
}
|
|
|
|
define <2 x i64> @fcmp_oeq_v2f64(<2 x double> %op1, <2 x double> %op2) {
|
|
; CHECK-LABEL: fcmp_oeq_v2f64:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ptrue p0.d, vl2
|
|
; CHECK-NEXT: fcmeq p0.d, p0/z, z0.d, z1.d
|
|
; CHECK-NEXT: mov z0.d, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v2f64:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [sp, #-48]!
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 48
|
|
; NONEON-NOSVE-NEXT: ldp d0, d2, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldp d1, d3, [sp]
|
|
; NONEON-NOSVE-NEXT: fcmp d3, d2
|
|
; NONEON-NOSVE-NEXT: csetm x9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp d1, d0
|
|
; NONEON-NOSVE-NEXT: csetm x8, eq
|
|
; NONEON-NOSVE-NEXT: stp x8, x9, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #48
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%cmp = fcmp oeq <2 x double> %op1, %op2
|
|
%sext = sext <2 x i1> %cmp to <2 x i64>
|
|
ret <2 x i64> %sext
|
|
}
|
|
|
|
define void @fcmp_oeq_v4f64(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_oeq_v4f64:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.d, vl2
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmeq p1.d, p0/z, z1.d, z0.d
|
|
; CHECK-NEXT: fcmeq p0.d, p0/z, z2.d, z3.d
|
|
; CHECK-NEXT: mov z0.d, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.d, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oeq_v4f64:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldp d0, d2, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldp d1, d3, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcmp d3, d2
|
|
; NONEON-NOSVE-NEXT: csetm x9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp d1, d0
|
|
; NONEON-NOSVE-NEXT: ldp d0, d2, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldp d1, d3, [sp]
|
|
; NONEON-NOSVE-NEXT: csetm x8, eq
|
|
; NONEON-NOSVE-NEXT: stp x8, x9, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp d3, d2
|
|
; NONEON-NOSVE-NEXT: csetm x9, eq
|
|
; NONEON-NOSVE-NEXT: fcmp d1, d0
|
|
; NONEON-NOSVE-NEXT: csetm x8, eq
|
|
; NONEON-NOSVE-NEXT: stp x8, x9, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <4 x double>, ptr %a
|
|
%op2 = load <4 x double>, ptr %b
|
|
%cmp = fcmp oeq <4 x double> %op1, %op2
|
|
%sext = sext <4 x i1> %cmp to <4 x i64>
|
|
store <4 x i64> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP UEQ
|
|
;
|
|
|
|
define void @fcmp_ueq_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ueq_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q3, q0, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q2, q1, [x0]
|
|
; CHECK-NEXT: fcmuo p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmuo p2.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: fcmeq p3.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: fcmeq p0.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: mov p2.b, p3/m, p3.b
|
|
; CHECK-NEXT: sel p0.b, p0, p0.b, p1.b
|
|
; CHECK-NEXT: mov z0.h, p2/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ueq_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp ueq <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP ONE
|
|
;
|
|
|
|
define void @fcmp_one_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_one_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q3, q0, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q2, q1, [x0]
|
|
; CHECK-NEXT: fcmgt p1.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: fcmgt p2.h, p0/z, z3.h, z2.h
|
|
; CHECK-NEXT: fcmgt p3.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: fcmgt p0.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: mov p2.b, p3/m, p3.b
|
|
; CHECK-NEXT: sel p0.b, p0, p0.b, p1.b
|
|
; CHECK-NEXT: mov z0.h, p2/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_one_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: csinv w8, w8, wzr, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp one <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP UNE
|
|
;
|
|
|
|
define void @fcmp_une_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_une_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmne p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmne p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_une_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp une <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP OGT
|
|
;
|
|
|
|
define void @fcmp_ogt_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ogt_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmgt p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmgt p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ogt_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp ogt <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP UGT
|
|
;
|
|
|
|
define void @fcmp_ugt_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ugt_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmge p1.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: fcmge p0.h, p0/z, z3.h, z2.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: subr z0.b, z0.b, #255 // =0xff
|
|
; CHECK-NEXT: subr z1.b, z1.b, #255 // =0xff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ugt_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, hi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp ugt <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP OLT
|
|
;
|
|
|
|
define void @fcmp_olt_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_olt_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmgt p1.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: fcmgt p0.h, p0/z, z3.h, z2.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_olt_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, mi
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp olt <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP ULT
|
|
;
|
|
|
|
define void @fcmp_ult_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ult_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmge p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmge p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: subr z0.b, z0.b, #255 // =0xff
|
|
; CHECK-NEXT: subr z1.b, z1.b, #255 // =0xff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ult_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp ult <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP OGE
|
|
;
|
|
|
|
define void @fcmp_oge_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_oge_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmge p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmge p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_oge_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp oge <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP UGE
|
|
;
|
|
|
|
define void @fcmp_uge_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_uge_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmgt p1.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: fcmgt p0.h, p0/z, z3.h, z2.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: subr z0.b, z0.b, #255 // =0xff
|
|
; CHECK-NEXT: subr z1.b, z1.b, #255 // =0xff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_uge_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, pl
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp uge <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP OLE
|
|
;
|
|
|
|
define void @fcmp_ole_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ole_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmge p1.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: fcmge p0.h, p0/z, z3.h, z2.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ole_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, ls
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp ole <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP ULE
|
|
;
|
|
|
|
define void @fcmp_ule_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ule_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmgt p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmgt p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: subr z0.b, z0.b, #255 // =0xff
|
|
; CHECK-NEXT: subr z1.b, z1.b, #255 // =0xff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ule_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp ule <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP UNO
|
|
;
|
|
|
|
define void @fcmp_uno_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_uno_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmuo p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmuo p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_uno_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, vs
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp uno <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP ORD
|
|
;
|
|
|
|
define void @fcmp_ord_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ord_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmuo p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmuo p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: subr z0.b, z0.b, #255 // =0xff
|
|
; CHECK-NEXT: subr z1.b, z1.b, #255 // =0xff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ord_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, vc
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp ord <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP EQ
|
|
;
|
|
|
|
define void @fcmp_eq_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_eq_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmeq p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmeq p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_eq_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, eq
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp fast oeq <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP NE
|
|
;
|
|
|
|
define void @fcmp_ne_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ne_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmne p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmne p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ne_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, ne
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp fast one <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP GT
|
|
;
|
|
|
|
define void @fcmp_gt_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_gt_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmgt p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmgt p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_gt_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, gt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp fast ogt <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP LT
|
|
;
|
|
|
|
define void @fcmp_lt_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_lt_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmgt p1.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: fcmgt p0.h, p0/z, z3.h, z2.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_lt_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, lt
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp fast olt <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP GE
|
|
;
|
|
|
|
define void @fcmp_ge_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_ge_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmge p1.h, p0/z, z1.h, z0.h
|
|
; CHECK-NEXT: fcmge p0.h, p0/z, z2.h, z3.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_ge_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, ge
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp fast oge <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|
|
|
|
;
|
|
; FCMP LE
|
|
;
|
|
|
|
define void @fcmp_le_v16f16(ptr %a, ptr %b, ptr %c) {
|
|
; CHECK-LABEL: fcmp_le_v16f16:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: ldp q0, q3, [x1]
|
|
; CHECK-NEXT: ptrue p0.h, vl8
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: fcmge p1.h, p0/z, z0.h, z1.h
|
|
; CHECK-NEXT: fcmge p0.h, p0/z, z3.h, z2.h
|
|
; CHECK-NEXT: mov z0.h, p1/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: mov z1.h, p0/z, #-1 // =0xffffffffffffffff
|
|
; CHECK-NEXT: stp q0, q1, [x2]
|
|
; CHECK-NEXT: ret
|
|
;
|
|
; NONEON-NOSVE-LABEL: fcmp_le_v16f16:
|
|
; NONEON-NOSVE: // %bb.0:
|
|
; NONEON-NOSVE-NEXT: sub sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 96
|
|
; NONEON-NOSVE-NEXT: ldp q3, q0, [x1]
|
|
; NONEON-NOSVE-NEXT: ldp q2, q1, [x0]
|
|
; NONEON-NOSVE-NEXT: stp q2, q3, [sp]
|
|
; NONEON-NOSVE-NEXT: stp q1, q0, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #62]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #46]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #60]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #44]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #94]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #58]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #42]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #92]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #56]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #40]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #90]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #54]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #38]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #88]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #52]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #36]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #86]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #50]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #34]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #84]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #48]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #32]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #82]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #30]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #14]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #80]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #28]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #12]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #78]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #26]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #10]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #76]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #24]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #8]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #74]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #22]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #6]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #72]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #20]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #4]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #70]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #18]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp, #2]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #68]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: ldr h0, [sp, #16]
|
|
; NONEON-NOSVE-NEXT: ldr h1, [sp]
|
|
; NONEON-NOSVE-NEXT: fcvt s0, h0
|
|
; NONEON-NOSVE-NEXT: fcvt s1, h1
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #66]
|
|
; NONEON-NOSVE-NEXT: fcmp s1, s0
|
|
; NONEON-NOSVE-NEXT: csetm w8, le
|
|
; NONEON-NOSVE-NEXT: strh w8, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: ldp q0, q1, [sp, #64]
|
|
; NONEON-NOSVE-NEXT: stp q0, q1, [x2]
|
|
; NONEON-NOSVE-NEXT: add sp, sp, #96
|
|
; NONEON-NOSVE-NEXT: ret
|
|
%op1 = load <16 x half>, ptr %a
|
|
%op2 = load <16 x half>, ptr %b
|
|
%cmp = fcmp fast ole <16 x half> %op1, %op2
|
|
%sext = sext <16 x i1> %cmp to <16 x i16>
|
|
store <16 x i16> %sext, ptr %c
|
|
ret void
|
|
}
|