From 90c397fc56b7a04dd53cdad8103de1ead9686104 Mon Sep 17 00:00:00 2001 From: David Green Date: Sun, 31 Dec 2023 18:45:01 +0000 Subject: [PATCH] [AArch64] Add icmp and fcmp tests for GlobalISel. NFC --- llvm/test/CodeGen/AArch64/fcmp.ll | 1856 +++++++++++++++++++++++++++-- llvm/test/CodeGen/AArch64/icmp.ll | 257 ++++ 2 files changed, 2034 insertions(+), 79 deletions(-) create mode 100644 llvm/test/CodeGen/AArch64/icmp.ll diff --git a/llvm/test/CodeGen/AArch64/fcmp.ll b/llvm/test/CodeGen/AArch64/fcmp.ll index 3c74508bb12b..54ab4af413f9 100644 --- a/llvm/test/CodeGen/AArch64/fcmp.ll +++ b/llvm/test/CodeGen/AArch64/fcmp.ll @@ -1,81 +1,1779 @@ -; RUN: llc -verify-machineinstrs -o - %s -mtriple=aarch64-none-linux-gnu | FileCheck %s - -declare void @bar(i32) - -define void @test_float(float %a, float %b) { -; CHECK-LABEL: test_float: - - %tst1 = fcmp oeq float %a, %b - br i1 %tst1, label %end, label %t2 -; CHECK: fcmp {{s[0-9]+}}, {{s[0-9]+}} -; CHECK: b.eq .L - -t2: - %tst2 = fcmp une float %b, 0.0 - br i1 %tst2, label %t3, label %end -; CHECK: fcmp {{s[0-9]+}}, #0.0 -; CHECK: b.eq .L - - -t3: -; This test can't be implemented with just one A64 conditional -; branch. LLVM converts "ordered and not equal" to "unordered or -; equal" before instruction selection, which is what we currently -; test. Obviously, other sequences are valid. - %tst3 = fcmp one float %a, %b - br i1 %tst3, label %t4, label %end -; CHECK: fcmp {{s[0-9]+}}, {{s[0-9]+}} -; CHECK-NEXT: b.eq .[[T4:LBB[0-9]+_[0-9]+]] -; CHECK-NEXT: b.vs .[[T4]] -t4: - %tst4 = fcmp uge float %a, -0.0 - br i1 %tst4, label %t5, label %end -; CHECK-NOT: fcmp {{s[0-9]+}}, #0.0 -; CHECK: b.mi .LBB - -t5: - call void @bar(i32 0) - ret void -end: - ret void - -} - -define void @test_double(double %a, double %b) { -; CHECK-LABEL: test_double: - - %tst1 = fcmp oeq double %a, %b - br i1 %tst1, label %end, label %t2 -; CHECK: fcmp {{d[0-9]+}}, {{d[0-9]+}} -; CHECK: b.eq .L - -t2: - %tst2 = fcmp une double %b, 0.0 - br i1 %tst2, label %t3, label %end -; CHECK: fcmp {{d[0-9]+}}, #0.0 -; CHECK: b.eq .L - - -t3: -; This test can't be implemented with just one A64 conditional -; branch. LLVM converts "ordered and not equal" to "unordered or -; equal" before instruction selection, which is what we currently -; test. Obviously, other sequences are valid. - %tst3 = fcmp one double %a, %b - br i1 %tst3, label %t4, label %end -; CHECK: fcmp {{d[0-9]+}}, {{d[0-9]+}} -; CHECK-NEXT: b.eq .[[T4:LBB[0-9]+_[0-9]+]] -; CHECK-NEXT: b.vs .[[T4]] -t4: - %tst4 = fcmp uge double %a, -0.0 - br i1 %tst4, label %t5, label %end -; CHECK-NOT: fcmp {{d[0-9]+}}, #0.0 -; CHECK: b.mi .LBB - -t5: - call void @bar(i32 0) - ret void -end: - ret void +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4 +; RUN: llc -mtriple=aarch64-none-eabi -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-SD-NOFP16 +; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-SD-FP16 +; RUN: llc -mtriple=aarch64-none-eabi -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-GI-NOFP16 +; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-GI-FP16 +; CHECK-GI: warning: Instruction selection used fallback path for v3f64_double +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v3f64_i32 +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v3f32_float +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v3f32_i32 +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v7f16_half +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v16f16_half +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v7f16_i32 +; CHECK-GI-FP16-NEXT: warning: Instruction selection used fallback path for v16f16_i32 + +define double @f64_double(double %a, double %b, double %d, double %e) { +; CHECK-LABEL: f64_double: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmp d0, d1 +; CHECK-NEXT: fcsel d0, d2, d3, mi +; CHECK-NEXT: ret +entry: + %c = fcmp olt double %a, %b + %s = select i1 %c, double %d, double %e + ret double %s +} + +define i32 @f64_i32(double %a, double %b, i32 %d, i32 %e) { +; CHECK-LABEL: f64_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmp d0, d1 +; CHECK-NEXT: csel w0, w0, w1, mi +; CHECK-NEXT: ret +entry: + %c = fcmp olt double %a, %b + %s = select i1 %c, i32 %d, i32 %e + ret i32 %s +} + +define float @f32_float(float %a, float %b, float %d, float %e) { +; CHECK-LABEL: f32_float: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmp s0, s1 +; CHECK-NEXT: fcsel s0, s2, s3, mi +; CHECK-NEXT: ret +entry: + %c = fcmp olt float %a, %b + %s = select i1 %c, float %d, float %e + ret float %s +} + +define i32 @f32_i32(float %a, float %b, i32 %d, i32 %e) { +; CHECK-LABEL: f32_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmp s0, s1 +; CHECK-NEXT: csel w0, w0, w1, mi +; CHECK-NEXT: ret +entry: + %c = fcmp olt float %a, %b + %s = select i1 %c, i32 %d, i32 %e + ret i32 %s +} + +define half @f16_half(half %a, half %b, half %d, half %e) { +; CHECK-SD-NOFP16-LABEL: f16_half: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s0, h0 +; CHECK-SD-NOFP16-NEXT: // kill: def $h3 killed $h3 def $s3 +; CHECK-SD-NOFP16-NEXT: // kill: def $h2 killed $h2 def $s2 +; CHECK-SD-NOFP16-NEXT: fcmp s0, s1 +; CHECK-SD-NOFP16-NEXT: fcsel s0, s2, s3, mi +; CHECK-SD-NOFP16-NEXT: // kill: def $h0 killed $h0 killed $s0 +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: f16_half: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmp h0, h1 +; CHECK-SD-FP16-NEXT: fcsel h0, h2, h3, mi +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-LABEL: f16_half: +; CHECK-GI: // %bb.0: // %entry +; CHECK-GI-NEXT: fcvt s0, h0 +; CHECK-GI-NEXT: fcvt s1, h1 +; CHECK-GI-NEXT: // kill: def $h2 killed $h2 def $s2 +; CHECK-GI-NEXT: // kill: def $h3 killed $h3 def $s3 +; CHECK-GI-NEXT: fmov w8, s2 +; CHECK-GI-NEXT: fmov w9, s3 +; CHECK-GI-NEXT: fcmp s0, s1 +; CHECK-GI-NEXT: csel w8, w8, w9, mi +; CHECK-GI-NEXT: fmov s0, w8 +; CHECK-GI-NEXT: // kill: def $h0 killed $h0 killed $s0 +; CHECK-GI-NEXT: ret +entry: + %c = fcmp olt half %a, %b + %s = select i1 %c, half %d, half %e + ret half %s +} + +define i32 @f16_i32(half %a, half %b, i32 %d, i32 %e) { +; CHECK-SD-NOFP16-LABEL: f16_i32: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s0, h0 +; CHECK-SD-NOFP16-NEXT: fcmp s0, s1 +; CHECK-SD-NOFP16-NEXT: csel w0, w0, w1, mi +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: f16_i32: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmp h0, h1 +; CHECK-SD-FP16-NEXT: csel w0, w0, w1, mi +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-LABEL: f16_i32: +; CHECK-GI: // %bb.0: // %entry +; CHECK-GI-NEXT: fcvt s0, h0 +; CHECK-GI-NEXT: fcvt s1, h1 +; CHECK-GI-NEXT: fcmp s0, s1 +; CHECK-GI-NEXT: csel w0, w0, w1, mi +; CHECK-GI-NEXT: ret +entry: + %c = fcmp olt half %a, %b + %s = select i1 %c, i32 %d, i32 %e + ret i32 %s +} + +define <2 x double> @v2f64_double(<2 x double> %a, <2 x double> %b, <2 x double> %d, <2 x double> %e) { +; CHECK-LABEL: v2f64_double: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmgt v0.2d, v1.2d, v0.2d +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <2 x double> %a, %b + %s = select <2 x i1> %c, <2 x double> %d, <2 x double> %e + ret <2 x double> %s +} + +define <3 x double> @v3f64_double(<3 x double> %a, <3 x double> %b, <3 x double> %d, <3 x double> %e) { +; CHECK-LABEL: v3f64_double: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3 +; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0 +; CHECK-NEXT: // kill: def $d4 killed $d4 def $q4 +; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1 +; CHECK-NEXT: // kill: def $d6 killed $d6 def $q6 +; CHECK-NEXT: // kill: def $d7 killed $d7 def $q7 +; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2 +; CHECK-NEXT: // kill: def $d5 killed $d5 def $q5 +; CHECK-NEXT: ldr d16, [sp, #24] +; CHECK-NEXT: ldr d17, [sp] +; CHECK-NEXT: mov v3.d[1], v4.d[0] +; CHECK-NEXT: mov v0.d[1], v1.d[0] +; CHECK-NEXT: mov v6.d[1], v7.d[0] +; CHECK-NEXT: ldp d1, d4, [sp, #8] +; CHECK-NEXT: fcmgt v2.2d, v5.2d, v2.2d +; CHECK-NEXT: mov v1.d[1], v4.d[0] +; CHECK-NEXT: fcmgt v0.2d, v3.2d, v0.2d +; CHECK-NEXT: bsl v2.16b, v17.16b, v16.16b +; CHECK-NEXT: // kill: def $d2 killed $d2 killed $q2 +; CHECK-NEXT: bsl v0.16b, v6.16b, v1.16b +; CHECK-NEXT: ext v1.16b, v0.16b, v0.16b, #8 +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: // kill: def $d1 killed $d1 killed $q1 +; CHECK-NEXT: ret +entry: + %c = fcmp olt <3 x double> %a, %b + %s = select <3 x i1> %c, <3 x double> %d, <3 x double> %e + ret <3 x double> %s +} + +define <4 x double> @v4f64_double(<4 x double> %a, <4 x double> %b, <4 x double> %d, <4 x double> %e) { +; CHECK-SD-LABEL: v4f64_double: +; CHECK-SD: // %bb.0: // %entry +; CHECK-SD-NEXT: fcmgt v1.2d, v3.2d, v1.2d +; CHECK-SD-NEXT: fcmgt v0.2d, v2.2d, v0.2d +; CHECK-SD-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-SD-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: v4f64_double: +; CHECK-GI: // %bb.0: // %entry +; CHECK-GI-NEXT: fcmgt v0.2d, v2.2d, v0.2d +; CHECK-GI-NEXT: fcmgt v1.2d, v3.2d, v1.2d +; CHECK-GI-NEXT: shl v0.2d, v0.2d, #63 +; CHECK-GI-NEXT: shl v1.2d, v1.2d, #63 +; CHECK-GI-NEXT: sshr v0.2d, v0.2d, #63 +; CHECK-GI-NEXT: sshr v1.2d, v1.2d, #63 +; CHECK-GI-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-GI-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-GI-NEXT: ret +entry: + %c = fcmp olt <4 x double> %a, %b + %s = select <4 x i1> %c, <4 x double> %d, <4 x double> %e + ret <4 x double> %s +} + +define <2 x i32> @v2f64_i32(<2 x double> %a, <2 x double> %b, <2 x i32> %d, <2 x i32> %e) { +; CHECK-LABEL: v2f64_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmgt v0.2d, v1.2d, v0.2d +; CHECK-NEXT: xtn v0.2s, v0.2d +; CHECK-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <2 x double> %a, %b + %s = select <2 x i1> %c, <2 x i32> %d, <2 x i32> %e + ret <2 x i32> %s +} + +define <3 x i32> @v3f64_i32(<3 x double> %a, <3 x double> %b, <3 x i32> %d, <3 x i32> %e) { +; CHECK-LABEL: v3f64_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3 +; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0 +; CHECK-NEXT: // kill: def $d4 killed $d4 def $q4 +; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1 +; CHECK-NEXT: // kill: def $d5 killed $d5 def $q5 +; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2 +; CHECK-NEXT: mov v0.d[1], v1.d[0] +; CHECK-NEXT: mov v3.d[1], v4.d[0] +; CHECK-NEXT: fcmgt v1.2d, v5.2d, v2.2d +; CHECK-NEXT: fcmgt v0.2d, v3.2d, v0.2d +; CHECK-NEXT: uzp1 v0.4s, v0.4s, v1.4s +; CHECK-NEXT: bsl v0.16b, v6.16b, v7.16b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <3 x double> %a, %b + %s = select <3 x i1> %c, <3 x i32> %d, <3 x i32> %e + ret <3 x i32> %s +} + +define <4 x i32> @v4f64_i32(<4 x double> %a, <4 x double> %b, <4 x i32> %d, <4 x i32> %e) { +; CHECK-SD-LABEL: v4f64_i32: +; CHECK-SD: // %bb.0: // %entry +; CHECK-SD-NEXT: fcmgt v1.2d, v3.2d, v1.2d +; CHECK-SD-NEXT: fcmgt v0.2d, v2.2d, v0.2d +; CHECK-SD-NEXT: uzp1 v0.4s, v0.4s, v1.4s +; CHECK-SD-NEXT: bsl v0.16b, v4.16b, v5.16b +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: v4f64_i32: +; CHECK-GI: // %bb.0: // %entry +; CHECK-GI-NEXT: fcmgt v0.2d, v2.2d, v0.2d +; CHECK-GI-NEXT: fcmgt v1.2d, v3.2d, v1.2d +; CHECK-GI-NEXT: uzp1 v0.4s, v0.4s, v1.4s +; CHECK-GI-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-NEXT: sshr v0.4s, v0.4s, #31 +; CHECK-GI-NEXT: bsl v0.16b, v4.16b, v5.16b +; CHECK-GI-NEXT: ret +entry: + %c = fcmp olt <4 x double> %a, %b + %s = select <4 x i1> %c, <4 x i32> %d, <4 x i32> %e + ret <4 x i32> %s +} + +define <2 x float> @v2f32_float(<2 x float> %a, <2 x float> %b, <2 x float> %d, <2 x float> %e) { +; CHECK-LABEL: v2f32_float: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmgt v0.2s, v1.2s, v0.2s +; CHECK-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <2 x float> %a, %b + %s = select <2 x i1> %c, <2 x float> %d, <2 x float> %e + ret <2 x float> %s +} + +define <3 x float> @v3f32_float(<3 x float> %a, <3 x float> %b, <3 x float> %d, <3 x float> %e) { +; CHECK-LABEL: v3f32_float: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmgt v0.4s, v1.4s, v0.4s +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <3 x float> %a, %b + %s = select <3 x i1> %c, <3 x float> %d, <3 x float> %e + ret <3 x float> %s +} + +define <4 x float> @v4f32_float(<4 x float> %a, <4 x float> %b, <4 x float> %d, <4 x float> %e) { +; CHECK-LABEL: v4f32_float: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmgt v0.4s, v1.4s, v0.4s +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <4 x float> %a, %b + %s = select <4 x i1> %c, <4 x float> %d, <4 x float> %e + ret <4 x float> %s +} + +define <8 x float> @v8f32_float(<8 x float> %a, <8 x float> %b, <8 x float> %d, <8 x float> %e) { +; CHECK-SD-LABEL: v8f32_float: +; CHECK-SD: // %bb.0: // %entry +; CHECK-SD-NEXT: fcmgt v1.4s, v3.4s, v1.4s +; CHECK-SD-NEXT: fcmgt v0.4s, v2.4s, v0.4s +; CHECK-SD-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-SD-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: v8f32_float: +; CHECK-GI: // %bb.0: // %entry +; CHECK-GI-NEXT: fcmgt v0.4s, v2.4s, v0.4s +; CHECK-GI-NEXT: fcmgt v1.4s, v3.4s, v1.4s +; CHECK-GI-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-NEXT: shl v1.4s, v1.4s, #31 +; CHECK-GI-NEXT: sshr v0.4s, v0.4s, #31 +; CHECK-GI-NEXT: sshr v1.4s, v1.4s, #31 +; CHECK-GI-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-GI-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-GI-NEXT: ret +entry: + %c = fcmp olt <8 x float> %a, %b + %s = select <8 x i1> %c, <8 x float> %d, <8 x float> %e + ret <8 x float> %s +} + +define <2 x i32> @v2f32_i32(<2 x float> %a, <2 x float> %b, <2 x i32> %d, <2 x i32> %e) { +; CHECK-LABEL: v2f32_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmgt v0.2s, v1.2s, v0.2s +; CHECK-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <2 x float> %a, %b + %s = select <2 x i1> %c, <2 x i32> %d, <2 x i32> %e + ret <2 x i32> %s +} + +define <3 x i32> @v3f32_i32(<3 x float> %a, <3 x float> %b, <3 x i32> %d, <3 x i32> %e) { +; CHECK-LABEL: v3f32_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmgt v0.4s, v1.4s, v0.4s +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <3 x float> %a, %b + %s = select <3 x i1> %c, <3 x i32> %d, <3 x i32> %e + ret <3 x i32> %s +} + +define <4 x i32> @v4f32_i32(<4 x float> %a, <4 x float> %b, <4 x i32> %d, <4 x i32> %e) { +; CHECK-LABEL: v4f32_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: fcmgt v0.4s, v1.4s, v0.4s +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = fcmp olt <4 x float> %a, %b + %s = select <4 x i1> %c, <4 x i32> %d, <4 x i32> %e + ret <4 x i32> %s +} + +define <8 x i32> @v8f32_i32(<8 x float> %a, <8 x float> %b, <8 x i32> %d, <8 x i32> %e) { +; CHECK-SD-LABEL: v8f32_i32: +; CHECK-SD: // %bb.0: // %entry +; CHECK-SD-NEXT: fcmgt v1.4s, v3.4s, v1.4s +; CHECK-SD-NEXT: fcmgt v0.4s, v2.4s, v0.4s +; CHECK-SD-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-SD-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: v8f32_i32: +; CHECK-GI: // %bb.0: // %entry +; CHECK-GI-NEXT: fcmgt v0.4s, v2.4s, v0.4s +; CHECK-GI-NEXT: fcmgt v1.4s, v3.4s, v1.4s +; CHECK-GI-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-NEXT: shl v1.4s, v1.4s, #31 +; CHECK-GI-NEXT: sshr v0.4s, v0.4s, #31 +; CHECK-GI-NEXT: sshr v1.4s, v1.4s, #31 +; CHECK-GI-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-GI-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-GI-NEXT: ret +entry: + %c = fcmp olt <8 x float> %a, %b + %s = select <8 x i1> %c, <8 x i32> %d, <8 x i32> %e + ret <8 x i32> %s +} + +define <7 x half> @v7f16_half(<7 x half> %a, <7 x half> %b, <7 x half> %d, <7 x half> %e) { +; CHECK-SD-NOFP16-LABEL: v7f16_half: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: mov h4, v1.h[1] +; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[1] +; CHECK-SD-NOFP16-NEXT: fcvt s6, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h0 +; CHECK-SD-NOFP16-NEXT: mov h16, v1.h[2] +; CHECK-SD-NOFP16-NEXT: fcvt s4, h4 +; CHECK-SD-NOFP16-NEXT: fcvt s5, h5 +; CHECK-SD-NOFP16-NEXT: fcmp s5, s4 +; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[2] +; CHECK-SD-NOFP16-NEXT: mov h5, v1.h[3] +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s7, s6 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s4, h4 +; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[3] +; CHECK-SD-NOFP16-NEXT: fcvt s5, h5 +; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[4] +; CHECK-SD-NOFP16-NEXT: csetm w9, mi +; CHECK-SD-NOFP16-NEXT: fcmp s4, s7 +; CHECK-SD-NOFP16-NEXT: fmov s4, w9 +; CHECK-SD-NOFP16-NEXT: fcvt s6, h6 +; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[4] +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: mov v4.h[1], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s6, s5 +; CHECK-SD-NOFP16-NEXT: mov h5, v1.h[5] +; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[5] +; CHECK-SD-NOFP16-NEXT: fcvt s7, h7 +; CHECK-SD-NOFP16-NEXT: mov v4.h[2], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcvt s5, h5 +; CHECK-SD-NOFP16-NEXT: fcvt s6, h6 +; CHECK-SD-NOFP16-NEXT: fcmp s16, s7 +; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[6] +; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[6] +; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-SD-NOFP16-NEXT: mov v4.h[3], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s6, s5 +; CHECK-SD-NOFP16-NEXT: fcvt s5, h7 +; CHECK-SD-NOFP16-NEXT: fcvt s6, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s0, h0 +; CHECK-SD-NOFP16-NEXT: mov v4.h[4], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s6, s5 +; CHECK-SD-NOFP16-NEXT: mov v4.h[5], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s0, s1 +; CHECK-SD-NOFP16-NEXT: mov v4.h[6], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: mov v4.h[7], w8 +; CHECK-SD-NOFP16-NEXT: mov v0.16b, v4.16b +; CHECK-SD-NOFP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: v7f16_half: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmgt v0.8h, v1.8h, v0.8h +; CHECK-SD-FP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-NOFP16-LABEL: v7f16_half: +; CHECK-GI-NOFP16: // %bb.0: // %entry +; CHECK-GI-NOFP16-NEXT: mov h4, v1.h[1] +; CHECK-GI-NOFP16-NEXT: mov h5, v0.h[1] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h1 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h0 +; CHECK-GI-NOFP16-NEXT: mov h16, v1.h[2] +; CHECK-GI-NOFP16-NEXT: fcvt s4, h4 +; CHECK-GI-NOFP16-NEXT: fcvt s5, h5 +; CHECK-GI-NOFP16-NEXT: fcmp s5, s4 +; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[2] +; CHECK-GI-NOFP16-NEXT: mov h5, v1.h[3] +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s7, s6 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s4, h4 +; CHECK-GI-NOFP16-NEXT: mov h6, v0.h[3] +; CHECK-GI-NOFP16-NEXT: fcvt s5, h5 +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[4] +; CHECK-GI-NOFP16-NEXT: csetm w9, mi +; CHECK-GI-NOFP16-NEXT: fcmp s4, s7 +; CHECK-GI-NOFP16-NEXT: fmov s4, w9 +; CHECK-GI-NOFP16-NEXT: fcvt s6, h6 +; CHECK-GI-NOFP16-NEXT: mov h7, v1.h[4] +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: mov v4.h[1], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s5 +; CHECK-GI-NOFP16-NEXT: mov h5, v1.h[5] +; CHECK-GI-NOFP16-NEXT: mov h6, v0.h[5] +; CHECK-GI-NOFP16-NEXT: fcvt s7, h7 +; CHECK-GI-NOFP16-NEXT: mov v4.h[2], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcvt s5, h5 +; CHECK-GI-NOFP16-NEXT: fcvt s6, h6 +; CHECK-GI-NOFP16-NEXT: fcmp s16, s7 +; CHECK-GI-NOFP16-NEXT: mov h7, v1.h[6] +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[6] +; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-GI-NOFP16-NEXT: mov v4.h[3], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s5 +; CHECK-GI-NOFP16-NEXT: fcvt s5, h7 +; CHECK-GI-NOFP16-NEXT: fcvt s6, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: mov v4.h[4], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s5 +; CHECK-GI-NOFP16-NEXT: mov v4.h[5], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s1 +; CHECK-GI-NOFP16-NEXT: mov v4.h[6], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: mov v4.h[7], w8 +; CHECK-GI-NOFP16-NEXT: mov v0.16b, v4.16b +; CHECK-GI-NOFP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-GI-NOFP16-NEXT: ret +; +; CHECK-GI-FP16-LABEL: v7f16_half: +; CHECK-GI-FP16: // %bb.0: // %entry +; CHECK-GI-FP16-NEXT: fcmgt v0.8h, v1.8h, v0.8h +; CHECK-GI-FP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-GI-FP16-NEXT: ret +entry: + %c = fcmp olt <7 x half> %a, %b + %s = select <7 x i1> %c, <7 x half> %d, <7 x half> %e + ret <7 x half> %s +} + +define <4 x half> @v4f16_half(<4 x half> %a, <4 x half> %b, <4 x half> %d, <4 x half> %e) { +; CHECK-SD-NOFP16-LABEL: v4f16_half: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: fcvtl v0.4s, v0.4h +; CHECK-SD-NOFP16-NEXT: fcvtl v1.4s, v1.4h +; CHECK-SD-NOFP16-NEXT: fcmgt v0.4s, v1.4s, v0.4s +; CHECK-SD-NOFP16-NEXT: xtn v0.4h, v0.4s +; CHECK-SD-NOFP16-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: v4f16_half: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmgt v0.4h, v1.4h, v0.4h +; CHECK-SD-FP16-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-NOFP16-LABEL: v4f16_half: +; CHECK-GI-NOFP16: // %bb.0: // %entry +; CHECK-GI-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0 +; CHECK-GI-NOFP16-NEXT: // kill: def $d1 killed $d1 def $q1 +; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[1] +; CHECK-GI-NOFP16-NEXT: mov h5, v1.h[1] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h1 +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[2] +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[2] +; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[3] +; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[3] +; CHECK-GI-NOFP16-NEXT: fcvt s4, h4 +; CHECK-GI-NOFP16-NEXT: fcvt s5, h5 +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: fcvt s6, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h17 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s4, s5 +; CHECK-GI-NOFP16-NEXT: fmov s4, w8 +; CHECK-GI-NOFP16-NEXT: cset w9, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: fmov s5, w9 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s1 +; CHECK-GI-NOFP16-NEXT: mov v4.h[1], v5.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s0, w8 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: mov v4.h[2], v0.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s0, w8 +; CHECK-GI-NOFP16-NEXT: mov v4.h[3], v0.h[0] +; CHECK-GI-NOFP16-NEXT: shl v0.4h, v4.4h, #15 +; CHECK-GI-NOFP16-NEXT: sshr v0.4h, v0.4h, #15 +; CHECK-GI-NOFP16-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-GI-NOFP16-NEXT: ret +; +; CHECK-GI-FP16-LABEL: v4f16_half: +; CHECK-GI-FP16: // %bb.0: // %entry +; CHECK-GI-FP16-NEXT: fcmgt v0.4h, v1.4h, v0.4h +; CHECK-GI-FP16-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-GI-FP16-NEXT: ret +entry: + %c = fcmp olt <4 x half> %a, %b + %s = select <4 x i1> %c, <4 x half> %d, <4 x half> %e + ret <4 x half> %s +} + +define <8 x half> @v8f16_half(<8 x half> %a, <8 x half> %b, <8 x half> %d, <8 x half> %e) { +; CHECK-SD-NOFP16-LABEL: v8f16_half: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: mov h4, v1.h[1] +; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[1] +; CHECK-SD-NOFP16-NEXT: fcvt s6, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h0 +; CHECK-SD-NOFP16-NEXT: mov h16, v1.h[2] +; CHECK-SD-NOFP16-NEXT: fcvt s4, h4 +; CHECK-SD-NOFP16-NEXT: fcvt s5, h5 +; CHECK-SD-NOFP16-NEXT: fcmp s5, s4 +; CHECK-SD-NOFP16-NEXT: mov h4, v0.h[2] +; CHECK-SD-NOFP16-NEXT: mov h5, v1.h[3] +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s7, s6 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s4, h4 +; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[3] +; CHECK-SD-NOFP16-NEXT: fcvt s5, h5 +; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[4] +; CHECK-SD-NOFP16-NEXT: csetm w9, mi +; CHECK-SD-NOFP16-NEXT: fcmp s4, s7 +; CHECK-SD-NOFP16-NEXT: fmov s4, w9 +; CHECK-SD-NOFP16-NEXT: fcvt s6, h6 +; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[4] +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: mov v4.h[1], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s6, s5 +; CHECK-SD-NOFP16-NEXT: mov h5, v1.h[5] +; CHECK-SD-NOFP16-NEXT: mov h6, v0.h[5] +; CHECK-SD-NOFP16-NEXT: fcvt s7, h7 +; CHECK-SD-NOFP16-NEXT: mov v4.h[2], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcvt s5, h5 +; CHECK-SD-NOFP16-NEXT: fcvt s6, h6 +; CHECK-SD-NOFP16-NEXT: fcmp s16, s7 +; CHECK-SD-NOFP16-NEXT: mov h7, v1.h[6] +; CHECK-SD-NOFP16-NEXT: mov h16, v0.h[6] +; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-SD-NOFP16-NEXT: mov v4.h[3], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s6, s5 +; CHECK-SD-NOFP16-NEXT: fcvt s5, h7 +; CHECK-SD-NOFP16-NEXT: fcvt s6, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s0, h0 +; CHECK-SD-NOFP16-NEXT: mov v4.h[4], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s6, s5 +; CHECK-SD-NOFP16-NEXT: mov v4.h[5], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s0, s1 +; CHECK-SD-NOFP16-NEXT: mov v4.h[6], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: mov v4.h[7], w8 +; CHECK-SD-NOFP16-NEXT: mov v0.16b, v4.16b +; CHECK-SD-NOFP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: v8f16_half: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmgt v0.8h, v1.8h, v0.8h +; CHECK-SD-FP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-NOFP16-LABEL: v8f16_half: +; CHECK-GI-NOFP16: // %bb.0: // %entry +; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[1] +; CHECK-GI-NOFP16-NEXT: mov h5, v1.h[1] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h1 +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[2] +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[2] +; CHECK-GI-NOFP16-NEXT: mov h18, v1.h[4] +; CHECK-GI-NOFP16-NEXT: fcvt s4, h4 +; CHECK-GI-NOFP16-NEXT: fcvt s5, h5 +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: mov h6, v0.h[3] +; CHECK-GI-NOFP16-NEXT: fcvt s7, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h17 +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[3] +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s4, s5 +; CHECK-GI-NOFP16-NEXT: mov h5, v0.h[4] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h6 +; CHECK-GI-NOFP16-NEXT: fmov s4, w8 +; CHECK-GI-NOFP16-NEXT: cset w9, mi +; CHECK-GI-NOFP16-NEXT: fcmp s7, s16 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h17 +; CHECK-GI-NOFP16-NEXT: fmov s16, w9 +; CHECK-GI-NOFP16-NEXT: fcvt s5, h5 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h18 +; CHECK-GI-NOFP16-NEXT: mov h18, v1.h[5] +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: mov v4.h[1], v16.h[0] +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[5] +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: fmov s6, w8 +; CHECK-GI-NOFP16-NEXT: mov h7, v1.h[6] +; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s5, s17 +; CHECK-GI-NOFP16-NEXT: mov v4.h[2], v6.h[0] +; CHECK-GI-NOFP16-NEXT: mov h6, v0.h[6] +; CHECK-GI-NOFP16-NEXT: fcvt s5, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h18 +; CHECK-GI-NOFP16-NEXT: fmov s17, w8 +; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcvt s7, h7 +; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: mov v4.h[3], v17.h[0] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h6 +; CHECK-GI-NOFP16-NEXT: fcmp s5, s16 +; CHECK-GI-NOFP16-NEXT: fmov s5, w8 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: mov v4.h[4], v5.h[0] +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: fmov s5, w8 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s1 +; CHECK-GI-NOFP16-NEXT: mov v4.h[5], v5.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s0, w8 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: mov v4.h[6], v0.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s0, w8 +; CHECK-GI-NOFP16-NEXT: mov v4.h[7], v0.h[0] +; CHECK-GI-NOFP16-NEXT: shl v0.8h, v4.8h, #15 +; CHECK-GI-NOFP16-NEXT: sshr v0.8h, v0.8h, #15 +; CHECK-GI-NOFP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-GI-NOFP16-NEXT: ret +; +; CHECK-GI-FP16-LABEL: v8f16_half: +; CHECK-GI-FP16: // %bb.0: // %entry +; CHECK-GI-FP16-NEXT: fcmgt v0.8h, v1.8h, v0.8h +; CHECK-GI-FP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-GI-FP16-NEXT: ret +entry: + %c = fcmp olt <8 x half> %a, %b + %s = select <8 x i1> %c, <8 x half> %d, <8 x half> %e + ret <8 x half> %s +} + +define <16 x half> @v16f16_half(<16 x half> %a, <16 x half> %b, <16 x half> %d, <16 x half> %e) { +; CHECK-SD-NOFP16-LABEL: v16f16_half: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: mov h16, v3.h[1] +; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[1] +; CHECK-SD-NOFP16-NEXT: mov h18, v3.h[2] +; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[2] +; CHECK-SD-NOFP16-NEXT: fcvt s20, h3 +; CHECK-SD-NOFP16-NEXT: fcvt s21, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v3.h[3] +; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[3] +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s21, s20 +; CHECK-SD-NOFP16-NEXT: mov h20, v3.h[4] +; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[4] +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: csetm w14, mi +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: mov h18, v3.h[5] +; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[5] +; CHECK-SD-NOFP16-NEXT: fcvt s20, h20 +; CHECK-SD-NOFP16-NEXT: fcvt s21, h21 +; CHECK-SD-NOFP16-NEXT: csetm w13, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v3.h[6] +; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[6] +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: mov h3, v3.h[7] +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-SD-NOFP16-NEXT: csetm w11, mi +; CHECK-SD-NOFP16-NEXT: fcmp s21, s20 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h3 +; CHECK-SD-NOFP16-NEXT: csetm w12, mi +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: mov h18, v2.h[1] +; CHECK-SD-NOFP16-NEXT: mov h19, v0.h[1] +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: csetm w10, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h18 +; CHECK-SD-NOFP16-NEXT: mov h18, v2.h[2] +; CHECK-SD-NOFP16-NEXT: fcvt s17, h19 +; CHECK-SD-NOFP16-NEXT: mov h19, v0.h[2] +; CHECK-SD-NOFP16-NEXT: csetm w9, mi +; CHECK-SD-NOFP16-NEXT: fcmp s1, s3 +; CHECK-SD-NOFP16-NEXT: fcvt s1, h2 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h0 +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: csetm w15, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v2.h[3] +; CHECK-SD-NOFP16-NEXT: mov h17, v0.h[3] +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: csetm w16, mi +; CHECK-SD-NOFP16-NEXT: fcmp s3, s1 +; CHECK-SD-NOFP16-NEXT: fmov s1, w14 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: csetm w14, mi +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: mov h18, v2.h[4] +; CHECK-SD-NOFP16-NEXT: fmov s3, w14 +; CHECK-SD-NOFP16-NEXT: mov h19, v0.h[4] +; CHECK-SD-NOFP16-NEXT: mov v1.h[1], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v2.h[5] +; CHECK-SD-NOFP16-NEXT: mov v3.h[1], w16 +; CHECK-SD-NOFP16-NEXT: mov h17, v0.h[5] +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: mov v1.h[2], w13 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: mov v3.h[2], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: mov h18, v2.h[6] +; CHECK-SD-NOFP16-NEXT: mov h19, v0.h[6] +; CHECK-SD-NOFP16-NEXT: mov v1.h[3], w11 +; CHECK-SD-NOFP16-NEXT: mov h2, v2.h[7] +; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-SD-NOFP16-NEXT: mov v3.h[3], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h18 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h19 +; CHECK-SD-NOFP16-NEXT: mov v1.h[4], w12 +; CHECK-SD-NOFP16-NEXT: fcvt s2, h2 +; CHECK-SD-NOFP16-NEXT: fcvt s0, h0 +; CHECK-SD-NOFP16-NEXT: mov v3.h[4], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov v1.h[5], w10 +; CHECK-SD-NOFP16-NEXT: mov v3.h[5], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s0, s2 +; CHECK-SD-NOFP16-NEXT: mov v1.h[6], w9 +; CHECK-SD-NOFP16-NEXT: mov v3.h[6], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: mov v1.h[7], w15 +; CHECK-SD-NOFP16-NEXT: mov v3.h[7], w8 +; CHECK-SD-NOFP16-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-SD-NOFP16-NEXT: mov v0.16b, v3.16b +; CHECK-SD-NOFP16-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: v16f16_half: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmgt v1.8h, v3.8h, v1.8h +; CHECK-SD-FP16-NEXT: fcmgt v0.8h, v2.8h, v0.8h +; CHECK-SD-FP16-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-SD-FP16-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-NOFP16-LABEL: v16f16_half: +; CHECK-GI-NOFP16: // %bb.0: // %entry +; CHECK-GI-NOFP16-NEXT: mov h16, v3.h[1] +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[1] +; CHECK-GI-NOFP16-NEXT: mov h18, v3.h[2] +; CHECK-GI-NOFP16-NEXT: mov h19, v1.h[2] +; CHECK-GI-NOFP16-NEXT: fcvt s20, h3 +; CHECK-GI-NOFP16-NEXT: fcvt s21, h1 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: fcvt s18, h18 +; CHECK-GI-NOFP16-NEXT: fcvt s19, h19 +; CHECK-GI-NOFP16-NEXT: fcmp s17, s16 +; CHECK-GI-NOFP16-NEXT: mov h16, v3.h[3] +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[3] +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s21, s20 +; CHECK-GI-NOFP16-NEXT: mov h20, v3.h[4] +; CHECK-GI-NOFP16-NEXT: mov h21, v1.h[4] +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: csetm w14, mi +; CHECK-GI-NOFP16-NEXT: fcmp s19, s18 +; CHECK-GI-NOFP16-NEXT: mov h18, v3.h[5] +; CHECK-GI-NOFP16-NEXT: mov h19, v1.h[5] +; CHECK-GI-NOFP16-NEXT: fcvt s20, h20 +; CHECK-GI-NOFP16-NEXT: fcvt s21, h21 +; CHECK-GI-NOFP16-NEXT: csetm w13, mi +; CHECK-GI-NOFP16-NEXT: fcmp s17, s16 +; CHECK-GI-NOFP16-NEXT: mov h16, v3.h[6] +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[6] +; CHECK-GI-NOFP16-NEXT: fcvt s18, h18 +; CHECK-GI-NOFP16-NEXT: mov h3, v3.h[7] +; CHECK-GI-NOFP16-NEXT: fcvt s19, h19 +; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-GI-NOFP16-NEXT: csetm w11, mi +; CHECK-GI-NOFP16-NEXT: fcmp s21, s20 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: fcvt s3, h3 +; CHECK-GI-NOFP16-NEXT: csetm w12, mi +; CHECK-GI-NOFP16-NEXT: fcmp s19, s18 +; CHECK-GI-NOFP16-NEXT: mov h18, v2.h[1] +; CHECK-GI-NOFP16-NEXT: mov h19, v0.h[1] +; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: csetm w10, mi +; CHECK-GI-NOFP16-NEXT: fcmp s17, s16 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h18 +; CHECK-GI-NOFP16-NEXT: mov h18, v2.h[2] +; CHECK-GI-NOFP16-NEXT: fcvt s17, h19 +; CHECK-GI-NOFP16-NEXT: mov h19, v0.h[2] +; CHECK-GI-NOFP16-NEXT: csetm w9, mi +; CHECK-GI-NOFP16-NEXT: fcmp s1, s3 +; CHECK-GI-NOFP16-NEXT: fcvt s1, h2 +; CHECK-GI-NOFP16-NEXT: fcvt s3, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s18, h18 +; CHECK-GI-NOFP16-NEXT: csetm w15, mi +; CHECK-GI-NOFP16-NEXT: fcmp s17, s16 +; CHECK-GI-NOFP16-NEXT: mov h16, v2.h[3] +; CHECK-GI-NOFP16-NEXT: mov h17, v0.h[3] +; CHECK-GI-NOFP16-NEXT: fcvt s19, h19 +; CHECK-GI-NOFP16-NEXT: csetm w16, mi +; CHECK-GI-NOFP16-NEXT: fcmp s3, s1 +; CHECK-GI-NOFP16-NEXT: fmov s1, w14 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: csetm w14, mi +; CHECK-GI-NOFP16-NEXT: fcmp s19, s18 +; CHECK-GI-NOFP16-NEXT: mov h18, v2.h[4] +; CHECK-GI-NOFP16-NEXT: fmov s3, w14 +; CHECK-GI-NOFP16-NEXT: mov h19, v0.h[4] +; CHECK-GI-NOFP16-NEXT: mov v1.h[1], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s17, s16 +; CHECK-GI-NOFP16-NEXT: mov h16, v2.h[5] +; CHECK-GI-NOFP16-NEXT: mov v3.h[1], w16 +; CHECK-GI-NOFP16-NEXT: mov h17, v0.h[5] +; CHECK-GI-NOFP16-NEXT: fcvt s18, h18 +; CHECK-GI-NOFP16-NEXT: fcvt s19, h19 +; CHECK-GI-NOFP16-NEXT: mov v1.h[2], w13 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: mov v3.h[2], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: fcmp s19, s18 +; CHECK-GI-NOFP16-NEXT: mov h18, v2.h[6] +; CHECK-GI-NOFP16-NEXT: mov h19, v0.h[6] +; CHECK-GI-NOFP16-NEXT: mov v1.h[3], w11 +; CHECK-GI-NOFP16-NEXT: mov h2, v2.h[7] +; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-GI-NOFP16-NEXT: mov v3.h[3], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s17, s16 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h18 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h19 +; CHECK-GI-NOFP16-NEXT: mov v1.h[4], w12 +; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: mov v3.h[4], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s17, s16 +; CHECK-GI-NOFP16-NEXT: mov v1.h[5], w10 +; CHECK-GI-NOFP16-NEXT: mov v3.h[5], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s2 +; CHECK-GI-NOFP16-NEXT: mov v1.h[6], w9 +; CHECK-GI-NOFP16-NEXT: mov v3.h[6], w8 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: mov v1.h[7], w15 +; CHECK-GI-NOFP16-NEXT: mov v3.h[7], w8 +; CHECK-GI-NOFP16-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-GI-NOFP16-NEXT: mov v0.16b, v3.16b +; CHECK-GI-NOFP16-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-GI-NOFP16-NEXT: ret +; +; CHECK-GI-FP16-LABEL: v16f16_half: +; CHECK-GI-FP16: // %bb.0: // %entry +; CHECK-GI-FP16-NEXT: fcmgt v1.8h, v3.8h, v1.8h +; CHECK-GI-FP16-NEXT: fcmgt v0.8h, v2.8h, v0.8h +; CHECK-GI-FP16-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-GI-FP16-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-GI-FP16-NEXT: ret +entry: + %c = fcmp olt <16 x half> %a, %b + %s = select <16 x i1> %c, <16 x half> %d, <16 x half> %e + ret <16 x half> %s +} + +define <7 x i32> @v7f16_i32(<7 x half> %a, <7 x half> %b, <7 x i32> %d, <7 x i32> %e) { +; CHECK-SD-NOFP16-LABEL: v7f16_i32: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: mov h2, v1.h[1] +; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[1] +; CHECK-SD-NOFP16-NEXT: mov h4, v1.h[2] +; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[2] +; CHECK-SD-NOFP16-NEXT: fcvt s6, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h0 +; CHECK-SD-NOFP16-NEXT: fcvt s2, h2 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h3 +; CHECK-SD-NOFP16-NEXT: fcvt s4, h4 +; CHECK-SD-NOFP16-NEXT: fcvt s5, h5 +; CHECK-SD-NOFP16-NEXT: fcmp s3, s2 +; CHECK-SD-NOFP16-NEXT: mov h2, v1.h[3] +; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[3] +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s7, s6 +; CHECK-SD-NOFP16-NEXT: mov h6, v1.h[5] +; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5] +; CHECK-SD-NOFP16-NEXT: fcvt s2, h2 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h3 +; CHECK-SD-NOFP16-NEXT: csetm w9, mi +; CHECK-SD-NOFP16-NEXT: fcmp s5, s4 +; CHECK-SD-NOFP16-NEXT: mov h4, v1.h[4] +; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[4] +; CHECK-SD-NOFP16-NEXT: fcvt s6, h6 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h7 +; CHECK-SD-NOFP16-NEXT: csetm w10, mi +; CHECK-SD-NOFP16-NEXT: fcmp s3, s2 +; CHECK-SD-NOFP16-NEXT: fcvt s2, h4 +; CHECK-SD-NOFP16-NEXT: mov h4, v1.h[6] +; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-SD-NOFP16-NEXT: fcvt s3, h5 +; CHECK-SD-NOFP16-NEXT: mov h5, v0.h[6] +; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-SD-NOFP16-NEXT: csetm w11, mi +; CHECK-SD-NOFP16-NEXT: fcmp s7, s6 +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: csetm w12, mi +; CHECK-SD-NOFP16-NEXT: fcmp s3, s2 +; CHECK-SD-NOFP16-NEXT: fcvt s2, h4 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h5 +; CHECK-SD-NOFP16-NEXT: fmov s4, w9 +; CHECK-SD-NOFP16-NEXT: fcvt s0, h0 +; CHECK-SD-NOFP16-NEXT: add x9, sp, #8 +; CHECK-SD-NOFP16-NEXT: csetm w13, mi +; CHECK-SD-NOFP16-NEXT: fmov s5, w13 +; CHECK-SD-NOFP16-NEXT: mov v4.h[1], w8 +; CHECK-SD-NOFP16-NEXT: mov x8, sp +; CHECK-SD-NOFP16-NEXT: fcmp s3, s2 +; CHECK-SD-NOFP16-NEXT: fmov s2, w7 +; CHECK-SD-NOFP16-NEXT: fmov s3, w0 +; CHECK-SD-NOFP16-NEXT: mov v5.h[1], w12 +; CHECK-SD-NOFP16-NEXT: ld1 { v2.s }[1], [x8] +; CHECK-SD-NOFP16-NEXT: mov v3.s[1], w1 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: mov v4.h[2], w10 +; CHECK-SD-NOFP16-NEXT: fcmp s0, s1 +; CHECK-SD-NOFP16-NEXT: fmov s1, w4 +; CHECK-SD-NOFP16-NEXT: ldr s0, [sp, #24] +; CHECK-SD-NOFP16-NEXT: mov v5.h[2], w8 +; CHECK-SD-NOFP16-NEXT: ld1 { v2.s }[2], [x9] +; CHECK-SD-NOFP16-NEXT: add x9, sp, #32 +; CHECK-SD-NOFP16-NEXT: mov v3.s[2], w2 +; CHECK-SD-NOFP16-NEXT: mov v1.s[1], w5 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: mov v4.h[3], w11 +; CHECK-SD-NOFP16-NEXT: ld1 { v0.s }[1], [x9] +; CHECK-SD-NOFP16-NEXT: mov v5.h[3], w8 +; CHECK-SD-NOFP16-NEXT: add x8, sp, #16 +; CHECK-SD-NOFP16-NEXT: ld1 { v2.s }[3], [x8] +; CHECK-SD-NOFP16-NEXT: mov v3.s[3], w3 +; CHECK-SD-NOFP16-NEXT: add x8, sp, #40 +; CHECK-SD-NOFP16-NEXT: mov v1.s[2], w6 +; CHECK-SD-NOFP16-NEXT: sshll v4.4s, v4.4h, #0 +; CHECK-SD-NOFP16-NEXT: ld1 { v0.s }[2], [x8] +; CHECK-SD-NOFP16-NEXT: sshll v5.4s, v5.4h, #0 +; CHECK-SD-NOFP16-NEXT: bit v2.16b, v3.16b, v4.16b +; CHECK-SD-NOFP16-NEXT: bit v0.16b, v1.16b, v5.16b +; CHECK-SD-NOFP16-NEXT: mov w1, v2.s[1] +; CHECK-SD-NOFP16-NEXT: mov w2, v2.s[2] +; CHECK-SD-NOFP16-NEXT: mov w3, v2.s[3] +; CHECK-SD-NOFP16-NEXT: fmov w0, s2 +; CHECK-SD-NOFP16-NEXT: mov w5, v0.s[1] +; CHECK-SD-NOFP16-NEXT: mov w6, v0.s[2] +; CHECK-SD-NOFP16-NEXT: fmov w4, s0 +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: v7f16_i32: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fmov s2, w0 +; CHECK-SD-FP16-NEXT: fmov s3, w7 +; CHECK-SD-FP16-NEXT: mov x8, sp +; CHECK-SD-FP16-NEXT: fmov s5, w4 +; CHECK-SD-FP16-NEXT: ldr s4, [sp, #24] +; CHECK-SD-FP16-NEXT: fcmgt v0.8h, v1.8h, v0.8h +; CHECK-SD-FP16-NEXT: add x9, sp, #32 +; CHECK-SD-FP16-NEXT: mov v2.s[1], w1 +; CHECK-SD-FP16-NEXT: ld1 { v3.s }[1], [x8] +; CHECK-SD-FP16-NEXT: add x8, sp, #8 +; CHECK-SD-FP16-NEXT: mov v5.s[1], w5 +; CHECK-SD-FP16-NEXT: ld1 { v4.s }[1], [x9] +; CHECK-SD-FP16-NEXT: add x9, sp, #16 +; CHECK-SD-FP16-NEXT: sshll v1.4s, v0.4h, #0 +; CHECK-SD-FP16-NEXT: sshll2 v0.4s, v0.8h, #0 +; CHECK-SD-FP16-NEXT: ld1 { v3.s }[2], [x8] +; CHECK-SD-FP16-NEXT: add x8, sp, #40 +; CHECK-SD-FP16-NEXT: mov v2.s[2], w2 +; CHECK-SD-FP16-NEXT: ld1 { v4.s }[2], [x8] +; CHECK-SD-FP16-NEXT: mov v5.s[2], w6 +; CHECK-SD-FP16-NEXT: ld1 { v3.s }[3], [x9] +; CHECK-SD-FP16-NEXT: mov v2.s[3], w3 +; CHECK-SD-FP16-NEXT: bsl v0.16b, v5.16b, v4.16b +; CHECK-SD-FP16-NEXT: bsl v1.16b, v2.16b, v3.16b +; CHECK-SD-FP16-NEXT: mov w5, v0.s[1] +; CHECK-SD-FP16-NEXT: mov w6, v0.s[2] +; CHECK-SD-FP16-NEXT: fmov w4, s0 +; CHECK-SD-FP16-NEXT: mov w1, v1.s[1] +; CHECK-SD-FP16-NEXT: mov w2, v1.s[2] +; CHECK-SD-FP16-NEXT: mov w3, v1.s[3] +; CHECK-SD-FP16-NEXT: fmov w0, s1 +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-NOFP16-LABEL: v7f16_i32: +; CHECK-GI-NOFP16: // %bb.0: // %entry +; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[1] +; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[1] +; CHECK-GI-NOFP16-NEXT: mov h4, v1.h[2] +; CHECK-GI-NOFP16-NEXT: mov h5, v0.h[2] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h1 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 +; CHECK-GI-NOFP16-NEXT: fcvt s3, h3 +; CHECK-GI-NOFP16-NEXT: fcvt s4, h4 +; CHECK-GI-NOFP16-NEXT: fcvt s5, h5 +; CHECK-GI-NOFP16-NEXT: fcmp s3, s2 +; CHECK-GI-NOFP16-NEXT: mov h2, v1.h[3] +; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[3] +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s7, s6 +; CHECK-GI-NOFP16-NEXT: mov h6, v1.h[5] +; CHECK-GI-NOFP16-NEXT: mov h7, v0.h[5] +; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 +; CHECK-GI-NOFP16-NEXT: fcvt s3, h3 +; CHECK-GI-NOFP16-NEXT: csetm w9, mi +; CHECK-GI-NOFP16-NEXT: fcmp s5, s4 +; CHECK-GI-NOFP16-NEXT: mov h4, v1.h[4] +; CHECK-GI-NOFP16-NEXT: mov h5, v0.h[4] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h6 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h7 +; CHECK-GI-NOFP16-NEXT: csetm w10, mi +; CHECK-GI-NOFP16-NEXT: fcmp s3, s2 +; CHECK-GI-NOFP16-NEXT: fcvt s2, h4 +; CHECK-GI-NOFP16-NEXT: mov h4, v1.h[6] +; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-GI-NOFP16-NEXT: fcvt s3, h5 +; CHECK-GI-NOFP16-NEXT: mov h5, v0.h[6] +; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-GI-NOFP16-NEXT: csetm w11, mi +; CHECK-GI-NOFP16-NEXT: fcmp s7, s6 +; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: csetm w12, mi +; CHECK-GI-NOFP16-NEXT: fcmp s3, s2 +; CHECK-GI-NOFP16-NEXT: fcvt s2, h4 +; CHECK-GI-NOFP16-NEXT: fcvt s3, h5 +; CHECK-GI-NOFP16-NEXT: fmov s4, w9 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: add x9, sp, #8 +; CHECK-GI-NOFP16-NEXT: csetm w13, mi +; CHECK-GI-NOFP16-NEXT: fmov s5, w13 +; CHECK-GI-NOFP16-NEXT: mov v4.h[1], w8 +; CHECK-GI-NOFP16-NEXT: mov x8, sp +; CHECK-GI-NOFP16-NEXT: fcmp s3, s2 +; CHECK-GI-NOFP16-NEXT: fmov s2, w7 +; CHECK-GI-NOFP16-NEXT: fmov s3, w0 +; CHECK-GI-NOFP16-NEXT: mov v5.h[1], w12 +; CHECK-GI-NOFP16-NEXT: ld1 { v2.s }[1], [x8] +; CHECK-GI-NOFP16-NEXT: mov v3.s[1], w1 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: mov v4.h[2], w10 +; CHECK-GI-NOFP16-NEXT: fcmp s0, s1 +; CHECK-GI-NOFP16-NEXT: fmov s1, w4 +; CHECK-GI-NOFP16-NEXT: ldr s0, [sp, #24] +; CHECK-GI-NOFP16-NEXT: mov v5.h[2], w8 +; CHECK-GI-NOFP16-NEXT: ld1 { v2.s }[2], [x9] +; CHECK-GI-NOFP16-NEXT: add x9, sp, #32 +; CHECK-GI-NOFP16-NEXT: mov v3.s[2], w2 +; CHECK-GI-NOFP16-NEXT: mov v1.s[1], w5 +; CHECK-GI-NOFP16-NEXT: csetm w8, mi +; CHECK-GI-NOFP16-NEXT: mov v4.h[3], w11 +; CHECK-GI-NOFP16-NEXT: ld1 { v0.s }[1], [x9] +; CHECK-GI-NOFP16-NEXT: mov v5.h[3], w8 +; CHECK-GI-NOFP16-NEXT: add x8, sp, #16 +; CHECK-GI-NOFP16-NEXT: ld1 { v2.s }[3], [x8] +; CHECK-GI-NOFP16-NEXT: mov v3.s[3], w3 +; CHECK-GI-NOFP16-NEXT: add x8, sp, #40 +; CHECK-GI-NOFP16-NEXT: mov v1.s[2], w6 +; CHECK-GI-NOFP16-NEXT: sshll v4.4s, v4.4h, #0 +; CHECK-GI-NOFP16-NEXT: ld1 { v0.s }[2], [x8] +; CHECK-GI-NOFP16-NEXT: sshll v5.4s, v5.4h, #0 +; CHECK-GI-NOFP16-NEXT: bit v2.16b, v3.16b, v4.16b +; CHECK-GI-NOFP16-NEXT: bit v0.16b, v1.16b, v5.16b +; CHECK-GI-NOFP16-NEXT: mov w1, v2.s[1] +; CHECK-GI-NOFP16-NEXT: mov w2, v2.s[2] +; CHECK-GI-NOFP16-NEXT: mov w3, v2.s[3] +; CHECK-GI-NOFP16-NEXT: fmov w0, s2 +; CHECK-GI-NOFP16-NEXT: mov w5, v0.s[1] +; CHECK-GI-NOFP16-NEXT: mov w6, v0.s[2] +; CHECK-GI-NOFP16-NEXT: fmov w4, s0 +; CHECK-GI-NOFP16-NEXT: ret +; +; CHECK-GI-FP16-LABEL: v7f16_i32: +; CHECK-GI-FP16: // %bb.0: // %entry +; CHECK-GI-FP16-NEXT: fmov s2, w0 +; CHECK-GI-FP16-NEXT: fmov s3, w7 +; CHECK-GI-FP16-NEXT: mov x8, sp +; CHECK-GI-FP16-NEXT: fmov s5, w4 +; CHECK-GI-FP16-NEXT: ldr s4, [sp, #24] +; CHECK-GI-FP16-NEXT: fcmgt v0.8h, v1.8h, v0.8h +; CHECK-GI-FP16-NEXT: add x9, sp, #32 +; CHECK-GI-FP16-NEXT: mov v2.s[1], w1 +; CHECK-GI-FP16-NEXT: ld1 { v3.s }[1], [x8] +; CHECK-GI-FP16-NEXT: add x8, sp, #8 +; CHECK-GI-FP16-NEXT: mov v5.s[1], w5 +; CHECK-GI-FP16-NEXT: ld1 { v4.s }[1], [x9] +; CHECK-GI-FP16-NEXT: add x9, sp, #16 +; CHECK-GI-FP16-NEXT: sshll v1.4s, v0.4h, #0 +; CHECK-GI-FP16-NEXT: sshll2 v0.4s, v0.8h, #0 +; CHECK-GI-FP16-NEXT: ld1 { v3.s }[2], [x8] +; CHECK-GI-FP16-NEXT: add x8, sp, #40 +; CHECK-GI-FP16-NEXT: mov v2.s[2], w2 +; CHECK-GI-FP16-NEXT: ld1 { v4.s }[2], [x8] +; CHECK-GI-FP16-NEXT: mov v5.s[2], w6 +; CHECK-GI-FP16-NEXT: ld1 { v3.s }[3], [x9] +; CHECK-GI-FP16-NEXT: mov v2.s[3], w3 +; CHECK-GI-FP16-NEXT: bsl v0.16b, v5.16b, v4.16b +; CHECK-GI-FP16-NEXT: bsl v1.16b, v2.16b, v3.16b +; CHECK-GI-FP16-NEXT: mov w5, v0.s[1] +; CHECK-GI-FP16-NEXT: mov w6, v0.s[2] +; CHECK-GI-FP16-NEXT: fmov w4, s0 +; CHECK-GI-FP16-NEXT: mov w1, v1.s[1] +; CHECK-GI-FP16-NEXT: mov w2, v1.s[2] +; CHECK-GI-FP16-NEXT: mov w3, v1.s[3] +; CHECK-GI-FP16-NEXT: fmov w0, s1 +; CHECK-GI-FP16-NEXT: ret +entry: + %c = fcmp olt <7 x half> %a, %b + %s = select <7 x i1> %c, <7 x i32> %d, <7 x i32> %e + ret <7 x i32> %s +} + +define <4 x i32> @v4f16_i32(<4 x half> %a, <4 x half> %b, <4 x i32> %d, <4 x i32> %e) { +; CHECK-SD-NOFP16-LABEL: v4f16_i32: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: fcvtl v0.4s, v0.4h +; CHECK-SD-NOFP16-NEXT: fcvtl v1.4s, v1.4h +; CHECK-SD-NOFP16-NEXT: fcmgt v0.4s, v1.4s, v0.4s +; CHECK-SD-NOFP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: v4f16_i32: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmgt v0.4h, v1.4h, v0.4h +; CHECK-SD-FP16-NEXT: sshll v0.4s, v0.4h, #0 +; CHECK-SD-FP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-NOFP16-LABEL: v4f16_i32: +; CHECK-GI-NOFP16: // %bb.0: // %entry +; CHECK-GI-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0 +; CHECK-GI-NOFP16-NEXT: // kill: def $d1 killed $d1 def $q1 +; CHECK-GI-NOFP16-NEXT: mov h4, v0.h[1] +; CHECK-GI-NOFP16-NEXT: mov h5, v1.h[1] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h1 +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[2] +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[2] +; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[3] +; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[3] +; CHECK-GI-NOFP16-NEXT: fcvt s4, h4 +; CHECK-GI-NOFP16-NEXT: fcvt s5, h5 +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: fcvt s6, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h17 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s4, s5 +; CHECK-GI-NOFP16-NEXT: fmov s4, w8 +; CHECK-GI-NOFP16-NEXT: cset w9, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: fmov s5, w9 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s1 +; CHECK-GI-NOFP16-NEXT: mov v4.h[1], v5.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s0, w8 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: mov v4.h[2], v0.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s0, w8 +; CHECK-GI-NOFP16-NEXT: mov v4.h[3], v0.h[0] +; CHECK-GI-NOFP16-NEXT: ushll v0.4s, v4.4h, #0 +; CHECK-GI-NOFP16-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-NOFP16-NEXT: sshr v0.4s, v0.4s, #31 +; CHECK-GI-NOFP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-GI-NOFP16-NEXT: ret +; +; CHECK-GI-FP16-LABEL: v4f16_i32: +; CHECK-GI-FP16: // %bb.0: // %entry +; CHECK-GI-FP16-NEXT: fcmgt v0.4h, v1.4h, v0.4h +; CHECK-GI-FP16-NEXT: ushll v0.4s, v0.4h, #0 +; CHECK-GI-FP16-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-FP16-NEXT: sshr v0.4s, v0.4s, #31 +; CHECK-GI-FP16-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-GI-FP16-NEXT: ret +entry: + %c = fcmp olt <4 x half> %a, %b + %s = select <4 x i1> %c, <4 x i32> %d, <4 x i32> %e + ret <4 x i32> %s +} + +define <8 x i32> @v8f16_i32(<8 x half> %a, <8 x half> %b, <8 x i32> %d, <8 x i32> %e) { +; CHECK-SD-NOFP16-LABEL: v8f16_i32: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: mov h6, v1.h[5] +; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[5] +; CHECK-SD-NOFP16-NEXT: mov h16, v1.h[4] +; CHECK-SD-NOFP16-NEXT: mov h17, v0.h[4] +; CHECK-SD-NOFP16-NEXT: mov h18, v1.h[6] +; CHECK-SD-NOFP16-NEXT: mov h19, v0.h[6] +; CHECK-SD-NOFP16-NEXT: fcvt s6, h6 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h7 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: fcmp s7, s6 +; CHECK-SD-NOFP16-NEXT: mov h6, v1.h[7] +; CHECK-SD-NOFP16-NEXT: mov h7, v0.h[7] +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v1.h[1] +; CHECK-SD-NOFP16-NEXT: mov h17, v0.h[1] +; CHECK-SD-NOFP16-NEXT: fcvt s6, h6 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h7 +; CHECK-SD-NOFP16-NEXT: csetm w9, mi +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: csetm w10, mi +; CHECK-SD-NOFP16-NEXT: fcmp s7, s6 +; CHECK-SD-NOFP16-NEXT: fcvt s6, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h0 +; CHECK-SD-NOFP16-NEXT: csetm w11, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v1.h[2] +; CHECK-SD-NOFP16-NEXT: mov h17, v0.h[2] +; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[3] +; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[3] +; CHECK-SD-NOFP16-NEXT: csetm w12, mi +; CHECK-SD-NOFP16-NEXT: fcmp s7, s6 +; CHECK-SD-NOFP16-NEXT: fcvt s6, h16 +; CHECK-SD-NOFP16-NEXT: fmov s16, w9 +; CHECK-SD-NOFP16-NEXT: fcvt s7, h17 +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s0, h0 +; CHECK-SD-NOFP16-NEXT: csetm w13, mi +; CHECK-SD-NOFP16-NEXT: fmov s17, w13 +; CHECK-SD-NOFP16-NEXT: mov v16.h[1], w8 +; CHECK-SD-NOFP16-NEXT: fcmp s7, s6 +; CHECK-SD-NOFP16-NEXT: mov v17.h[1], w12 +; CHECK-SD-NOFP16-NEXT: mov v16.h[2], w10 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s0, s1 +; CHECK-SD-NOFP16-NEXT: mov v17.h[2], w8 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: mov v16.h[3], w11 +; CHECK-SD-NOFP16-NEXT: mov v17.h[3], w8 +; CHECK-SD-NOFP16-NEXT: sshll v1.4s, v16.4h, #0 +; CHECK-SD-NOFP16-NEXT: sshll v0.4s, v17.4h, #0 +; CHECK-SD-NOFP16-NEXT: bsl v1.16b, v3.16b, v5.16b +; CHECK-SD-NOFP16-NEXT: bsl v0.16b, v2.16b, v4.16b +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: v8f16_i32: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmgt v0.8h, v1.8h, v0.8h +; CHECK-SD-FP16-NEXT: sshll v6.4s, v0.4h, #0 +; CHECK-SD-FP16-NEXT: sshll2 v0.4s, v0.8h, #0 +; CHECK-SD-FP16-NEXT: mov v1.16b, v0.16b +; CHECK-SD-FP16-NEXT: mov v0.16b, v6.16b +; CHECK-SD-FP16-NEXT: bsl v1.16b, v3.16b, v5.16b +; CHECK-SD-FP16-NEXT: bsl v0.16b, v2.16b, v4.16b +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-NOFP16-LABEL: v8f16_i32: +; CHECK-GI-NOFP16: // %bb.0: // %entry +; CHECK-GI-NOFP16-NEXT: mov h6, v0.h[1] +; CHECK-GI-NOFP16-NEXT: mov h7, v1.h[1] +; CHECK-GI-NOFP16-NEXT: fcvt s16, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h1 +; CHECK-GI-NOFP16-NEXT: mov h18, v0.h[2] +; CHECK-GI-NOFP16-NEXT: mov h19, v1.h[2] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h6 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h7 +; CHECK-GI-NOFP16-NEXT: fcmp s16, s17 +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[3] +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[3] +; CHECK-GI-NOFP16-NEXT: fcvt s18, h18 +; CHECK-GI-NOFP16-NEXT: fcvt s19, h19 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: mov h6, v0.h[4] +; CHECK-GI-NOFP16-NEXT: mov h7, v1.h[4] +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: cset w9, mi +; CHECK-GI-NOFP16-NEXT: fcmp s18, s19 +; CHECK-GI-NOFP16-NEXT: mov h18, v0.h[5] +; CHECK-GI-NOFP16-NEXT: mov h19, v1.h[5] +; CHECK-GI-NOFP16-NEXT: fcvt s6, h6 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h7 +; CHECK-GI-NOFP16-NEXT: cset w10, mi +; CHECK-GI-NOFP16-NEXT: fcmp s16, s17 +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[6] +; CHECK-GI-NOFP16-NEXT: mov h17, v1.h[6] +; CHECK-GI-NOFP16-NEXT: fcvt s18, h18 +; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-GI-NOFP16-NEXT: fcvt s19, h19 +; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-GI-NOFP16-NEXT: cset w11, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: fcvt s6, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s7, h17 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: cset w12, mi +; CHECK-GI-NOFP16-NEXT: fcmp s18, s19 +; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: fmov s16, w12 +; CHECK-GI-NOFP16-NEXT: cset w13, mi +; CHECK-GI-NOFP16-NEXT: fcmp s6, s7 +; CHECK-GI-NOFP16-NEXT: fmov s6, w8 +; CHECK-GI-NOFP16-NEXT: fmov s7, w9 +; CHECK-GI-NOFP16-NEXT: fmov s17, w13 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s1 +; CHECK-GI-NOFP16-NEXT: fmov s0, w10 +; CHECK-GI-NOFP16-NEXT: mov v6.h[1], v7.h[0] +; CHECK-GI-NOFP16-NEXT: mov v16.h[1], v17.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s1, w8 +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: mov v6.h[2], v0.h[0] +; CHECK-GI-NOFP16-NEXT: mov v16.h[2], v1.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s0, w11 +; CHECK-GI-NOFP16-NEXT: fmov s1, w8 +; CHECK-GI-NOFP16-NEXT: mov v6.h[3], v0.h[0] +; CHECK-GI-NOFP16-NEXT: mov v16.h[3], v1.h[0] +; CHECK-GI-NOFP16-NEXT: ushll v0.4s, v6.4h, #0 +; CHECK-GI-NOFP16-NEXT: ushll v1.4s, v16.4h, #0 +; CHECK-GI-NOFP16-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-NOFP16-NEXT: shl v1.4s, v1.4s, #31 +; CHECK-GI-NOFP16-NEXT: sshr v0.4s, v0.4s, #31 +; CHECK-GI-NOFP16-NEXT: sshr v1.4s, v1.4s, #31 +; CHECK-GI-NOFP16-NEXT: bsl v0.16b, v2.16b, v4.16b +; CHECK-GI-NOFP16-NEXT: bsl v1.16b, v3.16b, v5.16b +; CHECK-GI-NOFP16-NEXT: ret +; +; CHECK-GI-FP16-LABEL: v8f16_i32: +; CHECK-GI-FP16: // %bb.0: // %entry +; CHECK-GI-FP16-NEXT: fcmgt v0.8h, v1.8h, v0.8h +; CHECK-GI-FP16-NEXT: ushll v1.4s, v0.4h, #0 +; CHECK-GI-FP16-NEXT: ushll2 v0.4s, v0.8h, #0 +; CHECK-GI-FP16-NEXT: shl v1.4s, v1.4s, #31 +; CHECK-GI-FP16-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-FP16-NEXT: sshr v1.4s, v1.4s, #31 +; CHECK-GI-FP16-NEXT: sshr v6.4s, v0.4s, #31 +; CHECK-GI-FP16-NEXT: mov v0.16b, v1.16b +; CHECK-GI-FP16-NEXT: mov v1.16b, v6.16b +; CHECK-GI-FP16-NEXT: bsl v0.16b, v2.16b, v4.16b +; CHECK-GI-FP16-NEXT: bsl v1.16b, v3.16b, v5.16b +; CHECK-GI-FP16-NEXT: ret +entry: + %c = fcmp olt <8 x half> %a, %b + %s = select <8 x i1> %c, <8 x i32> %d, <8 x i32> %e + ret <8 x i32> %s +} + +define <16 x i32> @v16f16_i32(<16 x half> %a, <16 x half> %b, <16 x i32> %d, <16 x i32> %e) { +; CHECK-SD-NOFP16-LABEL: v16f16_i32: +; CHECK-SD-NOFP16: // %bb.0: // %entry +; CHECK-SD-NOFP16-NEXT: mov h16, v3.h[5] +; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[5] +; CHECK-SD-NOFP16-NEXT: mov h18, v3.h[4] +; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[4] +; CHECK-SD-NOFP16-NEXT: mov h20, v3.h[6] +; CHECK-SD-NOFP16-NEXT: mov h21, v1.h[6] +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: fcvt s20, h20 +; CHECK-SD-NOFP16-NEXT: fcvt s21, h21 +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v3.h[7] +; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[7] +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: mov h18, v3.h[1] +; CHECK-SD-NOFP16-NEXT: mov h19, v1.h[1] +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: csetm w10, mi +; CHECK-SD-NOFP16-NEXT: fcmp s21, s20 +; CHECK-SD-NOFP16-NEXT: fcvt s20, h3 +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: fcvt s21, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: csetm w9, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v3.h[2] +; CHECK-SD-NOFP16-NEXT: mov h17, v1.h[2] +; CHECK-SD-NOFP16-NEXT: mov h3, v3.h[3] +; CHECK-SD-NOFP16-NEXT: mov h1, v1.h[3] +; CHECK-SD-NOFP16-NEXT: csetm w11, mi +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: mov h18, v2.h[5] +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: mov h19, v0.h[5] +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h3 +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: csetm w12, mi +; CHECK-SD-NOFP16-NEXT: fcmp s21, s20 +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: csetm w14, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v2.h[4] +; CHECK-SD-NOFP16-NEXT: mov h17, v0.h[4] +; CHECK-SD-NOFP16-NEXT: csetm w13, mi +; CHECK-SD-NOFP16-NEXT: fcmp s1, s3 +; CHECK-SD-NOFP16-NEXT: mov h1, v2.h[6] +; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[6] +; CHECK-SD-NOFP16-NEXT: fcvt s16, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h17 +; CHECK-SD-NOFP16-NEXT: csetm w15, mi +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: mov h18, v2.h[7] +; CHECK-SD-NOFP16-NEXT: mov h19, v0.h[7] +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h3 +; CHECK-SD-NOFP16-NEXT: csetm w16, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: mov h16, v2.h[1] +; CHECK-SD-NOFP16-NEXT: mov h17, v0.h[1] +; CHECK-SD-NOFP16-NEXT: fcvt s18, h18 +; CHECK-SD-NOFP16-NEXT: fcvt s19, h19 +; CHECK-SD-NOFP16-NEXT: csetm w17, mi +; CHECK-SD-NOFP16-NEXT: fcmp s3, s1 +; CHECK-SD-NOFP16-NEXT: fcvt s1, h16 +; CHECK-SD-NOFP16-NEXT: fcvt s16, h2 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h17 +; CHECK-SD-NOFP16-NEXT: fcvt s17, h0 +; CHECK-SD-NOFP16-NEXT: csetm w18, mi +; CHECK-SD-NOFP16-NEXT: fcmp s19, s18 +; CHECK-SD-NOFP16-NEXT: fmov s18, w14 +; CHECK-SD-NOFP16-NEXT: fmov s19, w17 +; CHECK-SD-NOFP16-NEXT: csetm w0, mi +; CHECK-SD-NOFP16-NEXT: fcmp s3, s1 +; CHECK-SD-NOFP16-NEXT: mov h1, v2.h[2] +; CHECK-SD-NOFP16-NEXT: mov h3, v0.h[2] +; CHECK-SD-NOFP16-NEXT: mov h2, v2.h[3] +; CHECK-SD-NOFP16-NEXT: mov h0, v0.h[3] +; CHECK-SD-NOFP16-NEXT: mov v18.h[1], w12 +; CHECK-SD-NOFP16-NEXT: mov v19.h[1], w16 +; CHECK-SD-NOFP16-NEXT: csetm w1, mi +; CHECK-SD-NOFP16-NEXT: fcmp s17, s16 +; CHECK-SD-NOFP16-NEXT: fmov s16, w10 +; CHECK-SD-NOFP16-NEXT: fcvt s1, h1 +; CHECK-SD-NOFP16-NEXT: fcvt s3, h3 +; CHECK-SD-NOFP16-NEXT: fcvt s2, h2 +; CHECK-SD-NOFP16-NEXT: fcvt s0, h0 +; CHECK-SD-NOFP16-NEXT: csetm w2, mi +; CHECK-SD-NOFP16-NEXT: mov v16.h[1], w8 +; CHECK-SD-NOFP16-NEXT: mov v18.h[2], w13 +; CHECK-SD-NOFP16-NEXT: fmov s17, w2 +; CHECK-SD-NOFP16-NEXT: mov v19.h[2], w18 +; CHECK-SD-NOFP16-NEXT: fcmp s3, s1 +; CHECK-SD-NOFP16-NEXT: mov v17.h[1], w1 +; CHECK-SD-NOFP16-NEXT: mov v16.h[2], w9 +; CHECK-SD-NOFP16-NEXT: mov v18.h[3], w15 +; CHECK-SD-NOFP16-NEXT: mov v19.h[3], w0 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: fcmp s0, s2 +; CHECK-SD-NOFP16-NEXT: mov v17.h[2], w8 +; CHECK-SD-NOFP16-NEXT: mov v16.h[3], w11 +; CHECK-SD-NOFP16-NEXT: csetm w8, mi +; CHECK-SD-NOFP16-NEXT: mov v17.h[3], w8 +; CHECK-SD-NOFP16-NEXT: sshll v2.4s, v16.4h, #0 +; CHECK-SD-NOFP16-NEXT: sshll v16.4s, v18.4h, #0 +; CHECK-SD-NOFP16-NEXT: ldp q0, q18, [sp] +; CHECK-SD-NOFP16-NEXT: sshll v1.4s, v17.4h, #0 +; CHECK-SD-NOFP16-NEXT: sshll v17.4s, v19.4h, #0 +; CHECK-SD-NOFP16-NEXT: ldp q19, q3, [sp, #32] +; CHECK-SD-NOFP16-NEXT: bit v0.16b, v4.16b, v1.16b +; CHECK-SD-NOFP16-NEXT: mov v1.16b, v17.16b +; CHECK-SD-NOFP16-NEXT: bit v3.16b, v7.16b, v2.16b +; CHECK-SD-NOFP16-NEXT: mov v2.16b, v16.16b +; CHECK-SD-NOFP16-NEXT: bsl v1.16b, v5.16b, v18.16b +; CHECK-SD-NOFP16-NEXT: bsl v2.16b, v6.16b, v19.16b +; CHECK-SD-NOFP16-NEXT: ret +; +; CHECK-SD-FP16-LABEL: v16f16_i32: +; CHECK-SD-FP16: // %bb.0: // %entry +; CHECK-SD-FP16-NEXT: fcmgt v0.8h, v2.8h, v0.8h +; CHECK-SD-FP16-NEXT: fcmgt v1.8h, v3.8h, v1.8h +; CHECK-SD-FP16-NEXT: ldp q2, q20, [sp] +; CHECK-SD-FP16-NEXT: ldp q18, q19, [sp, #32] +; CHECK-SD-FP16-NEXT: sshll v3.4s, v0.4h, #0 +; CHECK-SD-FP16-NEXT: sshll v16.4s, v1.4h, #0 +; CHECK-SD-FP16-NEXT: sshll2 v17.4s, v1.8h, #0 +; CHECK-SD-FP16-NEXT: sshll2 v1.4s, v0.8h, #0 +; CHECK-SD-FP16-NEXT: mov v0.16b, v3.16b +; CHECK-SD-FP16-NEXT: mov v3.16b, v17.16b +; CHECK-SD-FP16-NEXT: bsl v1.16b, v5.16b, v20.16b +; CHECK-SD-FP16-NEXT: bsl v0.16b, v4.16b, v2.16b +; CHECK-SD-FP16-NEXT: mov v2.16b, v16.16b +; CHECK-SD-FP16-NEXT: bsl v3.16b, v7.16b, v19.16b +; CHECK-SD-FP16-NEXT: bsl v2.16b, v6.16b, v18.16b +; CHECK-SD-FP16-NEXT: ret +; +; CHECK-GI-NOFP16-LABEL: v16f16_i32: +; CHECK-GI-NOFP16: // %bb.0: // %entry +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[1] +; CHECK-GI-NOFP16-NEXT: mov h17, v2.h[1] +; CHECK-GI-NOFP16-NEXT: fcvt s18, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s19, h2 +; CHECK-GI-NOFP16-NEXT: mov h20, v0.h[2] +; CHECK-GI-NOFP16-NEXT: mov h21, v2.h[2] +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: fcmp s18, s19 +; CHECK-GI-NOFP16-NEXT: mov h18, v0.h[3] +; CHECK-GI-NOFP16-NEXT: mov h19, v2.h[3] +; CHECK-GI-NOFP16-NEXT: fcvt s20, h20 +; CHECK-GI-NOFP16-NEXT: fcvt s21, h21 +; CHECK-GI-NOFP16-NEXT: cset w14, mi +; CHECK-GI-NOFP16-NEXT: fcmp s16, s17 +; CHECK-GI-NOFP16-NEXT: mov h16, v0.h[4] +; CHECK-GI-NOFP16-NEXT: mov h17, v2.h[4] +; CHECK-GI-NOFP16-NEXT: fcvt s18, h18 +; CHECK-GI-NOFP16-NEXT: fcvt s19, h19 +; CHECK-GI-NOFP16-NEXT: cset w15, mi +; CHECK-GI-NOFP16-NEXT: fcmp s20, s21 +; CHECK-GI-NOFP16-NEXT: mov h20, v0.h[5] +; CHECK-GI-NOFP16-NEXT: mov h21, v2.h[5] +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: cset w9, mi +; CHECK-GI-NOFP16-NEXT: fcmp s18, s19 +; CHECK-GI-NOFP16-NEXT: mov h18, v0.h[6] +; CHECK-GI-NOFP16-NEXT: mov h19, v2.h[6] +; CHECK-GI-NOFP16-NEXT: fcvt s20, h20 +; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[7] +; CHECK-GI-NOFP16-NEXT: fcvt s21, h21 +; CHECK-GI-NOFP16-NEXT: mov h2, v2.h[7] +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fcmp s16, s17 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h18 +; CHECK-GI-NOFP16-NEXT: fcvt s18, h1 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h19 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s19, h3 +; CHECK-GI-NOFP16-NEXT: cset w16, mi +; CHECK-GI-NOFP16-NEXT: fcmp s20, s21 +; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 +; CHECK-GI-NOFP16-NEXT: cset w17, mi +; CHECK-GI-NOFP16-NEXT: fcmp s16, s17 +; CHECK-GI-NOFP16-NEXT: mov h16, v1.h[1] +; CHECK-GI-NOFP16-NEXT: mov h17, v3.h[1] +; CHECK-GI-NOFP16-NEXT: cset w11, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s2 +; CHECK-GI-NOFP16-NEXT: mov h0, v1.h[2] +; CHECK-GI-NOFP16-NEXT: mov h2, v3.h[2] +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: cset w10, mi +; CHECK-GI-NOFP16-NEXT: fcmp s18, s19 +; CHECK-GI-NOFP16-NEXT: mov h18, v1.h[3] +; CHECK-GI-NOFP16-NEXT: mov h19, v3.h[3] +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 +; CHECK-GI-NOFP16-NEXT: cset w18, mi +; CHECK-GI-NOFP16-NEXT: fcmp s16, s17 +; CHECK-GI-NOFP16-NEXT: mov h16, v1.h[4] +; CHECK-GI-NOFP16-NEXT: mov h17, v3.h[4] +; CHECK-GI-NOFP16-NEXT: fcvt s18, h18 +; CHECK-GI-NOFP16-NEXT: fcvt s19, h19 +; CHECK-GI-NOFP16-NEXT: cset w0, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s2 +; CHECK-GI-NOFP16-NEXT: mov h0, v1.h[5] +; CHECK-GI-NOFP16-NEXT: mov h2, v3.h[5] +; CHECK-GI-NOFP16-NEXT: fcvt s16, h16 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h17 +; CHECK-GI-NOFP16-NEXT: cset w13, mi +; CHECK-GI-NOFP16-NEXT: fcmp s18, s19 +; CHECK-GI-NOFP16-NEXT: mov h18, v1.h[6] +; CHECK-GI-NOFP16-NEXT: mov h19, v3.h[6] +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: mov h1, v1.h[7] +; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 +; CHECK-GI-NOFP16-NEXT: mov h3, v3.h[7] +; CHECK-GI-NOFP16-NEXT: cset w12, mi +; CHECK-GI-NOFP16-NEXT: fcmp s16, s17 +; CHECK-GI-NOFP16-NEXT: fcvt s16, h18 +; CHECK-GI-NOFP16-NEXT: fmov s18, w15 +; CHECK-GI-NOFP16-NEXT: fcvt s17, h19 +; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: fmov s19, w17 +; CHECK-GI-NOFP16-NEXT: cset w1, mi +; CHECK-GI-NOFP16-NEXT: fcmp s0, s2 +; CHECK-GI-NOFP16-NEXT: fcvt s3, h3 +; CHECK-GI-NOFP16-NEXT: fmov s2, w16 +; CHECK-GI-NOFP16-NEXT: fmov s0, w14 +; CHECK-GI-NOFP16-NEXT: cset w14, mi +; CHECK-GI-NOFP16-NEXT: fcmp s16, s17 +; CHECK-GI-NOFP16-NEXT: fmov s16, w18 +; CHECK-GI-NOFP16-NEXT: mov v2.h[1], v19.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s17, w0 +; CHECK-GI-NOFP16-NEXT: fmov s19, w14 +; CHECK-GI-NOFP16-NEXT: mov v0.h[1], v18.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s18, w1 +; CHECK-GI-NOFP16-NEXT: cset w14, mi +; CHECK-GI-NOFP16-NEXT: fcmp s1, s3 +; CHECK-GI-NOFP16-NEXT: fmov s1, w11 +; CHECK-GI-NOFP16-NEXT: mov v16.h[1], v17.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s17, w9 +; CHECK-GI-NOFP16-NEXT: fmov s3, w14 +; CHECK-GI-NOFP16-NEXT: mov v18.h[1], v19.h[0] +; CHECK-GI-NOFP16-NEXT: mov v2.h[2], v1.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s1, w13 +; CHECK-GI-NOFP16-NEXT: mov v0.h[2], v17.h[0] +; CHECK-GI-NOFP16-NEXT: mov v16.h[2], v1.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s1, w8 +; CHECK-GI-NOFP16-NEXT: mov v18.h[2], v3.h[0] +; CHECK-GI-NOFP16-NEXT: cset w8, mi +; CHECK-GI-NOFP16-NEXT: fmov s3, w8 +; CHECK-GI-NOFP16-NEXT: mov v0.h[3], v1.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s1, w10 +; CHECK-GI-NOFP16-NEXT: mov v18.h[3], v3.h[0] +; CHECK-GI-NOFP16-NEXT: mov v2.h[3], v1.h[0] +; CHECK-GI-NOFP16-NEXT: fmov s1, w12 +; CHECK-GI-NOFP16-NEXT: ushll v0.4s, v0.4h, #0 +; CHECK-GI-NOFP16-NEXT: mov v16.h[3], v1.h[0] +; CHECK-GI-NOFP16-NEXT: ushll v3.4s, v18.4h, #0 +; CHECK-GI-NOFP16-NEXT: ushll v1.4s, v2.4h, #0 +; CHECK-GI-NOFP16-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-NOFP16-NEXT: ldp q18, q19, [sp, #32] +; CHECK-GI-NOFP16-NEXT: shl v3.4s, v3.4s, #31 +; CHECK-GI-NOFP16-NEXT: ushll v2.4s, v16.4h, #0 +; CHECK-GI-NOFP16-NEXT: shl v1.4s, v1.4s, #31 +; CHECK-GI-NOFP16-NEXT: sshr v0.4s, v0.4s, #31 +; CHECK-GI-NOFP16-NEXT: ldp q16, q17, [sp] +; CHECK-GI-NOFP16-NEXT: sshr v3.4s, v3.4s, #31 +; CHECK-GI-NOFP16-NEXT: shl v2.4s, v2.4s, #31 +; CHECK-GI-NOFP16-NEXT: sshr v1.4s, v1.4s, #31 +; CHECK-GI-NOFP16-NEXT: bsl v0.16b, v4.16b, v16.16b +; CHECK-GI-NOFP16-NEXT: bsl v3.16b, v7.16b, v19.16b +; CHECK-GI-NOFP16-NEXT: sshr v2.4s, v2.4s, #31 +; CHECK-GI-NOFP16-NEXT: bsl v1.16b, v5.16b, v17.16b +; CHECK-GI-NOFP16-NEXT: bsl v2.16b, v6.16b, v18.16b +; CHECK-GI-NOFP16-NEXT: ret +; +; CHECK-GI-FP16-LABEL: v16f16_i32: +; CHECK-GI-FP16: // %bb.0: // %entry +; CHECK-GI-FP16-NEXT: fcmgt v0.8h, v2.8h, v0.8h +; CHECK-GI-FP16-NEXT: fcmgt v1.8h, v3.8h, v1.8h +; CHECK-GI-FP16-NEXT: ldp q2, q20, [sp] +; CHECK-GI-FP16-NEXT: ldp q18, q19, [sp, #32] +; CHECK-GI-FP16-NEXT: sshll v3.4s, v0.4h, #0 +; CHECK-GI-FP16-NEXT: sshll v16.4s, v1.4h, #0 +; CHECK-GI-FP16-NEXT: sshll2 v17.4s, v1.8h, #0 +; CHECK-GI-FP16-NEXT: sshll2 v1.4s, v0.8h, #0 +; CHECK-GI-FP16-NEXT: mov v0.16b, v3.16b +; CHECK-GI-FP16-NEXT: mov v3.16b, v17.16b +; CHECK-GI-FP16-NEXT: bsl v1.16b, v5.16b, v20.16b +; CHECK-GI-FP16-NEXT: bsl v0.16b, v4.16b, v2.16b +; CHECK-GI-FP16-NEXT: mov v2.16b, v16.16b +; CHECK-GI-FP16-NEXT: bsl v3.16b, v7.16b, v19.16b +; CHECK-GI-FP16-NEXT: bsl v2.16b, v6.16b, v18.16b +; CHECK-GI-FP16-NEXT: ret +entry: + %c = fcmp olt <16 x half> %a, %b + %s = select <16 x i1> %c, <16 x i32> %d, <16 x i32> %e + ret <16 x i32> %s } diff --git a/llvm/test/CodeGen/AArch64/icmp.ll b/llvm/test/CodeGen/AArch64/icmp.ll new file mode 100644 index 000000000000..8505caa39959 --- /dev/null +++ b/llvm/test/CodeGen/AArch64/icmp.ll @@ -0,0 +1,257 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4 +; RUN: llc -mtriple=aarch64-none-eabi -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD +; RUN: llc -mtriple=aarch64-none-eabi -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI + +; CHECK-GI: warning: Instruction selection used fallback path for v3i64_i64 +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v4i64_i64 +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v3i32_i32 +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v16i16_i16 +; CHECK-GI-NEXT: warning: Instruction selection used fallback path for v32i8_i8 + +define i64 @i64_i64(i64 %a, i64 %b, i64 %d, i64 %e) { +; CHECK-LABEL: i64_i64: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmp x0, x1 +; CHECK-NEXT: csel x0, x2, x3, lt +; CHECK-NEXT: ret +entry: + %c = icmp slt i64 %a, %b + %s = select i1 %c, i64 %d, i64 %e + ret i64 %s +} + +define i32 @i32_i32(i32 %a, i32 %b, i32 %d, i32 %e) { +; CHECK-LABEL: i32_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmp w0, w1 +; CHECK-NEXT: csel w0, w2, w3, lt +; CHECK-NEXT: ret +entry: + %c = icmp slt i32 %a, %b + %s = select i1 %c, i32 %d, i32 %e + ret i32 %s +} + +define i16 @i16_i16(i16 %a, i16 %b, i16 %d, i16 %e) { +; CHECK-LABEL: i16_i16: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: sxth w8, w0 +; CHECK-NEXT: cmp w8, w1, sxth +; CHECK-NEXT: csel w0, w2, w3, lt +; CHECK-NEXT: ret +entry: + %c = icmp slt i16 %a, %b + %s = select i1 %c, i16 %d, i16 %e + ret i16 %s +} + +define i8 @i8_i8(i8 %a, i8 %b, i8 %d, i8 %e) { +; CHECK-LABEL: i8_i8: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: sxtb w8, w0 +; CHECK-NEXT: cmp w8, w1, sxtb +; CHECK-NEXT: csel w0, w2, w3, lt +; CHECK-NEXT: ret +entry: + %c = icmp slt i8 %a, %b + %s = select i1 %c, i8 %d, i8 %e + ret i8 %s +} + +define <2 x i64> @v2i64_i64(<2 x i64> %a, <2 x i64> %b, <2 x i64> %d, <2 x i64> %e) { +; CHECK-LABEL: v2i64_i64: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v0.2d, v1.2d, v0.2d +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = icmp slt <2 x i64> %a, %b + %s = select <2 x i1> %c, <2 x i64> %d, <2 x i64> %e + ret <2 x i64> %s +} + +define <3 x i64> @v3i64_i64(<3 x i64> %a, <3 x i64> %b, <3 x i64> %d, <3 x i64> %e) { +; CHECK-LABEL: v3i64_i64: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: // kill: def $d4 killed $d4 def $q4 +; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3 +; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1 +; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0 +; CHECK-NEXT: // kill: def $d6 killed $d6 def $q6 +; CHECK-NEXT: // kill: def $d7 killed $d7 def $q7 +; CHECK-NEXT: // kill: def $d5 killed $d5 def $q5 +; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2 +; CHECK-NEXT: ldr d16, [sp, #24] +; CHECK-NEXT: ldr d17, [sp] +; CHECK-NEXT: mov v3.d[1], v4.d[0] +; CHECK-NEXT: mov v0.d[1], v1.d[0] +; CHECK-NEXT: mov v6.d[1], v7.d[0] +; CHECK-NEXT: ldp d1, d4, [sp, #8] +; CHECK-NEXT: mov v1.d[1], v4.d[0] +; CHECK-NEXT: cmgt v0.2d, v3.2d, v0.2d +; CHECK-NEXT: bsl v0.16b, v6.16b, v1.16b +; CHECK-NEXT: cmgt v1.2d, v5.2d, v2.2d +; CHECK-NEXT: mov v2.16b, v1.16b +; CHECK-NEXT: ext v1.16b, v0.16b, v0.16b, #8 +; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q0 +; CHECK-NEXT: // kill: def $d1 killed $d1 killed $q1 +; CHECK-NEXT: bsl v2.16b, v17.16b, v16.16b +; CHECK-NEXT: // kill: def $d2 killed $d2 killed $q2 +; CHECK-NEXT: ret +entry: + %c = icmp slt <3 x i64> %a, %b + %s = select <3 x i1> %c, <3 x i64> %d, <3 x i64> %e + ret <3 x i64> %s +} + +define <4 x i64> @v4i64_i64(<4 x i64> %a, <4 x i64> %b, <4 x i64> %d, <4 x i64> %e) { +; CHECK-LABEL: v4i64_i64: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v1.2d, v3.2d, v1.2d +; CHECK-NEXT: cmgt v0.2d, v2.2d, v0.2d +; CHECK-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-NEXT: ret +entry: + %c = icmp slt <4 x i64> %a, %b + %s = select <4 x i1> %c, <4 x i64> %d, <4 x i64> %e + ret <4 x i64> %s +} + +define <2 x i32> @v2i32_i32(<2 x i32> %a, <2 x i32> %b, <2 x i32> %d, <2 x i32> %e) { +; CHECK-LABEL: v2i32_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v0.2s, v1.2s, v0.2s +; CHECK-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-NEXT: ret +entry: + %c = icmp slt <2 x i32> %a, %b + %s = select <2 x i1> %c, <2 x i32> %d, <2 x i32> %e + ret <2 x i32> %s +} + +define <3 x i32> @v3i32_i32(<3 x i32> %a, <3 x i32> %b, <3 x i32> %d, <3 x i32> %e) { +; CHECK-LABEL: v3i32_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v0.4s, v1.4s, v0.4s +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = icmp slt <3 x i32> %a, %b + %s = select <3 x i1> %c, <3 x i32> %d, <3 x i32> %e + ret <3 x i32> %s +} + +define <4 x i32> @v4i32_i32(<4 x i32> %a, <4 x i32> %b, <4 x i32> %d, <4 x i32> %e) { +; CHECK-LABEL: v4i32_i32: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v0.4s, v1.4s, v0.4s +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = icmp slt <4 x i32> %a, %b + %s = select <4 x i1> %c, <4 x i32> %d, <4 x i32> %e + ret <4 x i32> %s +} + +define <8 x i32> @v8i32_i32(<8 x i32> %a, <8 x i32> %b, <8 x i32> %d, <8 x i32> %e) { +; CHECK-SD-LABEL: v8i32_i32: +; CHECK-SD: // %bb.0: // %entry +; CHECK-SD-NEXT: cmgt v1.4s, v3.4s, v1.4s +; CHECK-SD-NEXT: cmgt v0.4s, v2.4s, v0.4s +; CHECK-SD-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-SD-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: v8i32_i32: +; CHECK-GI: // %bb.0: // %entry +; CHECK-GI-NEXT: cmgt v0.4s, v2.4s, v0.4s +; CHECK-GI-NEXT: cmgt v1.4s, v3.4s, v1.4s +; CHECK-GI-NEXT: shl v0.4s, v0.4s, #31 +; CHECK-GI-NEXT: shl v1.4s, v1.4s, #31 +; CHECK-GI-NEXT: sshr v0.4s, v0.4s, #31 +; CHECK-GI-NEXT: sshr v1.4s, v1.4s, #31 +; CHECK-GI-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-GI-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-GI-NEXT: ret +entry: + %c = icmp slt <8 x i32> %a, %b + %s = select <8 x i1> %c, <8 x i32> %d, <8 x i32> %e + ret <8 x i32> %s +} + +define <4 x i16> @v4i16_i16(<4 x i16> %a, <4 x i16> %b, <4 x i16> %d, <4 x i16> %e) { +; CHECK-LABEL: v4i16_i16: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v0.4h, v1.4h, v0.4h +; CHECK-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-NEXT: ret +entry: + %c = icmp slt <4 x i16> %a, %b + %s = select <4 x i1> %c, <4 x i16> %d, <4 x i16> %e + ret <4 x i16> %s +} + +define <8 x i16> @v8i16_i16(<8 x i16> %a, <8 x i16> %b, <8 x i16> %d, <8 x i16> %e) { +; CHECK-LABEL: v8i16_i16: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v0.8h, v1.8h, v0.8h +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = icmp slt <8 x i16> %a, %b + %s = select <8 x i1> %c, <8 x i16> %d, <8 x i16> %e + ret <8 x i16> %s +} + +define <16 x i16> @v16i16_i16(<16 x i16> %a, <16 x i16> %b, <16 x i16> %d, <16 x i16> %e) { +; CHECK-LABEL: v16i16_i16: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v1.8h, v3.8h, v1.8h +; CHECK-NEXT: cmgt v0.8h, v2.8h, v0.8h +; CHECK-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-NEXT: ret +entry: + %c = icmp slt <16 x i16> %a, %b + %s = select <16 x i1> %c, <16 x i16> %d, <16 x i16> %e + ret <16 x i16> %s +} + +define <8 x i8> @v8i8_i8(<8 x i8> %a, <8 x i8> %b, <8 x i8> %d, <8 x i8> %e) { +; CHECK-LABEL: v8i8_i8: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v0.8b, v1.8b, v0.8b +; CHECK-NEXT: bsl v0.8b, v2.8b, v3.8b +; CHECK-NEXT: ret +entry: + %c = icmp slt <8 x i8> %a, %b + %s = select <8 x i1> %c, <8 x i8> %d, <8 x i8> %e + ret <8 x i8> %s +} + +define <16 x i8> @v16i8_i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> %d, <16 x i8> %e) { +; CHECK-LABEL: v16i8_i8: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v0.16b, v1.16b, v0.16b +; CHECK-NEXT: bsl v0.16b, v2.16b, v3.16b +; CHECK-NEXT: ret +entry: + %c = icmp slt <16 x i8> %a, %b + %s = select <16 x i1> %c, <16 x i8> %d, <16 x i8> %e + ret <16 x i8> %s +} + +define <32 x i8> @v32i8_i8(<32 x i8> %a, <32 x i8> %b, <32 x i8> %d, <32 x i8> %e) { +; CHECK-LABEL: v32i8_i8: +; CHECK: // %bb.0: // %entry +; CHECK-NEXT: cmgt v1.16b, v3.16b, v1.16b +; CHECK-NEXT: cmgt v0.16b, v2.16b, v0.16b +; CHECK-NEXT: bsl v1.16b, v5.16b, v7.16b +; CHECK-NEXT: bsl v0.16b, v4.16b, v6.16b +; CHECK-NEXT: ret +entry: + %c = icmp slt <32 x i8> %a, %b + %s = select <32 x i1> %c, <32 x i8> %d, <32 x i8> %e + ret <32 x i8> %s +}