[llvm] 0e75712 - [AArch64][GISel] Expand lowering for fminimum and fmaximum
David Green via llvm-commits
llvm-commits at lists.llvm.org
Sun Aug 6 06:36:58 PDT 2023
Author: David Green
Date: 2023-08-06T14:36:52+01:00
New Revision: 0e757122a1ca170695510b2d372ba7163b2dd96d
URL: https://github.com/llvm/llvm-project/commit/0e757122a1ca170695510b2d372ba7163b2dd96d
DIFF: https://github.com/llvm/llvm-project/commit/0e757122a1ca170695510b2d372ba7163b2dd96d.diff
LOG: [AArch64][GISel] Expand lowering for fminimum and fmaximum
This replicates the G_FMINNUM and G_FMAXNUM lowering to G_FMINIMUM and
G_FMAXIMUM, reusing the same action definition for lowering.
Added:
llvm/test/CodeGen/AArch64/fminimummaximum.ll
Modified:
llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
Removed:
################################################################################
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index 70b92b4283ce16..ecdab74ce0593e 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -919,7 +919,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
// TODO: Vector types.
getActionDefinitionsBuilder({G_SADDSAT, G_SSUBSAT}).lowerIf(isScalar(0));
- getActionDefinitionsBuilder({G_FMAXNUM, G_FMINNUM})
+ getActionDefinitionsBuilder({G_FMAXNUM, G_FMINNUM, G_FMAXIMUM, G_FMINIMUM})
.legalFor({MinFPScalar, s32, s64, v2s32, v4s32, v2s64})
.legalIf([=](const LegalityQuery &Query) {
const auto &Ty = Query.Types[0];
@@ -932,17 +932,6 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
.clampNumElements(0, v2s64, v2s64)
.moreElementsToNextPow2(0);
- getActionDefinitionsBuilder({G_FMAXIMUM, G_FMINIMUM})
- .legalFor({MinFPScalar, s32, s64, v2s32, v4s32, v2s64})
- .legalIf([=](const LegalityQuery &Query) {
- const auto &Ty = Query.Types[0];
- return (Ty == v8s16 || Ty == v4s16) && HasFP16;
- })
- .minScalar(0, MinFPScalar)
- .clampNumElements(0, v4s16, v8s16)
- .clampNumElements(0, v2s32, v4s32)
- .clampNumElements(0, v2s64, v2s64);
-
// TODO: Libcall support for s128.
// TODO: s16 should be legal with full FP16 support.
getActionDefinitionsBuilder({G_LROUND, G_LLROUND})
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
index 3f157ffdecfc42..65ab99627da29a 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir
@@ -543,6 +543,7 @@
# DEBUG-NEXT: .. type index coverage check SKIPPED: user-defined predicate detected
# DEBUG-NEXT: .. imm index coverage check SKIPPED: user-defined predicate detected
# DEBUG-NEXT: G_FMAXIMUM (opcode {{[0-9]+}}): 1 type index
+# DEBUG-NEXT: .. opcode {{[0-9]+}} is aliased to {{[0-9]+}}
# DEBUG-NEXT: .. type index coverage check SKIPPED: user-defined predicate detected
# DEBUG-NEXT: .. imm index coverage check SKIPPED: user-defined predicate detected
# DEBUG-NEXT: G_PTR_ADD (opcode {{[0-9]+}}): 2 type indices, 0 imm indices
diff --git a/llvm/test/CodeGen/AArch64/fminimummaximum.ll b/llvm/test/CodeGen/AArch64/fminimummaximum.ll
new file mode 100644
index 00000000000000..2492db229ceff5
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/fminimummaximum.ll
@@ -0,0 +1,1442 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-none-eabi -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-NOFP16-SD
+; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-FP16-SD
+; RUN: llc -mtriple=aarch64-none-eabi -global-isel -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-NOFP16-GI
+; RUN: llc -mtriple=aarch64-none-eabi -mattr=+fullfp16 -global-isel -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-FP16-GI
+
+define double @min_f64(double %a, double %b) {
+; CHECK-LABEL: min_f64:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmin d0, d0, d1
+; CHECK-NEXT: ret
+entry:
+ %c = call double @llvm.minimum.f64(double %a, double %b)
+ ret double %c
+}
+
+define double @max_f64(double %a, double %b) {
+; CHECK-LABEL: max_f64:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmax d0, d0, d1
+; CHECK-NEXT: ret
+entry:
+ %c = call double @llvm.maximum.f64(double %a, double %b)
+ ret double %c
+}
+
+define float @min_f32(float %a, float %b) {
+; CHECK-LABEL: min_f32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmin s0, s0, s1
+; CHECK-NEXT: ret
+entry:
+ %c = call float @llvm.minimum.f32(float %a, float %b)
+ ret float %c
+}
+
+define float @max_f32(float %a, float %b) {
+; CHECK-LABEL: max_f32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmax s0, s0, s1
+; CHECK-NEXT: ret
+entry:
+ %c = call float @llvm.maximum.f32(float %a, float %b)
+ ret float %c
+}
+
+define half @min_f16(half %a, half %b) {
+; CHECK-NOFP16-SD-LABEL: min_f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT: fmin s0, s0, s1
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: min_f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmin h0, h0, h1
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: min_f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: fcvt s0, h0
+; CHECK-NOFP16-GI-NEXT: fcvt s1, h1
+; CHECK-NOFP16-GI-NEXT: fmin s0, s0, s1
+; CHECK-NOFP16-GI-NEXT: fcvt h0, s0
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: min_f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: fmin h0, h0, h1
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call half @llvm.minimum.f16(half %a, half %b)
+ ret half %c
+}
+
+define half @max_f16(half %a, half %b) {
+; CHECK-NOFP16-SD-LABEL: max_f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT: fmax s0, s0, s1
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: max_f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmax h0, h0, h1
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: max_f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: fcvt s0, h0
+; CHECK-NOFP16-GI-NEXT: fcvt s1, h1
+; CHECK-NOFP16-GI-NEXT: fmax s0, s0, s1
+; CHECK-NOFP16-GI-NEXT: fcvt h0, s0
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: max_f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: fmax h0, h0, h1
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call half @llvm.maximum.f16(half %a, half %b)
+ ret half %c
+}
+
+define <2 x double> @min_v2f64(<2 x double> %a, <2 x double> %b) {
+; CHECK-LABEL: min_v2f64:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmin v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: ret
+entry:
+ %c = call <2 x double> @llvm.minimum.v2f64(<2 x double> %a, <2 x double> %b)
+ ret <2 x double> %c
+}
+
+define <2 x double> @max_v2f64(<2 x double> %a, <2 x double> %b) {
+; CHECK-LABEL: max_v2f64:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmax v0.2d, v0.2d, v1.2d
+; CHECK-NEXT: ret
+entry:
+ %c = call <2 x double> @llvm.maximum.v2f64(<2 x double> %a, <2 x double> %b)
+ ret <2 x double> %c
+}
+
+define <3 x double> @min_v3f64(<3 x double> %a, <3 x double> %b) {
+; CHECK-SD-LABEL: min_v3f64:
+; CHECK-SD: // %bb.0: // %entry
+; CHECK-SD-NEXT: // kill: def $d3 killed $d3 def $q3
+; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-SD-NEXT: // kill: def $d4 killed $d4 def $q4
+; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
+; CHECK-SD-NEXT: // kill: def $d5 killed $d5 def $q5
+; CHECK-SD-NEXT: mov v3.d[1], v4.d[0]
+; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
+; CHECK-SD-NEXT: fmin v2.2d, v2.2d, v5.2d
+; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
+; CHECK-SD-NEXT: fmin v0.2d, v0.2d, v3.2d
+; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: min_v3f64:
+; CHECK-GI: // %bb.0: // %entry
+; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-GI-NEXT: // kill: def $d3 killed $d3 def $q3
+; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-GI-NEXT: // kill: def $d4 killed $d4 def $q4
+; CHECK-GI-NEXT: fmin d2, d2, d5
+; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
+; CHECK-GI-NEXT: mov v3.d[1], v4.d[0]
+; CHECK-GI-NEXT: fmin v0.2d, v0.2d, v3.2d
+; CHECK-GI-NEXT: mov d1, v0.d[1]
+; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT: ret
+entry:
+ %c = call <3 x double> @llvm.minimum.v3f64(<3 x double> %a, <3 x double> %b)
+ ret <3 x double> %c
+}
+
+define <3 x double> @max_v3f64(<3 x double> %a, <3 x double> %b) {
+; CHECK-SD-LABEL: max_v3f64:
+; CHECK-SD: // %bb.0: // %entry
+; CHECK-SD-NEXT: // kill: def $d3 killed $d3 def $q3
+; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-SD-NEXT: // kill: def $d4 killed $d4 def $q4
+; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q2
+; CHECK-SD-NEXT: // kill: def $d5 killed $d5 def $q5
+; CHECK-SD-NEXT: mov v3.d[1], v4.d[0]
+; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]
+; CHECK-SD-NEXT: fmax v2.2d, v2.2d, v5.2d
+; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q2
+; CHECK-SD-NEXT: fmax v0.2d, v0.2d, v3.2d
+; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #8
+; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q1
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: max_v3f64:
+; CHECK-GI: // %bb.0: // %entry
+; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-GI-NEXT: // kill: def $d3 killed $d3 def $q3
+; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-GI-NEXT: // kill: def $d4 killed $d4 def $q4
+; CHECK-GI-NEXT: fmax d2, d2, d5
+; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]
+; CHECK-GI-NEXT: mov v3.d[1], v4.d[0]
+; CHECK-GI-NEXT: fmax v0.2d, v0.2d, v3.2d
+; CHECK-GI-NEXT: mov d1, v0.d[1]
+; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT: ret
+entry:
+ %c = call <3 x double> @llvm.maximum.v3f64(<3 x double> %a, <3 x double> %b)
+ ret <3 x double> %c
+}
+
+define <4 x double> @min_v4f64(<4 x double> %a, <4 x double> %b) {
+; CHECK-LABEL: min_v4f64:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmin v0.2d, v0.2d, v2.2d
+; CHECK-NEXT: fmin v1.2d, v1.2d, v3.2d
+; CHECK-NEXT: ret
+entry:
+ %c = call <4 x double> @llvm.minimum.v4f64(<4 x double> %a, <4 x double> %b)
+ ret <4 x double> %c
+}
+
+define <4 x double> @max_v4f64(<4 x double> %a, <4 x double> %b) {
+; CHECK-LABEL: max_v4f64:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmax v0.2d, v0.2d, v2.2d
+; CHECK-NEXT: fmax v1.2d, v1.2d, v3.2d
+; CHECK-NEXT: ret
+entry:
+ %c = call <4 x double> @llvm.maximum.v4f64(<4 x double> %a, <4 x double> %b)
+ ret <4 x double> %c
+}
+
+define <2 x float> @min_v2f32(<2 x float> %a, <2 x float> %b) {
+; CHECK-LABEL: min_v2f32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmin v0.2s, v0.2s, v1.2s
+; CHECK-NEXT: ret
+entry:
+ %c = call <2 x float> @llvm.minimum.v2f32(<2 x float> %a, <2 x float> %b)
+ ret <2 x float> %c
+}
+
+define <2 x float> @max_v2f32(<2 x float> %a, <2 x float> %b) {
+; CHECK-LABEL: max_v2f32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmax v0.2s, v0.2s, v1.2s
+; CHECK-NEXT: ret
+entry:
+ %c = call <2 x float> @llvm.maximum.v2f32(<2 x float> %a, <2 x float> %b)
+ ret <2 x float> %c
+}
+
+define <3 x float> @min_v3f32(<3 x float> %a, <3 x float> %b) {
+; CHECK-SD-LABEL: min_v3f32:
+; CHECK-SD: // %bb.0: // %entry
+; CHECK-SD-NEXT: fmin v0.4s, v0.4s, v1.4s
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: min_v3f32:
+; CHECK-GI: // %bb.0: // %entry
+; CHECK-GI-NEXT: mov s2, v0.s[1]
+; CHECK-GI-NEXT: mov s3, v1.s[1]
+; CHECK-GI-NEXT: mov s4, v0.s[2]
+; CHECK-GI-NEXT: mov s5, v1.s[2]
+; CHECK-GI-NEXT: mov v0.s[1], v2.s[0]
+; CHECK-GI-NEXT: mov v1.s[1], v3.s[0]
+; CHECK-GI-NEXT: mov v0.s[2], v4.s[0]
+; CHECK-GI-NEXT: mov v1.s[2], v5.s[0]
+; CHECK-GI-NEXT: mov v0.s[3], v0.s[0]
+; CHECK-GI-NEXT: mov v1.s[3], v0.s[0]
+; CHECK-GI-NEXT: fmin v0.4s, v0.4s, v1.4s
+; CHECK-GI-NEXT: mov s1, v0.s[1]
+; CHECK-GI-NEXT: mov s2, v0.s[2]
+; CHECK-GI-NEXT: mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT: mov v0.s[2], v2.s[0]
+; CHECK-GI-NEXT: mov v0.s[3], v0.s[0]
+; CHECK-GI-NEXT: ret
+entry:
+ %c = call <3 x float> @llvm.minimum.v3f32(<3 x float> %a, <3 x float> %b)
+ ret <3 x float> %c
+}
+
+define <3 x float> @max_v3f32(<3 x float> %a, <3 x float> %b) {
+; CHECK-SD-LABEL: max_v3f32:
+; CHECK-SD: // %bb.0: // %entry
+; CHECK-SD-NEXT: fmax v0.4s, v0.4s, v1.4s
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: max_v3f32:
+; CHECK-GI: // %bb.0: // %entry
+; CHECK-GI-NEXT: mov s2, v0.s[1]
+; CHECK-GI-NEXT: mov s3, v1.s[1]
+; CHECK-GI-NEXT: mov s4, v0.s[2]
+; CHECK-GI-NEXT: mov s5, v1.s[2]
+; CHECK-GI-NEXT: mov v0.s[1], v2.s[0]
+; CHECK-GI-NEXT: mov v1.s[1], v3.s[0]
+; CHECK-GI-NEXT: mov v0.s[2], v4.s[0]
+; CHECK-GI-NEXT: mov v1.s[2], v5.s[0]
+; CHECK-GI-NEXT: mov v0.s[3], v0.s[0]
+; CHECK-GI-NEXT: mov v1.s[3], v0.s[0]
+; CHECK-GI-NEXT: fmax v0.4s, v0.4s, v1.4s
+; CHECK-GI-NEXT: mov s1, v0.s[1]
+; CHECK-GI-NEXT: mov s2, v0.s[2]
+; CHECK-GI-NEXT: mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT: mov v0.s[2], v2.s[0]
+; CHECK-GI-NEXT: mov v0.s[3], v0.s[0]
+; CHECK-GI-NEXT: ret
+entry:
+ %c = call <3 x float> @llvm.maximum.v3f32(<3 x float> %a, <3 x float> %b)
+ ret <3 x float> %c
+}
+
+define <4 x float> @min_v4f32(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: min_v4f32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmin v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: ret
+entry:
+ %c = call <4 x float> @llvm.minimum.v4f32(<4 x float> %a, <4 x float> %b)
+ ret <4 x float> %c
+}
+
+define <4 x float> @max_v4f32(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: max_v4f32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmax v0.4s, v0.4s, v1.4s
+; CHECK-NEXT: ret
+entry:
+ %c = call <4 x float> @llvm.maximum.v4f32(<4 x float> %a, <4 x float> %b)
+ ret <4 x float> %c
+}
+
+define <7 x float> @min_v7f32(<7 x float> %a, <7 x float> %b) {
+; CHECK-SD-LABEL: min_v7f32:
+; CHECK-SD: // %bb.0: // %entry
+; CHECK-SD-NEXT: mov x8, sp
+; CHECK-SD-NEXT: // kill: def $s7 killed $s7 def $q7
+; CHECK-SD-NEXT: // kill: def $s0 killed $s0 def $q0
+; CHECK-SD-NEXT: // kill: def $s1 killed $s1 def $q1
+; CHECK-SD-NEXT: add x9, sp, #32
+; CHECK-SD-NEXT: mov v0.s[1], v1.s[0]
+; CHECK-SD-NEXT: ldr s1, [sp, #24]
+; CHECK-SD-NEXT: // kill: def $s4 killed $s4 def $q4
+; CHECK-SD-NEXT: // kill: def $s5 killed $s5 def $q5
+; CHECK-SD-NEXT: // kill: def $s2 killed $s2 def $q2
+; CHECK-SD-NEXT: // kill: def $s6 killed $s6 def $q6
+; CHECK-SD-NEXT: // kill: def $s3 killed $s3 def $q3
+; CHECK-SD-NEXT: ld1 { v7.s }[1], [x8]
+; CHECK-SD-NEXT: add x8, sp, #8
+; CHECK-SD-NEXT: ld1 { v1.s }[1], [x9]
+; CHECK-SD-NEXT: mov v0.s[2], v2.s[0]
+; CHECK-SD-NEXT: mov v4.s[1], v5.s[0]
+; CHECK-SD-NEXT: ld1 { v7.s }[2], [x8]
+; CHECK-SD-NEXT: add x8, sp, #16
+; CHECK-SD-NEXT: mov v0.s[3], v3.s[0]
+; CHECK-SD-NEXT: mov v4.s[2], v6.s[0]
+; CHECK-SD-NEXT: ld1 { v7.s }[3], [x8]
+; CHECK-SD-NEXT: add x8, sp, #40
+; CHECK-SD-NEXT: ld1 { v1.s }[2], [x8]
+; CHECK-SD-NEXT: fmin v0.4s, v0.4s, v7.4s
+; CHECK-SD-NEXT: fmin v4.4s, v4.4s, v1.4s
+; CHECK-SD-NEXT: mov s1, v0.s[1]
+; CHECK-SD-NEXT: mov s2, v0.s[2]
+; CHECK-SD-NEXT: mov s3, v0.s[3]
+; CHECK-SD-NEXT: // kill: def $s0 killed $s0 killed $q0
+; CHECK-SD-NEXT: mov s5, v4.s[1]
+; CHECK-SD-NEXT: mov s6, v4.s[2]
+; CHECK-SD-NEXT: // kill: def $s4 killed $s4 killed $q4
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: min_v7f32:
+; CHECK-GI: // %bb.0: // %entry
+; CHECK-GI-NEXT: // kill: def $s0 killed $s0 def $q0
+; CHECK-GI-NEXT: // kill: def $s1 killed $s1 def $q1
+; CHECK-GI-NEXT: ldr s16, [sp]
+; CHECK-GI-NEXT: // kill: def $s4 killed $s4 def $q4
+; CHECK-GI-NEXT: // kill: def $s7 killed $s7 def $q7
+; CHECK-GI-NEXT: // kill: def $s5 killed $s5 def $q5
+; CHECK-GI-NEXT: // kill: def $s2 killed $s2 def $q2
+; CHECK-GI-NEXT: // kill: def $s6 killed $s6 def $q6
+; CHECK-GI-NEXT: // kill: def $s3 killed $s3 def $q3
+; CHECK-GI-NEXT: mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT: ldr s19, [sp, #24]
+; CHECK-GI-NEXT: ldr s20, [sp, #32]
+; CHECK-GI-NEXT: mov v7.s[1], v16.s[0]
+; CHECK-GI-NEXT: ldr s17, [sp, #8]
+; CHECK-GI-NEXT: mov v4.s[1], v5.s[0]
+; CHECK-GI-NEXT: ldr s21, [sp, #40]
+; CHECK-GI-NEXT: mov v19.s[1], v20.s[0]
+; CHECK-GI-NEXT: ldr s18, [sp, #16]
+; CHECK-GI-NEXT: mov v0.s[2], v2.s[0]
+; CHECK-GI-NEXT: mov v7.s[2], v17.s[0]
+; CHECK-GI-NEXT: mov v4.s[2], v6.s[0]
+; CHECK-GI-NEXT: mov v19.s[2], v21.s[0]
+; CHECK-GI-NEXT: mov v0.s[3], v3.s[0]
+; CHECK-GI-NEXT: mov v7.s[3], v18.s[0]
+; CHECK-GI-NEXT: mov v4.s[3], v0.s[0]
+; CHECK-GI-NEXT: mov v19.s[3], v0.s[0]
+; CHECK-GI-NEXT: fmin v0.4s, v0.4s, v7.4s
+; CHECK-GI-NEXT: fmin v4.4s, v4.4s, v19.4s
+; CHECK-GI-NEXT: mov s1, v0.s[1]
+; CHECK-GI-NEXT: mov s2, v0.s[2]
+; CHECK-GI-NEXT: mov s3, v0.s[3]
+; CHECK-GI-NEXT: // kill: def $s0 killed $s0 killed $q0
+; CHECK-GI-NEXT: mov s5, v4.s[1]
+; CHECK-GI-NEXT: mov s6, v4.s[2]
+; CHECK-GI-NEXT: // kill: def $s4 killed $s4 killed $q4
+; CHECK-GI-NEXT: ret
+entry:
+ %c = call <7 x float> @llvm.minimum.v7f32(<7 x float> %a, <7 x float> %b)
+ ret <7 x float> %c
+}
+
+define <7 x float> @max_v7f32(<7 x float> %a, <7 x float> %b) {
+; CHECK-SD-LABEL: max_v7f32:
+; CHECK-SD: // %bb.0: // %entry
+; CHECK-SD-NEXT: mov x8, sp
+; CHECK-SD-NEXT: // kill: def $s7 killed $s7 def $q7
+; CHECK-SD-NEXT: // kill: def $s0 killed $s0 def $q0
+; CHECK-SD-NEXT: // kill: def $s1 killed $s1 def $q1
+; CHECK-SD-NEXT: add x9, sp, #32
+; CHECK-SD-NEXT: mov v0.s[1], v1.s[0]
+; CHECK-SD-NEXT: ldr s1, [sp, #24]
+; CHECK-SD-NEXT: // kill: def $s4 killed $s4 def $q4
+; CHECK-SD-NEXT: // kill: def $s5 killed $s5 def $q5
+; CHECK-SD-NEXT: // kill: def $s2 killed $s2 def $q2
+; CHECK-SD-NEXT: // kill: def $s6 killed $s6 def $q6
+; CHECK-SD-NEXT: // kill: def $s3 killed $s3 def $q3
+; CHECK-SD-NEXT: ld1 { v7.s }[1], [x8]
+; CHECK-SD-NEXT: add x8, sp, #8
+; CHECK-SD-NEXT: ld1 { v1.s }[1], [x9]
+; CHECK-SD-NEXT: mov v0.s[2], v2.s[0]
+; CHECK-SD-NEXT: mov v4.s[1], v5.s[0]
+; CHECK-SD-NEXT: ld1 { v7.s }[2], [x8]
+; CHECK-SD-NEXT: add x8, sp, #16
+; CHECK-SD-NEXT: mov v0.s[3], v3.s[0]
+; CHECK-SD-NEXT: mov v4.s[2], v6.s[0]
+; CHECK-SD-NEXT: ld1 { v7.s }[3], [x8]
+; CHECK-SD-NEXT: add x8, sp, #40
+; CHECK-SD-NEXT: ld1 { v1.s }[2], [x8]
+; CHECK-SD-NEXT: fmax v0.4s, v0.4s, v7.4s
+; CHECK-SD-NEXT: fmax v4.4s, v4.4s, v1.4s
+; CHECK-SD-NEXT: mov s1, v0.s[1]
+; CHECK-SD-NEXT: mov s2, v0.s[2]
+; CHECK-SD-NEXT: mov s3, v0.s[3]
+; CHECK-SD-NEXT: // kill: def $s0 killed $s0 killed $q0
+; CHECK-SD-NEXT: mov s5, v4.s[1]
+; CHECK-SD-NEXT: mov s6, v4.s[2]
+; CHECK-SD-NEXT: // kill: def $s4 killed $s4 killed $q4
+; CHECK-SD-NEXT: ret
+;
+; CHECK-GI-LABEL: max_v7f32:
+; CHECK-GI: // %bb.0: // %entry
+; CHECK-GI-NEXT: // kill: def $s0 killed $s0 def $q0
+; CHECK-GI-NEXT: // kill: def $s1 killed $s1 def $q1
+; CHECK-GI-NEXT: ldr s16, [sp]
+; CHECK-GI-NEXT: // kill: def $s4 killed $s4 def $q4
+; CHECK-GI-NEXT: // kill: def $s7 killed $s7 def $q7
+; CHECK-GI-NEXT: // kill: def $s5 killed $s5 def $q5
+; CHECK-GI-NEXT: // kill: def $s2 killed $s2 def $q2
+; CHECK-GI-NEXT: // kill: def $s6 killed $s6 def $q6
+; CHECK-GI-NEXT: // kill: def $s3 killed $s3 def $q3
+; CHECK-GI-NEXT: mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT: ldr s19, [sp, #24]
+; CHECK-GI-NEXT: ldr s20, [sp, #32]
+; CHECK-GI-NEXT: mov v7.s[1], v16.s[0]
+; CHECK-GI-NEXT: ldr s17, [sp, #8]
+; CHECK-GI-NEXT: mov v4.s[1], v5.s[0]
+; CHECK-GI-NEXT: ldr s21, [sp, #40]
+; CHECK-GI-NEXT: mov v19.s[1], v20.s[0]
+; CHECK-GI-NEXT: ldr s18, [sp, #16]
+; CHECK-GI-NEXT: mov v0.s[2], v2.s[0]
+; CHECK-GI-NEXT: mov v7.s[2], v17.s[0]
+; CHECK-GI-NEXT: mov v4.s[2], v6.s[0]
+; CHECK-GI-NEXT: mov v19.s[2], v21.s[0]
+; CHECK-GI-NEXT: mov v0.s[3], v3.s[0]
+; CHECK-GI-NEXT: mov v7.s[3], v18.s[0]
+; CHECK-GI-NEXT: mov v4.s[3], v0.s[0]
+; CHECK-GI-NEXT: mov v19.s[3], v0.s[0]
+; CHECK-GI-NEXT: fmax v0.4s, v0.4s, v7.4s
+; CHECK-GI-NEXT: fmax v4.4s, v4.4s, v19.4s
+; CHECK-GI-NEXT: mov s1, v0.s[1]
+; CHECK-GI-NEXT: mov s2, v0.s[2]
+; CHECK-GI-NEXT: mov s3, v0.s[3]
+; CHECK-GI-NEXT: // kill: def $s0 killed $s0 killed $q0
+; CHECK-GI-NEXT: mov s5, v4.s[1]
+; CHECK-GI-NEXT: mov s6, v4.s[2]
+; CHECK-GI-NEXT: // kill: def $s4 killed $s4 killed $q4
+; CHECK-GI-NEXT: ret
+entry:
+ %c = call <7 x float> @llvm.maximum.v7f32(<7 x float> %a, <7 x float> %b)
+ ret <7 x float> %c
+}
+
+define <8 x float> @min_v8f32(<8 x float> %a, <8 x float> %b) {
+; CHECK-LABEL: min_v8f32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmin v0.4s, v0.4s, v2.4s
+; CHECK-NEXT: fmin v1.4s, v1.4s, v3.4s
+; CHECK-NEXT: ret
+entry:
+ %c = call <8 x float> @llvm.minimum.v8f32(<8 x float> %a, <8 x float> %b)
+ ret <8 x float> %c
+}
+
+define <8 x float> @max_v8f32(<8 x float> %a, <8 x float> %b) {
+; CHECK-LABEL: max_v8f32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: fmax v0.4s, v0.4s, v2.4s
+; CHECK-NEXT: fmax v1.4s, v1.4s, v3.4s
+; CHECK-NEXT: ret
+entry:
+ %c = call <8 x float> @llvm.maximum.v8f32(<8 x float> %a, <8 x float> %b)
+ ret <8 x float> %c
+}
+
+define <4 x half> @min_v4f16(<4 x half> %a, <4 x half> %b) {
+; CHECK-NOFP16-SD-LABEL: min_v4f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-NOFP16-SD-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-NOFP16-SD-NEXT: mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h1, v1.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT: fmin s4, s5, s4
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h7
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fmin s2, s3, s2
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h6
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s4
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s2
+; CHECK-NOFP16-SD-NEXT: fmin s3, s5, s3
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h6
+; CHECK-NOFP16-SD-NEXT: mov v0.h[1], v2.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s3
+; CHECK-NOFP16-SD-NEXT: fmin s1, s4, s1
+; CHECK-NOFP16-SD-NEXT: mov v0.h[2], v2.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h1, s1
+; CHECK-NOFP16-SD-NEXT: mov v0.h[3], v1.h[0]
+; CHECK-NOFP16-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: min_v4f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmin v0.4h, v0.4h, v1.4h
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: min_v4f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: fcvtl v0.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v1.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT: fmin v0.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn v0.4h, v0.4s
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: min_v4f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: fmin v0.4h, v0.4h, v1.4h
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call <4 x half> @llvm.minimum.v4f16(<4 x half> %a, <4 x half> %b)
+ ret <4 x half> %c
+}
+
+define <4 x half> @max_v4f16(<4 x half> %a, <4 x half> %b) {
+; CHECK-NOFP16-SD-LABEL: max_v4f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: // kill: def $d1 killed $d1 def $q1
+; CHECK-NOFP16-SD-NEXT: // kill: def $d0 killed $d0 def $q0
+; CHECK-NOFP16-SD-NEXT: mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h1, v1.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT: fmax s4, s5, s4
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h7
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fmax s2, s3, s2
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h6
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s4
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s2
+; CHECK-NOFP16-SD-NEXT: fmax s3, s5, s3
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h6
+; CHECK-NOFP16-SD-NEXT: mov v0.h[1], v2.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s3
+; CHECK-NOFP16-SD-NEXT: fmax s1, s4, s1
+; CHECK-NOFP16-SD-NEXT: mov v0.h[2], v2.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h1, s1
+; CHECK-NOFP16-SD-NEXT: mov v0.h[3], v1.h[0]
+; CHECK-NOFP16-SD-NEXT: // kill: def $d0 killed $d0 killed $q0
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: max_v4f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmax v0.4h, v0.4h, v1.4h
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: max_v4f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: fcvtl v0.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v1.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT: fmax v0.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn v0.4h, v0.4s
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: max_v4f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: fmax v0.4h, v0.4h, v1.4h
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call <4 x half> @llvm.maximum.v4f16(<4 x half> %a, <4 x half> %b)
+ ret <4 x half> %c
+}
+
+define <7 x half> @min_v7f16(<7 x half> %a, <7 x half> %b) {
+; CHECK-NOFP16-SD-LABEL: min_v7f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h16, v1.h[3]
+; CHECK-NOFP16-SD-NEXT: mov h17, v0.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT: fmin s4, s5, s4
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h7
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: fmin s3, s3, s2
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s4
+; CHECK-NOFP16-SD-NEXT: fmin s4, s6, s5
+; CHECK-NOFP16-SD-NEXT: mov h5, v1.h[4]
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[4]
+; CHECK-NOFP16-SD-NEXT: fmin s7, s16, s7
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT: mov h16, v0.h[5]
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT: fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT: mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT: mov h4, v1.h[5]
+; CHECK-NOFP16-SD-NEXT: fmin s5, s6, s5
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[6]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[2], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h4
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h16
+; CHECK-NOFP16-SD-NEXT: mov h16, v0.h[6]
+; CHECK-NOFP16-SD-NEXT: mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[3], v7.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fmin s3, s4, s3
+; CHECK-NOFP16-SD-NEXT: fcvt h4, s5
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT: mov v2.h[4], v4.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT: fmin s4, s6, s5
+; CHECK-NOFP16-SD-NEXT: fmin s0, s0, s1
+; CHECK-NOFP16-SD-NEXT: mov v2.h[5], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT: mov v2.h[6], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v0.16b, v2.16b
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: min_v7f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmin v0.8h, v0.8h, v1.8h
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: min_v7f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: mov h2, v0.h[4]
+; CHECK-NOFP16-GI-NEXT: mov h3, v0.h[5]
+; CHECK-NOFP16-GI-NEXT: mov h4, v1.h[4]
+; CHECK-NOFP16-GI-NEXT: mov h5, v1.h[5]
+; CHECK-NOFP16-GI-NEXT: mov h6, v0.h[6]
+; CHECK-NOFP16-GI-NEXT: mov h7, v1.h[6]
+; CHECK-NOFP16-GI-NEXT: fcvtl v1.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT: mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v4.h[1], v5.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.h[2], v6.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v4.h[2], v7.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.h[3], v0.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v4.h[3], v0.h[0]
+; CHECK-NOFP16-GI-NEXT: fcvtl v2.4s, v2.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v3.4s, v4.4h
+; CHECK-NOFP16-GI-NEXT: mov s4, v2.s[1]
+; CHECK-NOFP16-GI-NEXT: mov s6, v2.s[2]
+; CHECK-NOFP16-GI-NEXT: mov s5, v3.s[1]
+; CHECK-NOFP16-GI-NEXT: mov s7, v3.s[2]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[1], v4.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v3.s[1], v5.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[2], v6.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v3.s[2], v7.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v3.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT: fcvtl v0.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT: fmin v2.4s, v2.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT: fmin v0.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT: mov s1, v2.s[1]
+; CHECK-NOFP16-GI-NEXT: mov s3, v2.s[2]
+; CHECK-NOFP16-GI-NEXT: fcvtn v0.4h, v0.4s
+; CHECK-NOFP16-GI-NEXT: mov v2.s[1], v1.s[0]
+; CHECK-NOFP16-GI-NEXT: mov h1, v0.h[1]
+; CHECK-NOFP16-GI-NEXT: mov h4, v0.h[2]
+; CHECK-NOFP16-GI-NEXT: mov h5, v0.h[3]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[2], v3.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[1], v1.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[2], v4.h[0]
+; CHECK-NOFP16-GI-NEXT: fcvtn v1.4h, v2.4s
+; CHECK-NOFP16-GI-NEXT: mov v0.h[3], v5.h[0]
+; CHECK-NOFP16-GI-NEXT: mov h2, v1.h[1]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[4], v1.h[0]
+; CHECK-NOFP16-GI-NEXT: mov h1, v1.h[2]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[5], v2.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[6], v1.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[7], v0.h[0]
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: min_v7f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: mov h2, v0.h[1]
+; CHECK-FP16-GI-NEXT: mov h3, v0.h[2]
+; CHECK-FP16-GI-NEXT: mov h4, v0.h[3]
+; CHECK-FP16-GI-NEXT: mov h5, v0.h[4]
+; CHECK-FP16-GI-NEXT: mov h6, v0.h[5]
+; CHECK-FP16-GI-NEXT: mov h7, v0.h[6]
+; CHECK-FP16-GI-NEXT: mov h16, v1.h[1]
+; CHECK-FP16-GI-NEXT: mov h17, v1.h[2]
+; CHECK-FP16-GI-NEXT: mov v0.h[1], v2.h[0]
+; CHECK-FP16-GI-NEXT: mov h18, v1.h[3]
+; CHECK-FP16-GI-NEXT: mov h19, v1.h[4]
+; CHECK-FP16-GI-NEXT: mov h20, v1.h[5]
+; CHECK-FP16-GI-NEXT: mov h21, v1.h[6]
+; CHECK-FP16-GI-NEXT: mov v1.h[1], v16.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[2], v3.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[2], v17.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[3], v4.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[3], v18.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[4], v5.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[4], v19.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[5], v6.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[5], v20.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[6], v7.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[6], v21.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT: fmin v0.8h, v0.8h, v1.8h
+; CHECK-FP16-GI-NEXT: mov h1, v0.h[1]
+; CHECK-FP16-GI-NEXT: mov h2, v0.h[2]
+; CHECK-FP16-GI-NEXT: mov h3, v0.h[3]
+; CHECK-FP16-GI-NEXT: mov h4, v0.h[4]
+; CHECK-FP16-GI-NEXT: mov h5, v0.h[5]
+; CHECK-FP16-GI-NEXT: mov h6, v0.h[6]
+; CHECK-FP16-GI-NEXT: mov v0.h[1], v1.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[2], v2.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[3], v3.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[4], v4.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[5], v5.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[6], v6.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call <7 x half> @llvm.minimum.v7f16(<7 x half> %a, <7 x half> %b)
+ ret <7 x half> %c
+}
+
+define <7 x half> @max_v7f16(<7 x half> %a, <7 x half> %b) {
+; CHECK-NOFP16-SD-LABEL: max_v7f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h16, v1.h[3]
+; CHECK-NOFP16-SD-NEXT: mov h17, v0.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT: fmax s4, s5, s4
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h7
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: fmax s3, s3, s2
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s4
+; CHECK-NOFP16-SD-NEXT: fmax s4, s6, s5
+; CHECK-NOFP16-SD-NEXT: mov h5, v1.h[4]
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[4]
+; CHECK-NOFP16-SD-NEXT: fmax s7, s16, s7
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT: mov h16, v0.h[5]
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT: fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT: mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT: mov h4, v1.h[5]
+; CHECK-NOFP16-SD-NEXT: fmax s5, s6, s5
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[6]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[2], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h4
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h16
+; CHECK-NOFP16-SD-NEXT: mov h16, v0.h[6]
+; CHECK-NOFP16-SD-NEXT: mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[3], v7.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fmax s3, s4, s3
+; CHECK-NOFP16-SD-NEXT: fcvt h4, s5
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT: mov v2.h[4], v4.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT: fmax s4, s6, s5
+; CHECK-NOFP16-SD-NEXT: fmax s0, s0, s1
+; CHECK-NOFP16-SD-NEXT: mov v2.h[5], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT: mov v2.h[6], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v0.16b, v2.16b
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: max_v7f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmax v0.8h, v0.8h, v1.8h
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: max_v7f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: mov h2, v0.h[4]
+; CHECK-NOFP16-GI-NEXT: mov h3, v0.h[5]
+; CHECK-NOFP16-GI-NEXT: mov h4, v1.h[4]
+; CHECK-NOFP16-GI-NEXT: mov h5, v1.h[5]
+; CHECK-NOFP16-GI-NEXT: mov h6, v0.h[6]
+; CHECK-NOFP16-GI-NEXT: mov h7, v1.h[6]
+; CHECK-NOFP16-GI-NEXT: fcvtl v1.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT: mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v4.h[1], v5.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.h[2], v6.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v4.h[2], v7.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.h[3], v0.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v4.h[3], v0.h[0]
+; CHECK-NOFP16-GI-NEXT: fcvtl v2.4s, v2.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v3.4s, v4.4h
+; CHECK-NOFP16-GI-NEXT: mov s4, v2.s[1]
+; CHECK-NOFP16-GI-NEXT: mov s6, v2.s[2]
+; CHECK-NOFP16-GI-NEXT: mov s5, v3.s[1]
+; CHECK-NOFP16-GI-NEXT: mov s7, v3.s[2]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[1], v4.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v3.s[1], v5.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[2], v6.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v3.s[2], v7.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v3.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT: fcvtl v0.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT: fmax v2.4s, v2.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT: fmax v0.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT: mov s1, v2.s[1]
+; CHECK-NOFP16-GI-NEXT: mov s3, v2.s[2]
+; CHECK-NOFP16-GI-NEXT: fcvtn v0.4h, v0.4s
+; CHECK-NOFP16-GI-NEXT: mov v2.s[1], v1.s[0]
+; CHECK-NOFP16-GI-NEXT: mov h1, v0.h[1]
+; CHECK-NOFP16-GI-NEXT: mov h4, v0.h[2]
+; CHECK-NOFP16-GI-NEXT: mov h5, v0.h[3]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[2], v3.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[1], v1.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v2.s[3], v0.s[0]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[2], v4.h[0]
+; CHECK-NOFP16-GI-NEXT: fcvtn v1.4h, v2.4s
+; CHECK-NOFP16-GI-NEXT: mov v0.h[3], v5.h[0]
+; CHECK-NOFP16-GI-NEXT: mov h2, v1.h[1]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[4], v1.h[0]
+; CHECK-NOFP16-GI-NEXT: mov h1, v1.h[2]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[5], v2.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[6], v1.h[0]
+; CHECK-NOFP16-GI-NEXT: mov v0.h[7], v0.h[0]
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: max_v7f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: mov h2, v0.h[1]
+; CHECK-FP16-GI-NEXT: mov h3, v0.h[2]
+; CHECK-FP16-GI-NEXT: mov h4, v0.h[3]
+; CHECK-FP16-GI-NEXT: mov h5, v0.h[4]
+; CHECK-FP16-GI-NEXT: mov h6, v0.h[5]
+; CHECK-FP16-GI-NEXT: mov h7, v0.h[6]
+; CHECK-FP16-GI-NEXT: mov h16, v1.h[1]
+; CHECK-FP16-GI-NEXT: mov h17, v1.h[2]
+; CHECK-FP16-GI-NEXT: mov v0.h[1], v2.h[0]
+; CHECK-FP16-GI-NEXT: mov h18, v1.h[3]
+; CHECK-FP16-GI-NEXT: mov h19, v1.h[4]
+; CHECK-FP16-GI-NEXT: mov h20, v1.h[5]
+; CHECK-FP16-GI-NEXT: mov h21, v1.h[6]
+; CHECK-FP16-GI-NEXT: mov v1.h[1], v16.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[2], v3.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[2], v17.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[3], v4.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[3], v18.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[4], v5.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[4], v19.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[5], v6.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[5], v20.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[6], v7.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[6], v21.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT: mov v1.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT: fmax v0.8h, v0.8h, v1.8h
+; CHECK-FP16-GI-NEXT: mov h1, v0.h[1]
+; CHECK-FP16-GI-NEXT: mov h2, v0.h[2]
+; CHECK-FP16-GI-NEXT: mov h3, v0.h[3]
+; CHECK-FP16-GI-NEXT: mov h4, v0.h[4]
+; CHECK-FP16-GI-NEXT: mov h5, v0.h[5]
+; CHECK-FP16-GI-NEXT: mov h6, v0.h[6]
+; CHECK-FP16-GI-NEXT: mov v0.h[1], v1.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[2], v2.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[3], v3.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[4], v4.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[5], v5.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[6], v6.h[0]
+; CHECK-FP16-GI-NEXT: mov v0.h[7], v0.h[0]
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call <7 x half> @llvm.maximum.v7f16(<7 x half> %a, <7 x half> %b)
+ ret <7 x half> %c
+}
+
+define <8 x half> @min_v8f16(<8 x half> %a, <8 x half> %b) {
+; CHECK-NOFP16-SD-LABEL: min_v8f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h16, v1.h[3]
+; CHECK-NOFP16-SD-NEXT: mov h17, v0.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT: fmin s4, s5, s4
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h7
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: fmin s3, s3, s2
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s4
+; CHECK-NOFP16-SD-NEXT: fmin s4, s6, s5
+; CHECK-NOFP16-SD-NEXT: mov h5, v1.h[4]
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[4]
+; CHECK-NOFP16-SD-NEXT: fmin s7, s16, s7
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT: mov h16, v0.h[5]
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT: fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT: mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT: mov h4, v1.h[5]
+; CHECK-NOFP16-SD-NEXT: fmin s5, s6, s5
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[6]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[2], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h4
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h16
+; CHECK-NOFP16-SD-NEXT: mov h16, v0.h[6]
+; CHECK-NOFP16-SD-NEXT: mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[3], v7.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fmin s3, s4, s3
+; CHECK-NOFP16-SD-NEXT: fcvt h4, s5
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT: mov v2.h[4], v4.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT: fmin s4, s6, s5
+; CHECK-NOFP16-SD-NEXT: fmin s0, s0, s1
+; CHECK-NOFP16-SD-NEXT: mov v2.h[5], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT: mov v2.h[6], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v0.16b, v2.16b
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: min_v8f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmin v0.8h, v0.8h, v1.8h
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: min_v8f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: fcvtl v2.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v3.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v0.4s, v0.8h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v1.4s, v1.8h
+; CHECK-NOFP16-GI-NEXT: fmin v2.4s, v2.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT: fmin v1.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn v0.4h, v2.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn2 v0.8h, v1.4s
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: min_v8f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: fmin v0.8h, v0.8h, v1.8h
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call <8 x half> @llvm.minimum.v8f16(<8 x half> %a, <8 x half> %b)
+ ret <8 x half> %c
+}
+
+define <8 x half> @max_v8f16(<8 x half> %a, <8 x half> %b) {
+; CHECK-NOFP16-SD-LABEL: max_v8f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: mov h2, v1.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h3, v0.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h0
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h7, v0.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h16, v1.h[3]
+; CHECK-NOFP16-SD-NEXT: mov h17, v0.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT: fmax s4, s5, s4
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h7
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: fmax s3, s3, s2
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s4
+; CHECK-NOFP16-SD-NEXT: fmax s4, s6, s5
+; CHECK-NOFP16-SD-NEXT: mov h5, v1.h[4]
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[4]
+; CHECK-NOFP16-SD-NEXT: fmax s7, s16, s7
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT: mov h16, v0.h[5]
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT: fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT: mov v2.h[1], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT: mov h4, v1.h[5]
+; CHECK-NOFP16-SD-NEXT: fmax s5, s6, s5
+; CHECK-NOFP16-SD-NEXT: mov h6, v1.h[6]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[2], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h4
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h16
+; CHECK-NOFP16-SD-NEXT: mov h16, v0.h[6]
+; CHECK-NOFP16-SD-NEXT: mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[3], v7.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fmax s3, s4, s3
+; CHECK-NOFP16-SD-NEXT: fcvt h4, s5
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h6
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT: mov v2.h[4], v4.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s3
+; CHECK-NOFP16-SD-NEXT: fmax s4, s6, s5
+; CHECK-NOFP16-SD-NEXT: fmax s0, s0, s1
+; CHECK-NOFP16-SD-NEXT: mov v2.h[5], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h3, s4
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT: mov v2.h[6], v3.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v2.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v0.16b, v2.16b
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: max_v8f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmax v0.8h, v0.8h, v1.8h
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: max_v8f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: fcvtl v2.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v3.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v0.4s, v0.8h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v1.4s, v1.8h
+; CHECK-NOFP16-GI-NEXT: fmax v2.4s, v2.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT: fmax v1.4s, v0.4s, v1.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn v0.4h, v2.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn2 v0.8h, v1.4s
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: max_v8f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: fmax v0.8h, v0.8h, v1.8h
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call <8 x half> @llvm.maximum.v8f16(<8 x half> %a, <8 x half> %b)
+ ret <8 x half> %c
+}
+
+define <16 x half> @min_v16f16(<16 x half> %a, <16 x half> %b) {
+; CHECK-NOFP16-SD-LABEL: min_v16f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: mov h4, v2.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h5, v0.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h2
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h0
+; CHECK-NOFP16-SD-NEXT: mov h16, v2.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h17, v0.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h18, v3.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h19, v1.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h4
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT: fmin s6, s7, s6
+; CHECK-NOFP16-SD-NEXT: fcvt s20, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: mov h17, v2.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT: fmin s5, s5, s4
+; CHECK-NOFP16-SD-NEXT: fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT: fcvt h4, s6
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[3]
+; CHECK-NOFP16-SD-NEXT: fmin s7, s16, s7
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h3
+; CHECK-NOFP16-SD-NEXT: mov h21, v0.h[4]
+; CHECK-NOFP16-SD-NEXT: fcvt s17, h17
+; CHECK-NOFP16-SD-NEXT: fcvt h5, s5
+; CHECK-NOFP16-SD-NEXT: fmin s18, s19, s18
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT: mov h19, v3.h[2]
+; CHECK-NOFP16-SD-NEXT: fmin s16, s20, s16
+; CHECK-NOFP16-SD-NEXT: mov h20, v1.h[2]
+; CHECK-NOFP16-SD-NEXT: fcvt s21, h21
+; CHECK-NOFP16-SD-NEXT: mov v4.h[1], v5.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h5, s7
+; CHECK-NOFP16-SD-NEXT: mov h7, v2.h[4]
+; CHECK-NOFP16-SD-NEXT: fmin s6, s6, s17
+; CHECK-NOFP16-SD-NEXT: mov h17, v3.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT: fcvt s20, h20
+; CHECK-NOFP16-SD-NEXT: mov v4.h[2], v5.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h5, s16
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h7
+; CHECK-NOFP16-SD-NEXT: fcvt h16, s18
+; CHECK-NOFP16-SD-NEXT: mov h18, v1.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s6
+; CHECK-NOFP16-SD-NEXT: fmin s7, s21, s7
+; CHECK-NOFP16-SD-NEXT: mov v5.h[1], v16.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: fmin s17, s20, s19
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT: mov h19, v3.h[4]
+; CHECK-NOFP16-SD-NEXT: mov h20, v1.h[4]
+; CHECK-NOFP16-SD-NEXT: mov v4.h[3], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT: fmin s7, s18, s16
+; CHECK-NOFP16-SD-NEXT: fcvt h16, s17
+; CHECK-NOFP16-SD-NEXT: fcvt s17, h19
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h20
+; CHECK-NOFP16-SD-NEXT: mov v4.h[4], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: mov h19, v1.h[5]
+; CHECK-NOFP16-SD-NEXT: mov v5.h[2], v16.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT: fmin s7, s18, s17
+; CHECK-NOFP16-SD-NEXT: mov h16, v2.h[5]
+; CHECK-NOFP16-SD-NEXT: mov h17, v0.h[5]
+; CHECK-NOFP16-SD-NEXT: mov h18, v3.h[5]
+; CHECK-NOFP16-SD-NEXT: mov v5.h[3], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: fcvt s17, h18
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h19
+; CHECK-NOFP16-SD-NEXT: mov h19, v2.h[6]
+; CHECK-NOFP16-SD-NEXT: mov h2, v2.h[7]
+; CHECK-NOFP16-SD-NEXT: mov v5.h[4], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[6]
+; CHECK-NOFP16-SD-NEXT: fmin s7, s16, s7
+; CHECK-NOFP16-SD-NEXT: fmin s16, s18, s17
+; CHECK-NOFP16-SD-NEXT: mov h17, v3.h[6]
+; CHECK-NOFP16-SD-NEXT: mov h18, v1.h[6]
+; CHECK-NOFP16-SD-NEXT: fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT: mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT: mov h3, v3.h[7]
+; CHECK-NOFP16-SD-NEXT: mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT: fcvt s17, h17
+; CHECK-NOFP16-SD-NEXT: fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT: fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT: fmin s6, s6, s19
+; CHECK-NOFP16-SD-NEXT: fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fcvt h16, s16
+; CHECK-NOFP16-SD-NEXT: fmin s17, s18, s17
+; CHECK-NOFP16-SD-NEXT: mov v4.h[5], v7.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s6
+; CHECK-NOFP16-SD-NEXT: fmin s0, s0, s2
+; CHECK-NOFP16-SD-NEXT: fmin s1, s1, s3
+; CHECK-NOFP16-SD-NEXT: mov v5.h[5], v16.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s17
+; CHECK-NOFP16-SD-NEXT: mov v4.h[6], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT: fcvt h1, s1
+; CHECK-NOFP16-SD-NEXT: mov v5.h[6], v2.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v4.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v5.h[7], v1.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v0.16b, v4.16b
+; CHECK-NOFP16-SD-NEXT: mov v1.16b, v5.16b
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: min_v16f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmin v0.8h, v0.8h, v2.8h
+; CHECK-FP16-SD-NEXT: fmin v1.8h, v1.8h, v3.8h
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: min_v16f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: fcvtl v4.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v5.4s, v2.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v6.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v7.4s, v3.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v0.4s, v0.8h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v2.4s, v2.8h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v1.4s, v1.8h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v3.4s, v3.8h
+; CHECK-NOFP16-GI-NEXT: fmin v4.4s, v4.4s, v5.4s
+; CHECK-NOFP16-GI-NEXT: fmin v5.4s, v6.4s, v7.4s
+; CHECK-NOFP16-GI-NEXT: fmin v2.4s, v0.4s, v2.4s
+; CHECK-NOFP16-GI-NEXT: fmin v3.4s, v1.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn v0.4h, v4.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn v1.4h, v5.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn2 v0.8h, v2.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn2 v1.8h, v3.4s
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: min_v16f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: fmin v0.8h, v0.8h, v2.8h
+; CHECK-FP16-GI-NEXT: fmin v1.8h, v1.8h, v3.8h
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call <16 x half> @llvm.minimum.v16f16(<16 x half> %a, <16 x half> %b)
+ ret <16 x half> %c
+}
+
+define <16 x half> @max_v16f16(<16 x half> %a, <16 x half> %b) {
+; CHECK-NOFP16-SD-LABEL: max_v16f16:
+; CHECK-NOFP16-SD: // %bb.0: // %entry
+; CHECK-NOFP16-SD-NEXT: mov h4, v2.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h5, v0.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h2
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h0
+; CHECK-NOFP16-SD-NEXT: mov h16, v2.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h17, v0.h[2]
+; CHECK-NOFP16-SD-NEXT: mov h18, v3.h[1]
+; CHECK-NOFP16-SD-NEXT: mov h19, v1.h[1]
+; CHECK-NOFP16-SD-NEXT: fcvt s4, h4
+; CHECK-NOFP16-SD-NEXT: fcvt s5, h5
+; CHECK-NOFP16-SD-NEXT: fmax s6, s7, s6
+; CHECK-NOFP16-SD-NEXT: fcvt s20, h1
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: mov h17, v2.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT: fmax s5, s5, s4
+; CHECK-NOFP16-SD-NEXT: fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT: fcvt h4, s6
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[3]
+; CHECK-NOFP16-SD-NEXT: fmax s7, s16, s7
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h3
+; CHECK-NOFP16-SD-NEXT: mov h21, v0.h[4]
+; CHECK-NOFP16-SD-NEXT: fcvt s17, h17
+; CHECK-NOFP16-SD-NEXT: fcvt h5, s5
+; CHECK-NOFP16-SD-NEXT: fmax s18, s19, s18
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT: mov h19, v3.h[2]
+; CHECK-NOFP16-SD-NEXT: fmax s16, s20, s16
+; CHECK-NOFP16-SD-NEXT: mov h20, v1.h[2]
+; CHECK-NOFP16-SD-NEXT: fcvt s21, h21
+; CHECK-NOFP16-SD-NEXT: mov v4.h[1], v5.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h5, s7
+; CHECK-NOFP16-SD-NEXT: mov h7, v2.h[4]
+; CHECK-NOFP16-SD-NEXT: fmax s6, s6, s17
+; CHECK-NOFP16-SD-NEXT: mov h17, v3.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT: fcvt s20, h20
+; CHECK-NOFP16-SD-NEXT: mov v4.h[2], v5.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h5, s16
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h7
+; CHECK-NOFP16-SD-NEXT: fcvt h16, s18
+; CHECK-NOFP16-SD-NEXT: mov h18, v1.h[3]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s6
+; CHECK-NOFP16-SD-NEXT: fmax s7, s21, s7
+; CHECK-NOFP16-SD-NEXT: mov v5.h[1], v16.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: fmax s17, s20, s19
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT: mov h19, v3.h[4]
+; CHECK-NOFP16-SD-NEXT: mov h20, v1.h[4]
+; CHECK-NOFP16-SD-NEXT: mov v4.h[3], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT: fmax s7, s18, s16
+; CHECK-NOFP16-SD-NEXT: fcvt h16, s17
+; CHECK-NOFP16-SD-NEXT: fcvt s17, h19
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h20
+; CHECK-NOFP16-SD-NEXT: mov v4.h[4], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: mov h19, v1.h[5]
+; CHECK-NOFP16-SD-NEXT: mov v5.h[2], v16.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT: fmax s7, s18, s17
+; CHECK-NOFP16-SD-NEXT: mov h16, v2.h[5]
+; CHECK-NOFP16-SD-NEXT: mov h17, v0.h[5]
+; CHECK-NOFP16-SD-NEXT: mov h18, v3.h[5]
+; CHECK-NOFP16-SD-NEXT: mov v5.h[3], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s7
+; CHECK-NOFP16-SD-NEXT: fcvt s7, h16
+; CHECK-NOFP16-SD-NEXT: fcvt s16, h17
+; CHECK-NOFP16-SD-NEXT: fcvt s17, h18
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h19
+; CHECK-NOFP16-SD-NEXT: mov h19, v2.h[6]
+; CHECK-NOFP16-SD-NEXT: mov h2, v2.h[7]
+; CHECK-NOFP16-SD-NEXT: mov v5.h[4], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: mov h6, v0.h[6]
+; CHECK-NOFP16-SD-NEXT: fmax s7, s16, s7
+; CHECK-NOFP16-SD-NEXT: fmax s16, s18, s17
+; CHECK-NOFP16-SD-NEXT: mov h17, v3.h[6]
+; CHECK-NOFP16-SD-NEXT: mov h18, v1.h[6]
+; CHECK-NOFP16-SD-NEXT: fcvt s19, h19
+; CHECK-NOFP16-SD-NEXT: fcvt s6, h6
+; CHECK-NOFP16-SD-NEXT: mov h0, v0.h[7]
+; CHECK-NOFP16-SD-NEXT: mov h3, v3.h[7]
+; CHECK-NOFP16-SD-NEXT: mov h1, v1.h[7]
+; CHECK-NOFP16-SD-NEXT: fcvt s17, h17
+; CHECK-NOFP16-SD-NEXT: fcvt h7, s7
+; CHECK-NOFP16-SD-NEXT: fcvt s18, h18
+; CHECK-NOFP16-SD-NEXT: fcvt s2, h2
+; CHECK-NOFP16-SD-NEXT: fmax s6, s6, s19
+; CHECK-NOFP16-SD-NEXT: fcvt s0, h0
+; CHECK-NOFP16-SD-NEXT: fcvt s3, h3
+; CHECK-NOFP16-SD-NEXT: fcvt s1, h1
+; CHECK-NOFP16-SD-NEXT: fcvt h16, s16
+; CHECK-NOFP16-SD-NEXT: fmax s17, s18, s17
+; CHECK-NOFP16-SD-NEXT: mov v4.h[5], v7.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h6, s6
+; CHECK-NOFP16-SD-NEXT: fmax s0, s0, s2
+; CHECK-NOFP16-SD-NEXT: fmax s1, s1, s3
+; CHECK-NOFP16-SD-NEXT: mov v5.h[5], v16.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h2, s17
+; CHECK-NOFP16-SD-NEXT: mov v4.h[6], v6.h[0]
+; CHECK-NOFP16-SD-NEXT: fcvt h0, s0
+; CHECK-NOFP16-SD-NEXT: fcvt h1, s1
+; CHECK-NOFP16-SD-NEXT: mov v5.h[6], v2.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v4.h[7], v0.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v5.h[7], v1.h[0]
+; CHECK-NOFP16-SD-NEXT: mov v0.16b, v4.16b
+; CHECK-NOFP16-SD-NEXT: mov v1.16b, v5.16b
+; CHECK-NOFP16-SD-NEXT: ret
+;
+; CHECK-FP16-SD-LABEL: max_v16f16:
+; CHECK-FP16-SD: // %bb.0: // %entry
+; CHECK-FP16-SD-NEXT: fmax v0.8h, v0.8h, v2.8h
+; CHECK-FP16-SD-NEXT: fmax v1.8h, v1.8h, v3.8h
+; CHECK-FP16-SD-NEXT: ret
+;
+; CHECK-NOFP16-GI-LABEL: max_v16f16:
+; CHECK-NOFP16-GI: // %bb.0: // %entry
+; CHECK-NOFP16-GI-NEXT: fcvtl v4.4s, v0.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v5.4s, v2.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v6.4s, v1.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl v7.4s, v3.4h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v0.4s, v0.8h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v2.4s, v2.8h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v1.4s, v1.8h
+; CHECK-NOFP16-GI-NEXT: fcvtl2 v3.4s, v3.8h
+; CHECK-NOFP16-GI-NEXT: fmax v4.4s, v4.4s, v5.4s
+; CHECK-NOFP16-GI-NEXT: fmax v5.4s, v6.4s, v7.4s
+; CHECK-NOFP16-GI-NEXT: fmax v2.4s, v0.4s, v2.4s
+; CHECK-NOFP16-GI-NEXT: fmax v3.4s, v1.4s, v3.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn v0.4h, v4.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn v1.4h, v5.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn2 v0.8h, v2.4s
+; CHECK-NOFP16-GI-NEXT: fcvtn2 v1.8h, v3.4s
+; CHECK-NOFP16-GI-NEXT: ret
+;
+; CHECK-FP16-GI-LABEL: max_v16f16:
+; CHECK-FP16-GI: // %bb.0: // %entry
+; CHECK-FP16-GI-NEXT: fmax v0.8h, v0.8h, v2.8h
+; CHECK-FP16-GI-NEXT: fmax v1.8h, v1.8h, v3.8h
+; CHECK-FP16-GI-NEXT: ret
+entry:
+ %c = call <16 x half> @llvm.maximum.v16f16(<16 x half> %a, <16 x half> %b)
+ ret <16 x half> %c
+}
+
+declare <16 x half> @llvm.maximum.v16f16(<16 x half>, <16 x half>)
+declare <16 x half> @llvm.minimum.v16f16(<16 x half>, <16 x half>)
+declare <2 x double> @llvm.maximum.v2f64(<2 x double>, <2 x double>)
+declare <2 x double> @llvm.minimum.v2f64(<2 x double>, <2 x double>)
+declare <2 x float> @llvm.maximum.v2f32(<2 x float>, <2 x float>)
+declare <2 x float> @llvm.minimum.v2f32(<2 x float>, <2 x float>)
+declare <3 x double> @llvm.maximum.v3f64(<3 x double>, <3 x double>)
+declare <3 x double> @llvm.minimum.v3f64(<3 x double>, <3 x double>)
+declare <3 x float> @llvm.maximum.v3f32(<3 x float>, <3 x float>)
+declare <3 x float> @llvm.minimum.v3f32(<3 x float>, <3 x float>)
+declare <4 x double> @llvm.maximum.v4f64(<4 x double>, <4 x double>)
+declare <4 x double> @llvm.minimum.v4f64(<4 x double>, <4 x double>)
+declare <4 x float> @llvm.maximum.v4f32(<4 x float>, <4 x float>)
+declare <4 x float> @llvm.minimum.v4f32(<4 x float>, <4 x float>)
+declare <4 x half> @llvm.maximum.v4f16(<4 x half>, <4 x half>)
+declare <4 x half> @llvm.minimum.v4f16(<4 x half>, <4 x half>)
+declare <7 x float> @llvm.maximum.v7f32(<7 x float>, <7 x float>)
+declare <7 x float> @llvm.minimum.v7f32(<7 x float>, <7 x float>)
+declare <7 x half> @llvm.maximum.v7f16(<7 x half>, <7 x half>)
+declare <7 x half> @llvm.minimum.v7f16(<7 x half>, <7 x half>)
+declare <8 x float> @llvm.maximum.v8f32(<8 x float>, <8 x float>)
+declare <8 x float> @llvm.minimum.v8f32(<8 x float>, <8 x float>)
+declare <8 x half> @llvm.maximum.v8f16(<8 x half>, <8 x half>)
+declare <8 x half> @llvm.minimum.v8f16(<8 x half>, <8 x half>)
+declare double @llvm.maximum.f64(double, double)
+declare double @llvm.minimum.f64(double, double)
+declare float @llvm.maximum.f32(float, float)
+declare float @llvm.minimum.f32(float, float)
+declare half @llvm.maximum.f16(half, half)
+declare half @llvm.minimum.f16(half, half)
More information about the llvm-commits
mailing list