2018-03-15 04:15:22 -06:00
|
|
|
// asmcheck
|
|
|
|
|
|
|
|
// Copyright 2018 The Go Authors. All rights reserved.
|
|
|
|
// Use of this source code is governed by a BSD-style
|
|
|
|
// license that can be found in the LICENSE file.
|
|
|
|
|
|
|
|
package codegen
|
|
|
|
|
|
|
|
// This file contains codegen tests related to arithmetic
|
|
|
|
// simplifications and optimizations on float types.
|
|
|
|
// For codegen tests on integer types, see arithmetic.go.
|
|
|
|
|
|
|
|
// --------------------- //
|
|
|
|
// Strength-reduce //
|
|
|
|
// --------------------- //
|
|
|
|
|
|
|
|
func Mul2(f float64) float64 {
|
2020-10-06 15:42:15 -06:00
|
|
|
// 386/sse2:"ADDSD",-"MULSD"
|
2018-03-15 04:15:22 -06:00
|
|
|
// amd64:"ADDSD",-"MULSD"
|
2018-04-15 11:00:27 -06:00
|
|
|
// arm/7:"ADDD",-"MULD"
|
2018-03-15 04:15:22 -06:00
|
|
|
// arm64:"FADDD",-"FMULD"
|
2023-01-25 10:53:10 -07:00
|
|
|
// ppc64x:"FADD",-"FMUL"
|
2023-06-24 21:20:35 -06:00
|
|
|
// riscv64:"FADDD",-"FMULD"
|
2018-03-15 04:15:22 -06:00
|
|
|
return f * 2.0
|
|
|
|
}
|
|
|
|
|
|
|
|
func DivPow2(f1, f2, f3 float64) (float64, float64, float64) {
|
2020-10-06 15:42:15 -06:00
|
|
|
// 386/sse2:"MULSD",-"DIVSD"
|
2018-03-15 04:15:22 -06:00
|
|
|
// amd64:"MULSD",-"DIVSD"
|
2018-04-15 11:00:27 -06:00
|
|
|
// arm/7:"MULD",-"DIVD"
|
2018-03-15 04:15:22 -06:00
|
|
|
// arm64:"FMULD",-"FDIVD"
|
2023-01-25 10:53:10 -07:00
|
|
|
// ppc64x:"FMUL",-"FDIV"
|
2023-06-24 21:20:35 -06:00
|
|
|
// riscv64:"FMULD",-"FDIVD"
|
2018-03-15 04:15:22 -06:00
|
|
|
x := f1 / 16.0
|
|
|
|
|
2020-10-06 15:42:15 -06:00
|
|
|
// 386/sse2:"MULSD",-"DIVSD"
|
2018-03-15 04:15:22 -06:00
|
|
|
// amd64:"MULSD",-"DIVSD"
|
2018-04-15 11:00:27 -06:00
|
|
|
// arm/7:"MULD",-"DIVD"
|
2018-03-15 04:15:22 -06:00
|
|
|
// arm64:"FMULD",-"FDIVD"
|
2023-01-25 10:53:10 -07:00
|
|
|
// ppc64x:"FMUL",-"FDIVD"
|
2023-06-24 21:20:35 -06:00
|
|
|
// riscv64:"FMULD",-"FDIVD"
|
2018-03-15 04:15:22 -06:00
|
|
|
y := f2 / 0.125
|
|
|
|
|
2020-10-06 15:42:15 -06:00
|
|
|
// 386/sse2:"ADDSD",-"DIVSD",-"MULSD"
|
2018-03-15 04:15:22 -06:00
|
|
|
// amd64:"ADDSD",-"DIVSD",-"MULSD"
|
2018-04-15 11:00:27 -06:00
|
|
|
// arm/7:"ADDD",-"MULD",-"DIVD"
|
2018-03-15 04:15:22 -06:00
|
|
|
// arm64:"FADDD",-"FMULD",-"FDIVD"
|
2023-01-25 10:53:10 -07:00
|
|
|
// ppc64x:"FADD",-"FMUL",-"FDIV"
|
2023-06-24 21:20:35 -06:00
|
|
|
// riscv64:"FADDD",-"FMULD",-"FDIVD"
|
2018-03-15 04:15:22 -06:00
|
|
|
z := f3 / 0.5
|
|
|
|
|
|
|
|
return x, y, z
|
|
|
|
}
|
|
|
|
|
2018-07-18 03:31:35 -06:00
|
|
|
func indexLoad(b0 []float32, b1 float32, idx int) float32 {
|
2020-11-27 08:10:33 -07:00
|
|
|
// arm64:`FMOVS\s\(R[0-9]+\)\(R[0-9]+<<2\),\sF[0-9]+`
|
2018-07-18 03:31:35 -06:00
|
|
|
return b0[idx] * b1
|
|
|
|
}
|
|
|
|
|
|
|
|
func indexStore(b0 []float64, b1 float64, idx int) {
|
2020-11-27 08:10:33 -07:00
|
|
|
// arm64:`FMOVD\sF[0-9]+,\s\(R[0-9]+\)\(R[0-9]+<<3\)`
|
2018-07-18 03:31:35 -06:00
|
|
|
b0[idx] = b1
|
|
|
|
}
|
|
|
|
|
2018-03-15 04:15:22 -06:00
|
|
|
// ----------- //
|
|
|
|
// Fused //
|
|
|
|
// ----------- //
|
|
|
|
|
|
|
|
func FusedAdd32(x, y, z float32) float32 {
|
|
|
|
// s390x:"FMADDS\t"
|
2023-01-25 10:53:10 -07:00
|
|
|
// ppc64x:"FMADDS\t"
|
2018-10-14 23:39:23 -06:00
|
|
|
// arm64:"FMADDS"
|
2018-03-15 04:15:22 -06:00
|
|
|
return x*y + z
|
|
|
|
}
|
|
|
|
|
2018-10-14 23:39:23 -06:00
|
|
|
func FusedSub32_a(x, y, z float32) float32 {
|
2018-03-15 04:15:22 -06:00
|
|
|
// s390x:"FMSUBS\t"
|
2023-01-25 10:53:10 -07:00
|
|
|
// ppc64x:"FMSUBS\t"
|
2018-03-15 04:15:22 -06:00
|
|
|
return x*y - z
|
|
|
|
}
|
|
|
|
|
2018-10-14 23:39:23 -06:00
|
|
|
func FusedSub32_b(x, y, z float32) float32 {
|
|
|
|
// arm64:"FMSUBS"
|
|
|
|
return z - x*y
|
|
|
|
}
|
|
|
|
|
2018-03-15 04:15:22 -06:00
|
|
|
func FusedAdd64(x, y, z float64) float64 {
|
|
|
|
// s390x:"FMADD\t"
|
2023-01-25 10:53:10 -07:00
|
|
|
// ppc64x:"FMADD\t"
|
2018-10-14 23:39:23 -06:00
|
|
|
// arm64:"FMADDD"
|
2023-06-26 06:46:49 -06:00
|
|
|
// riscv64:"FMADDD\t"
|
2018-03-15 04:15:22 -06:00
|
|
|
return x*y + z
|
|
|
|
}
|
|
|
|
|
2018-10-14 23:39:23 -06:00
|
|
|
func FusedSub64_a(x, y, z float64) float64 {
|
2018-03-15 04:15:22 -06:00
|
|
|
// s390x:"FMSUB\t"
|
2023-01-25 10:53:10 -07:00
|
|
|
// ppc64x:"FMSUB\t"
|
2023-06-26 06:46:49 -06:00
|
|
|
// riscv64:"FMSUBD\t"
|
2018-03-15 04:15:22 -06:00
|
|
|
return x*y - z
|
|
|
|
}
|
2018-04-15 11:17:41 -06:00
|
|
|
|
2018-10-14 23:39:23 -06:00
|
|
|
func FusedSub64_b(x, y, z float64) float64 {
|
|
|
|
// arm64:"FMSUBD"
|
2023-06-26 06:46:49 -06:00
|
|
|
// riscv64:"FNMSUBD\t"
|
2018-10-14 23:39:23 -06:00
|
|
|
return z - x*y
|
|
|
|
}
|
|
|
|
|
2019-05-04 21:35:37 -06:00
|
|
|
func Cmp(f float64) bool {
|
2020-03-06 15:20:45 -07:00
|
|
|
// arm64:"FCMPD","(BGT|BLE|BMI|BPL)",-"CSET\tGT",-"CBZ"
|
2019-05-04 21:35:37 -06:00
|
|
|
return f > 4 || f < -4
|
|
|
|
}
|
|
|
|
|
2019-11-26 13:33:37 -07:00
|
|
|
func CmpZero64(f float64) bool {
|
|
|
|
// s390x:"LTDBR",-"FCMPU"
|
|
|
|
return f <= 0
|
|
|
|
}
|
|
|
|
|
|
|
|
func CmpZero32(f float32) bool {
|
|
|
|
// s390x:"LTEBR",-"CEBR"
|
|
|
|
return f <= 0
|
|
|
|
}
|
|
|
|
|
2019-11-21 08:44:23 -07:00
|
|
|
func CmpWithSub(a float64, b float64) bool {
|
|
|
|
f := a - b
|
|
|
|
// s390x:-"LTDBR"
|
|
|
|
return f <= 0
|
|
|
|
}
|
|
|
|
|
|
|
|
func CmpWithAdd(a float64, b float64) bool {
|
|
|
|
f := a + b
|
|
|
|
// s390x:-"LTDBR"
|
|
|
|
return f <= 0
|
|
|
|
}
|
|
|
|
|
2018-04-15 11:17:41 -06:00
|
|
|
// ---------------- //
|
|
|
|
// Non-floats //
|
|
|
|
// ---------------- //
|
|
|
|
|
|
|
|
// We should make sure that the compiler doesn't generate floating point
|
|
|
|
// instructions for non-float operations on Plan 9, because floating point
|
|
|
|
// operations are not allowed in the note handler.
|
|
|
|
|
|
|
|
func ArrayZero() [16]byte {
|
|
|
|
// amd64:"MOVUPS"
|
|
|
|
// plan9/amd64/:-"MOVUPS"
|
|
|
|
var a [16]byte
|
|
|
|
return a
|
|
|
|
}
|
|
|
|
|
|
|
|
func ArrayCopy(a [16]byte) (b [16]byte) {
|
|
|
|
// amd64:"MOVUPS"
|
|
|
|
// plan9/amd64/:-"MOVUPS"
|
|
|
|
b = a
|
|
|
|
return
|
|
|
|
}
|