From 5c67ebbb31a296ca1593d0229b1d51d5ac73aa6d Mon Sep 17 00:00:00 2001
From: Jorropo <jorropo.pgm@gmail.com>
Date: Sun, 6 Nov 2022 06:37:13 +0100
Subject: cmd/compile: AMD64v3 remove unnecessary TEST comparision in
 isPowerOfTwo

With GOAMD64=V3 the canonical isPowerOfTwo function:
  func isPowerOfTwo(x uintptr) bool {
    return x&(x-1) == 0
  }

Used to compile to:
  temp := BLSR(x) // x&(x-1)
  flags = TEST(temp, temp)
  return flags.zf

However the blsr instruction already set ZF according to the result.
So we can remove the TEST instruction if we are just checking ZF.
Such as in multiple pieces of code around memory allocations.

This make the code smaller and faster.

Change-Id: Ia12d5a73aa3cb49188c0b647b1eff7b56c5a7b58
Reviewed-on: https://go-review.googlesource.com/c/go/+/448255
Run-TryBot: Jakub Ciolek <jakub@ciolek.dev>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
---
 test/codegen/bmi.go | 104 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 104 insertions(+)

(limited to 'test/codegen')

diff --git a/test/codegen/bmi.go b/test/codegen/bmi.go
index 3b125a1b59..aa61b03928 100644
--- a/test/codegen/bmi.go
+++ b/test/codegen/bmi.go
@@ -46,6 +46,110 @@ func blsr32(x int32) int32 {
 	return x & (x - 1)
 }
 
+func isPowerOfTwo64(x int64) bool {
+	// amd64/v3:"BLSRQ",-"TESTQ",-"CALL"
+	return blsr64(x) == 0
+}
+
+func isPowerOfTwo32(x int32) bool {
+	// amd64/v3:"BLSRL",-"TESTL",-"CALL"
+	return blsr32(x) == 0
+}
+
+func isPowerOfTwoSelect64(x, a, b int64) int64 {
+	var r int64
+	// amd64/v3:"BLSRQ",-"TESTQ",-"CALL"
+	if isPowerOfTwo64(x) {
+		r = a
+	} else {
+		r = b
+	}
+	// amd64/v3:"CMOVQEQ",-"TESTQ",-"CALL"
+	return r * 2 // force return blocks joining
+}
+
+func isPowerOfTwoSelect32(x, a, b int32) int32 {
+	var r int32
+	// amd64/v3:"BLSRL",-"TESTL",-"CALL"
+	if isPowerOfTwo32(x) {
+		r = a
+	} else {
+		r = b
+	}
+	// amd64/v3:"CMOVLEQ",-"TESTL",-"CALL"
+	return r * 2 // force return blocks joining
+}
+
+func isPowerOfTwoBranch64(x int64, a func(bool), b func(string)) {
+	// amd64/v3:"BLSRQ",-"TESTQ",-"CALL"
+	if isPowerOfTwo64(x) {
+		a(true)
+	} else {
+		b("false")
+	}
+}
+
+func isPowerOfTwoBranch32(x int32, a func(bool), b func(string)) {
+	// amd64/v3:"BLSRL",-"TESTL",-"CALL"
+	if isPowerOfTwo32(x) {
+		a(true)
+	} else {
+		b("false")
+	}
+}
+
+func isNotPowerOfTwo64(x int64) bool {
+	// amd64/v3:"BLSRQ",-"TESTQ",-"CALL"
+	return blsr64(x) != 0
+}
+
+func isNotPowerOfTwo32(x int32) bool {
+	// amd64/v3:"BLSRL",-"TESTL",-"CALL"
+	return blsr32(x) != 0
+}
+
+func isNotPowerOfTwoSelect64(x, a, b int64) int64 {
+	var r int64
+	// amd64/v3:"BLSRQ",-"TESTQ",-"CALL"
+	if isNotPowerOfTwo64(x) {
+		r = a
+	} else {
+		r = b
+	}
+	// amd64/v3:"CMOVQNE",-"TESTQ",-"CALL"
+	return r * 2 // force return blocks joining
+}
+
+func isNotPowerOfTwoSelect32(x, a, b int32) int32 {
+	var r int32
+	// amd64/v3:"BLSRL",-"TESTL",-"CALL"
+	if isNotPowerOfTwo32(x) {
+		r = a
+	} else {
+		r = b
+	}
+	// amd64/v3:"CMOVLNE",-"TESTL",-"CALL"
+	return r * 2 // force return blocks joining
+}
+
+func isNotPowerOfTwoBranch64(x int64, a func(bool), b func(string)) {
+	// amd64/v3:"BLSRQ",-"TESTQ",-"CALL"
+	if isNotPowerOfTwo64(x) {
+		a(true)
+	} else {
+		b("false")
+	}
+}
+
+func isNotPowerOfTwoBranch32(x int32, a func(bool), b func(string)) {
+	// amd64/v3:"BLSRL",-"TESTL",-"CALL"
+	if isNotPowerOfTwo32(x) {
+		a(true)
+	} else {
+		b("false")
+	}
+}
+
 func sarx64(x, y int64) int64 {
 	// amd64/v3:"SARXQ"
 	return x >> y
-- 
cgit v1.3