From 5f7dd2527fb39ce24feb24a4b75323cf64729d57 Mon Sep 17 00:00:00 2001
From: greatroar <61184462+greatroar@users.noreply.github.com>
Date: Thu, 30 May 2024 09:57:43 +0200
Subject: [PATCH] flate, zstd: Shave some bytes off amd64 matchLen (#963)

XORQ sets the flags, so there is no need for TESTQ.

Use a 32-bit shift after TZCNT/BSF.
---
 flate/matchlen_amd64.s | 10 ++++------
 zstd/matchlen_amd64.s  | 10 ++++------
 2 files changed, 8 insertions(+), 12 deletions(-)

diff --git a/flate/matchlen_amd64.s b/flate/matchlen_amd64.s
index 9a7655c0f7..0782b86e3d 100644
--- a/flate/matchlen_amd64.s
+++ b/flate/matchlen_amd64.s
@@ -5,7 +5,6 @@
 #include "textflag.h"
 
 // func matchLen(a []byte, b []byte) int
-// Requires: BMI
 TEXT ·matchLen(SB), NOSPLIT, $0-56
 	MOVQ a_base+0(FP), AX
 	MOVQ b_base+24(FP), CX
@@ -17,17 +16,16 @@ TEXT ·matchLen(SB), NOSPLIT, $0-56
 	JB   matchlen_match4_standalone
 
 matchlen_loopback_standalone:
-	MOVQ  (AX)(SI*1), BX
-	XORQ  (CX)(SI*1), BX
-	TESTQ BX, BX
-	JZ    matchlen_loop_standalone
+	MOVQ (AX)(SI*1), BX
+	XORQ (CX)(SI*1), BX
+	JZ   matchlen_loop_standalone
 
 #ifdef GOAMD64_v3
 	TZCNTQ BX, BX
 #else
 	BSFQ BX, BX
 #endif
-	SARQ $0x03, BX
+	SHRL $0x03, BX
 	LEAL (SI)(BX*1), SI
 	JMP  gen_match_len_end
 
diff --git a/zstd/matchlen_amd64.s b/zstd/matchlen_amd64.s
index 9a7655c0f7..0782b86e3d 100644
--- a/zstd/matchlen_amd64.s
+++ b/zstd/matchlen_amd64.s
@@ -5,7 +5,6 @@
 #include "textflag.h"
 
 // func matchLen(a []byte, b []byte) int
-// Requires: BMI
 TEXT ·matchLen(SB), NOSPLIT, $0-56
 	MOVQ a_base+0(FP), AX
 	MOVQ b_base+24(FP), CX
@@ -17,17 +16,16 @@ TEXT ·matchLen(SB), NOSPLIT, $0-56
 	JB   matchlen_match4_standalone
 
 matchlen_loopback_standalone:
-	MOVQ  (AX)(SI*1), BX
-	XORQ  (CX)(SI*1), BX
-	TESTQ BX, BX
-	JZ    matchlen_loop_standalone
+	MOVQ (AX)(SI*1), BX
+	XORQ (CX)(SI*1), BX
+	JZ   matchlen_loop_standalone
 
 #ifdef GOAMD64_v3
 	TZCNTQ BX, BX
 #else
 	BSFQ BX, BX
 #endif
-	SARQ $0x03, BX
+	SHRL $0x03, BX
 	LEAL (SI)(BX*1), SI
 	JMP  gen_match_len_end