commit 3c55c207564c0ae30d78d01689b4ae16bf38dd63
Author: Noah Goldstein <goldstein.w.n@gmail.com>
Date:   Wed Mar 23 16:57:16 2022 -0500

    x86: Code cleanup in strchr-avx2 and comment justifying branch
    
    Small code cleanup for size: -53 bytes.
    
    Add comment justifying using a branch to do NULL/non-null return.
    
    All string/memory tests pass and no regressions in benchtests.
    
    geometric_mean(N=20) of all benchmarks Original / New: 1.00
    Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
    
    (cherry picked from commit a6fbf4d51e9ba8063c4f8331564892ead9c67344)

diff --git a/sysdeps/x86_64/multiarch/strchr-avx2.S b/sysdeps/x86_64/multiarch/strchr-avx2.S
index 413942b96a835c4a..ef4ce0f3677e30c8 100644
--- a/sysdeps/x86_64/multiarch/strchr-avx2.S
+++ b/sysdeps/x86_64/multiarch/strchr-avx2.S
@@ -48,13 +48,13 @@
 # define PAGE_SIZE 4096
 
 	.section SECTION(.text),"ax",@progbits
-ENTRY (STRCHR)
+ENTRY_P2ALIGN (STRCHR, 5)
 	/* Broadcast CHAR to YMM0.	*/
 	vmovd	%esi, %xmm0
 	movl	%edi, %eax
 	andl	$(PAGE_SIZE - 1), %eax
 	VPBROADCAST	%xmm0, %ymm0
-	vpxor	%xmm9, %xmm9, %xmm9
+	vpxor	%xmm1, %xmm1, %xmm1
 
 	/* Check if we cross page boundary with one vector load.  */
 	cmpl	$(PAGE_SIZE - VEC_SIZE), %eax
@@ -62,37 +62,29 @@ ENTRY (STRCHR)
 
 	/* Check the first VEC_SIZE bytes.	Search for both CHAR and the
 	   null byte.  */
-	vmovdqu	(%rdi), %ymm8
-	VPCMPEQ	%ymm8, %ymm0, %ymm1
-	VPCMPEQ	%ymm8, %ymm9, %ymm2
-	vpor	%ymm1, %ymm2, %ymm1
-	vpmovmskb %ymm1, %eax
+	vmovdqu	(%rdi), %ymm2
+	VPCMPEQ	%ymm2, %ymm0, %ymm3
+	VPCMPEQ	%ymm2, %ymm1, %ymm2
+	vpor	%ymm3, %ymm2, %ymm3
+	vpmovmskb %ymm3, %eax
 	testl	%eax, %eax
 	jz	L(aligned_more)
 	tzcntl	%eax, %eax
 # ifndef USE_AS_STRCHRNUL
-	/* Found CHAR or the null byte.	 */
-	cmp	(%rdi, %rax), %CHAR_REG
-	jne	L(zero)
-# endif
-	addq	%rdi, %rax
-	VZEROUPPER_RETURN
-
-	/* .p2align 5 helps keep performance more consistent if ENTRY()
-	   alignment % 32 was either 16 or 0. As well this makes the
-	   alignment % 32 of the loop_4x_vec fixed which makes tuning it
-	   easier.  */
-	.p2align 5
-L(first_vec_x4):
-	tzcntl	%eax, %eax
-	addq	$(VEC_SIZE * 3 + 1), %rdi
-# ifndef USE_AS_STRCHRNUL
-	/* Found CHAR or the null byte.	 */
+	/* Found CHAR or the null byte.  */
 	cmp	(%rdi, %rax), %CHAR_REG
+	/* NB: Use a branch instead of cmovcc here. The expectation is
+	   that with strchr the user will branch based on input being
+	   null. Since this branch will be 100% predictive of the user
+	   branch a branch miss here should save what otherwise would
+	   be branch miss in the user code. Otherwise using a branch 1)
+	   saves code size and 2) is faster in highly predictable
+	   environments.  */
 	jne	L(zero)
 # endif
 	addq	%rdi, %rax
-	VZEROUPPER_RETURN
+L(return_vzeroupper):
+	ZERO_UPPER_VEC_REGISTERS_RETURN
 
 # ifndef USE_AS_STRCHRNUL
 L(zero):
@@ -103,7 +95,8 @@ L(zero):
 
 	.p2align 4
 L(first_vec_x1):
-	tzcntl	%eax, %eax
+	/* Use bsf to save code size.  */
+	bsfl	%eax, %eax
 	incq	%rdi
 # ifndef USE_AS_STRCHRNUL
 	/* Found CHAR or the null byte.	 */
@@ -113,9 +106,10 @@ L(first_vec_x1):
 	addq	%rdi, %rax
 	VZEROUPPER_RETURN
 
-	.p2align 4
+	.p2align 4,, 10
 L(first_vec_x2):
-	tzcntl	%eax, %eax
+	/* Use bsf to save code size.  */
+	bsfl	%eax, %eax
 	addq	$(VEC_SIZE + 1), %rdi
 # ifndef USE_AS_STRCHRNUL
 	/* Found CHAR or the null byte.	 */
@@ -125,9 +119,10 @@ L(first_vec_x2):
 	addq	%rdi, %rax
 	VZEROUPPER_RETURN
 
-	.p2align 4
+	.p2align 4,, 8
 L(first_vec_x3):
-	tzcntl	%eax, %eax
+	/* Use bsf to save code size.  */
+	bsfl	%eax, %eax
 	addq	$(VEC_SIZE * 2 + 1), %rdi
 # ifndef USE_AS_STRCHRNUL
 	/* Found CHAR or the null byte.	 */
@@ -137,6 +132,21 @@ L(first_vec_x3):
 	addq	%rdi, %rax
 	VZEROUPPER_RETURN
 
+	.p2align 4,, 10
+L(first_vec_x4):
+	/* Use bsf to save code size.  */
+	bsfl	%eax, %eax
+	addq	$(VEC_SIZE * 3 + 1), %rdi
+# ifndef USE_AS_STRCHRNUL
+	/* Found CHAR or the null byte.	 */
+	cmp	(%rdi, %rax), %CHAR_REG
+	jne	L(zero)
+# endif
+	addq	%rdi, %rax
+	VZEROUPPER_RETURN
+
+
+
 	.p2align 4
 L(aligned_more):
 	/* Align data to VEC_SIZE - 1. This is the same number of
@@ -146,90 +156,92 @@ L(aligned_more):
 L(cross_page_continue):
 	/* Check the next 4 * VEC_SIZE.  Only one VEC_SIZE at a time
 	   since data is only aligned to VEC_SIZE.  */
-	vmovdqa	1(%rdi), %ymm8
-	VPCMPEQ	%ymm8, %ymm0, %ymm1
-	VPCMPEQ	%ymm8, %ymm9, %ymm2
-	vpor	%ymm1, %ymm2, %ymm1
-	vpmovmskb %ymm1, %eax
+	vmovdqa	1(%rdi), %ymm2
+	VPCMPEQ	%ymm2, %ymm0, %ymm3
+	VPCMPEQ	%ymm2, %ymm1, %ymm2
+	vpor	%ymm3, %ymm2, %ymm3
+	vpmovmskb %ymm3, %eax
 	testl	%eax, %eax
 	jnz	L(first_vec_x1)
 
-	vmovdqa	(VEC_SIZE + 1)(%rdi), %ymm8
-	VPCMPEQ	%ymm8, %ymm0, %ymm1
-	VPCMPEQ	%ymm8, %ymm9, %ymm2
-	vpor	%ymm1, %ymm2, %ymm1
-	vpmovmskb %ymm1, %eax
+	vmovdqa	(VEC_SIZE + 1)(%rdi), %ymm2
+	VPCMPEQ	%ymm2, %ymm0, %ymm3
+	VPCMPEQ	%ymm2, %ymm1, %ymm2
+	vpor	%ymm3, %ymm2, %ymm3
+	vpmovmskb %ymm3, %eax
 	testl	%eax, %eax
 	jnz	L(first_vec_x2)
 
-	vmovdqa	(VEC_SIZE * 2 + 1)(%rdi), %ymm8
-	VPCMPEQ	%ymm8, %ymm0, %ymm1
-	VPCMPEQ	%ymm8, %ymm9, %ymm2
-	vpor	%ymm1, %ymm2, %ymm1
-	vpmovmskb %ymm1, %eax
+	vmovdqa	(VEC_SIZE * 2 + 1)(%rdi), %ymm2
+	VPCMPEQ	%ymm2, %ymm0, %ymm3
+	VPCMPEQ	%ymm2, %ymm1, %ymm2
+	vpor	%ymm3, %ymm2, %ymm3
+	vpmovmskb %ymm3, %eax
 	testl	%eax, %eax
 	jnz	L(first_vec_x3)
 
-	vmovdqa	(VEC_SIZE * 3 + 1)(%rdi), %ymm8
-	VPCMPEQ	%ymm8, %ymm0, %ymm1
-	VPCMPEQ	%ymm8, %ymm9, %ymm2
-	vpor	%ymm1, %ymm2, %ymm1
-	vpmovmskb %ymm1, %eax
+	vmovdqa	(VEC_SIZE * 3 + 1)(%rdi), %ymm2
+	VPCMPEQ	%ymm2, %ymm0, %ymm3
+	VPCMPEQ	%ymm2, %ymm1, %ymm2
+	vpor	%ymm3, %ymm2, %ymm3
+	vpmovmskb %ymm3, %eax
 	testl	%eax, %eax
 	jnz	L(first_vec_x4)
-	/* Align data to VEC_SIZE * 4 - 1.	*/
-	addq	$(VEC_SIZE * 4 + 1), %rdi
-	andq	$-(VEC_SIZE * 4), %rdi
+	/* Align data to VEC_SIZE * 4 - 1.  */
+	incq	%rdi
+	orq	$(VEC_SIZE * 4 - 1), %rdi
 	.p2align 4
 L(loop_4x_vec):
 	/* Compare 4 * VEC at a time forward.  */
-	vmovdqa	(%rdi), %ymm5
-	vmovdqa	(VEC_SIZE)(%rdi), %ymm6
-	vmovdqa	(VEC_SIZE * 2)(%rdi), %ymm7
-	vmovdqa	(VEC_SIZE * 3)(%rdi), %ymm8
+	vmovdqa	1(%rdi), %ymm6
+	vmovdqa	(VEC_SIZE + 1)(%rdi), %ymm7
 
 	/* Leaves only CHARS matching esi as 0.	 */
-	vpxor	%ymm5, %ymm0, %ymm1
 	vpxor	%ymm6, %ymm0, %ymm2
 	vpxor	%ymm7, %ymm0, %ymm3
-	vpxor	%ymm8, %ymm0, %ymm4
 
-	VPMINU	%ymm1, %ymm5, %ymm1
 	VPMINU	%ymm2, %ymm6, %ymm2
 	VPMINU	%ymm3, %ymm7, %ymm3
-	VPMINU	%ymm4, %ymm8, %ymm4
 
-	VPMINU	%ymm1, %ymm2, %ymm5
-	VPMINU	%ymm3, %ymm4, %ymm6
+	vmovdqa	(VEC_SIZE * 2 + 1)(%rdi), %ymm6
+	vmovdqa	(VEC_SIZE * 3 + 1)(%rdi), %ymm7
+
+	vpxor	%ymm6, %ymm0, %ymm4
+	vpxor	%ymm7, %ymm0, %ymm5
+
+	VPMINU	%ymm4, %ymm6, %ymm4
+	VPMINU	%ymm5, %ymm7, %ymm5
 
-	VPMINU	%ymm5, %ymm6, %ymm6
+	VPMINU	%ymm2, %ymm3, %ymm6
+	VPMINU	%ymm4, %ymm5, %ymm7
 
-	VPCMPEQ	%ymm6, %ymm9, %ymm6
-	vpmovmskb %ymm6, %ecx
+	VPMINU	%ymm6, %ymm7, %ymm7
+
+	VPCMPEQ	%ymm7, %ymm1, %ymm7
+	vpmovmskb %ymm7, %ecx
 	subq	$-(VEC_SIZE * 4), %rdi
 	testl	%ecx, %ecx
 	jz	L(loop_4x_vec)
 
-
-	VPCMPEQ	%ymm1, %ymm9, %ymm1
-	vpmovmskb %ymm1, %eax
+	VPCMPEQ	%ymm2, %ymm1, %ymm2
+	vpmovmskb %ymm2, %eax
 	testl	%eax, %eax
 	jnz	L(last_vec_x0)
 
 
-	VPCMPEQ	%ymm5, %ymm9, %ymm2
-	vpmovmskb %ymm2, %eax
+	VPCMPEQ	%ymm3, %ymm1, %ymm3
+	vpmovmskb %ymm3, %eax
 	testl	%eax, %eax
 	jnz	L(last_vec_x1)
 
-	VPCMPEQ	%ymm3, %ymm9, %ymm3
-	vpmovmskb %ymm3, %eax
+	VPCMPEQ	%ymm4, %ymm1, %ymm4
+	vpmovmskb %ymm4, %eax
 	/* rcx has combined result from all 4 VEC. It will only be used
 	   if the first 3 other VEC all did not contain a match.  */
 	salq	$32, %rcx
 	orq	%rcx, %rax
 	tzcntq	%rax, %rax
-	subq	$(VEC_SIZE * 2), %rdi
+	subq	$(VEC_SIZE * 2 - 1), %rdi
 # ifndef USE_AS_STRCHRNUL
 	/* Found CHAR or the null byte.	 */
 	cmp	(%rdi, %rax), %CHAR_REG
@@ -239,10 +251,11 @@ L(loop_4x_vec):
 	VZEROUPPER_RETURN
 
 
-	.p2align 4
+	.p2align 4,, 10
 L(last_vec_x0):
-	tzcntl	%eax, %eax
-	addq	$-(VEC_SIZE * 4), %rdi
+	/* Use bsf to save code size.  */
+	bsfl	%eax, %eax
+	addq	$-(VEC_SIZE * 4 - 1), %rdi
 # ifndef USE_AS_STRCHRNUL
 	/* Found CHAR or the null byte.	 */
 	cmp	(%rdi, %rax), %CHAR_REG
@@ -251,16 +264,11 @@ L(last_vec_x0):
 	addq	%rdi, %rax
 	VZEROUPPER_RETURN
 
-# ifndef USE_AS_STRCHRNUL
-L(zero_end):
-	xorl	%eax, %eax
-	VZEROUPPER_RETURN
-# endif
 
-	.p2align 4
+	.p2align 4,, 10
 L(last_vec_x1):
 	tzcntl	%eax, %eax
-	subq	$(VEC_SIZE * 3), %rdi
+	subq	$(VEC_SIZE * 3 - 1), %rdi
 # ifndef USE_AS_STRCHRNUL
 	/* Found CHAR or the null byte.	 */
 	cmp	(%rdi, %rax), %CHAR_REG
@@ -269,18 +277,23 @@ L(last_vec_x1):
 	addq	%rdi, %rax
 	VZEROUPPER_RETURN
 
+# ifndef USE_AS_STRCHRNUL
+L(zero_end):
+	xorl	%eax, %eax
+	VZEROUPPER_RETURN
+# endif
 
 	/* Cold case for crossing page with first load.	 */
-	.p2align 4
+	.p2align 4,, 8
 L(cross_page_boundary):
 	movq	%rdi, %rdx
 	/* Align rdi to VEC_SIZE - 1.  */
 	orq	$(VEC_SIZE - 1), %rdi
-	vmovdqa	-(VEC_SIZE - 1)(%rdi), %ymm8
-	VPCMPEQ	%ymm8, %ymm0, %ymm1
-	VPCMPEQ	%ymm8, %ymm9, %ymm2
-	vpor	%ymm1, %ymm2, %ymm1
-	vpmovmskb %ymm1, %eax
+	vmovdqa	-(VEC_SIZE - 1)(%rdi), %ymm2
+	VPCMPEQ	%ymm2, %ymm0, %ymm3
+	VPCMPEQ	%ymm2, %ymm1, %ymm2
+	vpor	%ymm3, %ymm2, %ymm3
+	vpmovmskb %ymm3, %eax
 	/* Remove the leading bytes. sarxl only uses bits [5:0] of COUNT
 	   so no need to manually mod edx.  */
 	sarxl	%edx, %eax, %eax
@@ -291,13 +304,10 @@ L(cross_page_boundary):
 	xorl	%ecx, %ecx
 	/* Found CHAR or the null byte.	 */
 	cmp	(%rdx, %rax), %CHAR_REG
-	leaq	(%rdx, %rax), %rax
-	cmovne	%rcx, %rax
-# else
-	addq	%rdx, %rax
+	jne	L(zero_end)
 # endif
-L(return_vzeroupper):
-	ZERO_UPPER_VEC_REGISTERS_RETURN
+	addq	%rdx, %rax
+	VZEROUPPER_RETURN
 
 END (STRCHR)
-# endif
+#endif