c65238
The memmove related fix is dropped in this patch because rhel-7.5
c65238
does not have optimized memmove for POWER7.
c65238
c65238
commit 63da5cd4a097d089033d980c42254c3356fa723f
c65238
Author: Rajalakshmi Srinivasaraghavan <raji@linux.vnet.ibm.com>
c65238
Date:   Wed Oct 25 13:13:53 2017 -0200
c65238
c65238
    powerpc: Replace lxvd2x/stxvd2x with lvx/stvx in P7's memcpy/memmove
c65238
    
c65238
    POWER9 DD2.1 and earlier has an issue where some cache inhibited
c65238
    vector load traps to the kernel, causing a performance degradation.  To
c65238
    handle this in memcpy and memmove, lvx/stvx is used for aligned
c65238
    addresses instead of lxvd2x/stxvd2x.
c65238
    
c65238
    Reference: https://patchwork.ozlabs.org/patch/814059/
c65238
    
c65238
            * sysdeps/powerpc/powerpc64/power7/memcpy.S: Replace
c65238
            lxvd2x/stxvd2x with lvx/stvx.
c65238
            * sysdeps/powerpc/powerpc64/power7/memmove.S: Likewise.
c65238
    
c65238
    Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.vnet.ibm.com>
c65238
    Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
c65238
c65238
diff --git a/sysdeps/powerpc/powerpc64/power7/memcpy.S b/sysdeps/powerpc/powerpc64/power7/memcpy.S
c65238
index 1ccbc2e..a7cdf8b 100644
c65238
--- a/sysdeps/powerpc/powerpc64/power7/memcpy.S
c65238
+++ b/sysdeps/powerpc/powerpc64/power7/memcpy.S
c65238
@@ -91,63 +91,63 @@ L(aligned_copy):
c65238
 	srdi	12,cnt,7
c65238
 	cmpdi	12,0
c65238
 	beq	L(aligned_tail)
c65238
-	lxvd2x	6,0,src
c65238
-	lxvd2x	7,src,6
c65238
+	lvx	6,0,src
c65238
+	lvx	7,src,6
c65238
 	mtctr	12
c65238
 	b	L(aligned_128loop)
c65238
 
c65238
 	.align  4
c65238
 L(aligned_128head):
c65238
 	/* for the 2nd + iteration of this loop. */
c65238
-	lxvd2x	6,0,src
c65238
-	lxvd2x	7,src,6
c65238
+	lvx	6,0,src
c65238
+	lvx	7,src,6
c65238
 L(aligned_128loop):
c65238
-	lxvd2x	8,src,7
c65238
-	lxvd2x	9,src,8
c65238
-	stxvd2x	6,0,dst
c65238
+	lvx	8,src,7
c65238
+	lvx	9,src,8
c65238
+	stvx	6,0,dst
c65238
 	addi	src,src,64
c65238
-	stxvd2x	7,dst,6
c65238
-	stxvd2x	8,dst,7
c65238
-	stxvd2x	9,dst,8
c65238
-	lxvd2x	6,0,src
c65238
-	lxvd2x	7,src,6
c65238
+	stvx	7,dst,6
c65238
+	stvx	8,dst,7
c65238
+	stvx	9,dst,8
c65238
+	lvx	6,0,src
c65238
+	lvx	7,src,6
c65238
 	addi	dst,dst,64
c65238
-	lxvd2x	8,src,7
c65238
-	lxvd2x	9,src,8
c65238
+	lvx	8,src,7
c65238
+	lvx	9,src,8
c65238
 	addi	src,src,64
c65238
-	stxvd2x	6,0,dst
c65238
-	stxvd2x	7,dst,6
c65238
-	stxvd2x	8,dst,7
c65238
-	stxvd2x	9,dst,8
c65238
+	stvx	6,0,dst
c65238
+	stvx	7,dst,6
c65238
+	stvx	8,dst,7
c65238
+	stvx	9,dst,8
c65238
 	addi	dst,dst,64
c65238
 	bdnz	L(aligned_128head)
c65238
 
c65238
 L(aligned_tail):
c65238
 	mtocrf	0x01,cnt
c65238
 	bf	25,32f
c65238
-	lxvd2x	6,0,src
c65238
-	lxvd2x	7,src,6
c65238
-	lxvd2x	8,src,7
c65238
-	lxvd2x	9,src,8
c65238
+	lvx	6,0,src
c65238
+	lvx	7,src,6
c65238
+	lvx	8,src,7
c65238
+	lvx	9,src,8
c65238
 	addi	src,src,64
c65238
-	stxvd2x	6,0,dst
c65238
-	stxvd2x	7,dst,6
c65238
-	stxvd2x	8,dst,7
c65238
-	stxvd2x	9,dst,8
c65238
+	stvx	6,0,dst
c65238
+	stvx	7,dst,6
c65238
+	stvx	8,dst,7
c65238
+	stvx	9,dst,8
c65238
 	addi	dst,dst,64
c65238
 32:
c65238
 	bf	26,16f
c65238
-	lxvd2x	6,0,src
c65238
-	lxvd2x	7,src,6
c65238
+	lvx	6,0,src
c65238
+	lvx	7,src,6
c65238
 	addi	src,src,32
c65238
-	stxvd2x	6,0,dst
c65238
-	stxvd2x	7,dst,6
c65238
+	stvx	6,0,dst
c65238
+	stvx	7,dst,6
c65238
 	addi	dst,dst,32
c65238
 16:
c65238
 	bf	27,8f
c65238
-	lxvd2x	6,0,src
c65238
+	lvx	6,0,src
c65238
 	addi	src,src,16
c65238
-	stxvd2x	6,0,dst
c65238
+	stvx	6,0,dst
c65238
 	addi	dst,dst,16
c65238
 8:
c65238
 	bf	28,4f