Eliminate the use of the MASKMOVDQU instruction, to speed up decompression performance by 10x on AMD Bobcat embedded processors (and ~5% on AMD desktop processors.)

git-svn-id: svn://svn.code.sf.net/p/libjpeg-turbo/code/trunk@836 632fc199-4ca6-4c93-a231-07263d6284db

4 files changed, 76 insertions, 121 deletions

diff --git a/simd/jdclrss2-64.asm b/simd/jdclrss2-64.asm
index 696a383..06cb213 100644
--- a/simd/jdclrss2-64.asm
+++ b/simd/jdclrss2-64.asm
@@ -251,17 +251,13 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	movntdq	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
 	movntdq	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
 	movntdq	XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
-	add	rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	jmp	short .out0
 .out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmH,xmmH			; xmmH=(all 1's)
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [rdi], xmmA
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [rdi], xmmD
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmF,xmmH			; movntdqu XMMWORD [rdi], xmmF
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+	movdqu	XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
 .out0:
+	add	rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	sub	rcx, byte SIZEOF_XMMWORD
 	jz	near .nextrow
 
@@ -275,17 +271,16 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	lea	rcx, [rcx+rcx*2]		; imul ecx, RGB_PIXELSIZE
 	cmp	rcx, byte 2*SIZEOF_XMMWORD
 	jb	short .column_st16
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [rdi], xmmA
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [rdi], xmmD
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+	add	rdi, byte 2*SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmF
 	sub	rcx, byte 2*SIZEOF_XMMWORD
 	jmp	short .column_st15
 .column_st16:
 	cmp	rcx, byte SIZEOF_XMMWORD
 	jb	short .column_st15
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [rdi], xmmA
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
 	add	rdi, byte SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmD
 	sub	rcx, byte SIZEOF_XMMWORD
@@ -363,7 +358,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	por	xmmA,xmmG
 	por	xmmE,xmmC
 .adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [rdi], xmmA
+	movdqu	XMMWORD [rdi], xmmA
 %endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %else ; RGB_PIXELSIZE == 4 ; -----------
@@ -409,19 +404,14 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	movntdq	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
 	movntdq	XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
 	movntdq	XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
-	add	rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	jmp	short .out0
 .out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmE,xmmE			; xmmE=(all 1's)
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [rdi], xmmA
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [rdi], xmmD
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmC,xmmE			; movntdqu XMMWORD [rdi], xmmC
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmH,xmmE			; movntdqu XMMWORD [rdi], xmmH
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+	movdqu	XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
+	movdqu	XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
 .out0:
+	add	rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	sub	rcx, byte SIZEOF_XMMWORD
 	jz	near .nextrow
 
@@ -434,17 +424,16 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	pcmpeqb	xmmE,xmmE			; xmmE=(all 1's)
 	cmp	rcx, byte SIZEOF_XMMWORD/2
 	jb	short .column_st16
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [rdi], xmmA
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [rdi], xmmD
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+	add	rdi, byte 2*SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmC
 	movdqa	xmmD,xmmH
 	sub	rcx, byte SIZEOF_XMMWORD/2
 .column_st16:
 	cmp	rcx, byte SIZEOF_XMMWORD/4
 	jb	short .column_st15
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [rdi], xmmA
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
 	add	rdi, byte SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmD
 	sub	rcx, byte SIZEOF_XMMWORD/4
@@ -503,7 +492,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	por	xmmA,xmmB
 	por	xmmE,xmmG
 .adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [rdi], xmmA
+	movdqu	XMMWORD [rdi], xmmA
 %endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %endif ; RGB_PIXELSIZE ; ---------------
diff --git a/simd/jdclrss2.asm b/simd/jdclrss2.asm
index 7f519e6..1354c3d 100644
--- a/simd/jdclrss2.asm
+++ b/simd/jdclrss2.asm
@@ -262,17 +262,13 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	movntdq	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
 	movntdq	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
 	movntdq	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	jmp	short .out0
 .out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmH,xmmH			; xmmH=(all 1's)
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmF,xmmH			; movntdqu XMMWORD [edi], xmmF
-	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	movdqu	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
 .out0:
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	sub	ecx, byte SIZEOF_XMMWORD
 	jz	near .nextrow
 
@@ -287,17 +283,16 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	lea	ecx, [ecx+ecx*2]		; imul ecx, RGB_PIXELSIZE
 	cmp	ecx, byte 2*SIZEOF_XMMWORD
 	jb	short .column_st16
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	add	edi, byte 2*SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmF
 	sub	ecx, byte 2*SIZEOF_XMMWORD
 	jmp	short .column_st15
 .column_st16:
 	cmp	ecx, byte SIZEOF_XMMWORD
 	jb	short .column_st15
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
 	add	edi, byte SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmD
 	sub	ecx, byte SIZEOF_XMMWORD
@@ -375,7 +370,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	por	xmmA,xmmG
 	por	xmmE,xmmC
 .adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [edi], xmmA
 %endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %else ; RGB_PIXELSIZE == 4 ; -----------
@@ -421,19 +416,14 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	movntdq	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
 	movntdq	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
 	movntdq	XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	jmp	short .out0
 .out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmE,xmmE			; xmmE=(all 1's)
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmC,xmmE			; movntdqu XMMWORD [edi], xmmC
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmH,xmmE			; movntdqu XMMWORD [edi], xmmH
-	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	movdqu	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
+	movdqu	XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
 .out0:
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	sub	ecx, byte SIZEOF_XMMWORD
 	jz	near .nextrow
 
@@ -447,17 +437,16 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	pcmpeqb	xmmE,xmmE			; xmmE=(all 1's)
 	cmp	ecx, byte SIZEOF_XMMWORD/2
 	jb	short .column_st16
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	add	edi, byte 2*SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmC
 	movdqa	xmmD,xmmH
 	sub	ecx, byte SIZEOF_XMMWORD/2
 .column_st16:
 	cmp	ecx, byte SIZEOF_XMMWORD/4
 	jb	short .column_st15
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
 	add	edi, byte SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmD
 	sub	ecx, byte SIZEOF_XMMWORD/4
@@ -516,7 +505,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
 	por	xmmA,xmmB
 	por	xmmE,xmmG
 .adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [edi], xmmA
 %endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %endif ; RGB_PIXELSIZE ; ---------------
diff --git a/simd/jdmrgss2-64.asm b/simd/jdmrgss2-64.asm
index a64a6b3..ffe0288 100644
--- a/simd/jdmrgss2-64.asm
+++ b/simd/jdmrgss2-64.asm
@@ -252,17 +252,13 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	movntdq	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
 	movntdq	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
 	movntdq	XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
-	add	rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	jmp	short .out0
 .out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmH,xmmH			; xmmH=(all 1's)
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [rdi], xmmA
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [rdi], xmmD
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmF,xmmH			; movntdqu XMMWORD [rdi], xmmF
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+	movdqu	XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF
 .out0:
+	add	rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	sub	rcx, byte SIZEOF_XMMWORD
 	jz	near .endcolumn
 
@@ -275,21 +271,19 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	jmp	near .columnloop
 
 .column_st32:
-	pcmpeqb	xmmH,xmmH			; xmmH=(all 1's)
 	lea	rcx, [rcx+rcx*2]		; imul ecx, RGB_PIXELSIZE
 	cmp	rcx, byte 2*SIZEOF_XMMWORD
 	jb	short .column_st16
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [rdi], xmmA
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [rdi], xmmD
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+	add	rdi, byte 2*SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmF
 	sub	rcx, byte 2*SIZEOF_XMMWORD
 	jmp	short .column_st15
 .column_st16:
 	cmp	rcx, byte SIZEOF_XMMWORD
 	jb	short .column_st15
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [rdi], xmmA
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
 	add	rdi, byte SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmD
 	sub	rcx, byte SIZEOF_XMMWORD
@@ -367,7 +361,7 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	por	xmmA,xmmG
 	por	xmmE,xmmC
 .adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [rdi],xmmA
 %endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %else ; RGB_PIXELSIZE == 4 ; -----------
@@ -413,19 +407,14 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	movntdq	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
 	movntdq	XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
 	movntdq	XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
-	add	rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	jmp	short .out0
 .out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmE,xmmE			; xmmE=(all 1's)
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [rdi], xmmA
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [rdi], xmmD
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmC,xmmE			; movntdqu XMMWORD [rdi], xmmC
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmH,xmmE			; movntdqu XMMWORD [rdi], xmmH
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+	movdqu	XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC
+	movdqu	XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH
 .out0:
+	add	rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	sub	rcx, byte SIZEOF_XMMWORD
 	jz	near .endcolumn
 
@@ -441,17 +430,16 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	pcmpeqb	xmmE,xmmE			; xmmE=(all 1's)
 	cmp	rcx, byte SIZEOF_XMMWORD/2
 	jb	short .column_st16
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [rdi], xmmA
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [rdi], xmmD
-	add	rdi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD
+	add	rdi, byte 2*SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmC
 	movdqa	xmmD,xmmH
 	sub	rcx, byte SIZEOF_XMMWORD/2
 .column_st16:
 	cmp	rcx, byte SIZEOF_XMMWORD/4
 	jb	short .column_st15
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA
 	add	rdi, byte SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmD
 	sub	rcx, byte SIZEOF_XMMWORD/4
@@ -510,7 +498,7 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	por	xmmA,xmmB
 	por	xmmE,xmmG
 .adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [rdi],xmmA
 %endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %endif ; RGB_PIXELSIZE ; ---------------
diff --git a/simd/jdmrgss2.asm b/simd/jdmrgss2.asm
index 04089aa..556a490 100644
--- a/simd/jdmrgss2.asm
+++ b/simd/jdmrgss2.asm
@@ -264,17 +264,13 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	movntdq	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
 	movntdq	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
 	movntdq	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	jmp	short .out0
 .out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmH,xmmH			; xmmH=(all 1's)
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmF,xmmH			; movntdqu XMMWORD [edi], xmmF
-	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	movdqu	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF
 .out0:
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	sub	ecx, byte SIZEOF_XMMWORD
 	jz	near .endcolumn
 
@@ -292,17 +288,16 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	lea	ecx, [ecx+ecx*2]		; imul ecx, RGB_PIXELSIZE
 	cmp	ecx, byte 2*SIZEOF_XMMWORD
 	jb	short .column_st16
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmH			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	add	edi, byte 2*SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmF
 	sub	ecx, byte 2*SIZEOF_XMMWORD
 	jmp	short .column_st15
 .column_st16:
 	cmp	ecx, byte SIZEOF_XMMWORD
 	jb	short .column_st15
-	maskmovdqu xmmA,xmmH			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
 	add	edi, byte SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmD
 	sub	ecx, byte SIZEOF_XMMWORD
@@ -380,7 +375,7 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	por	xmmA,xmmG
 	por	xmmE,xmmC
 .adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [edi], xmmA
 %endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %else ; RGB_PIXELSIZE == 4 ; -----------
@@ -426,19 +421,14 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	movntdq	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
 	movntdq	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
 	movntdq	XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
-	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	jmp	short .out0
 .out1:	; --(unaligned)-----------------
-	pcmpeqb    xmmE,xmmE			; xmmE=(all 1's)
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmC,xmmE			; movntdqu XMMWORD [edi], xmmC
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmH,xmmE			; movntdqu XMMWORD [edi], xmmH
-	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	movdqu	XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC
+	movdqu	XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH
 .out0:
+	add	edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD	; outptr
 	sub	ecx, byte SIZEOF_XMMWORD
 	jz	near .endcolumn
 
@@ -455,17 +445,16 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	pcmpeqb	xmmE,xmmE			; xmmE=(all 1's)
 	cmp	ecx, byte SIZEOF_XMMWORD/2
 	jb	short .column_st16
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
-	add	edi, byte SIZEOF_XMMWORD	; outptr
-	maskmovdqu xmmD,xmmE			; movntdqu XMMWORD [edi], xmmD
-	add	edi, byte SIZEOF_XMMWORD	; outptr
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
+	movdqu	XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD
+	add	edi, byte 2*SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmC
 	movdqa	xmmD,xmmH
 	sub	ecx, byte SIZEOF_XMMWORD/2
 .column_st16:
 	cmp	ecx, byte SIZEOF_XMMWORD/4
 	jb	short .column_st15
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA
 	add	edi, byte SIZEOF_XMMWORD	; outptr
 	movdqa	xmmA,xmmD
 	sub	ecx, byte SIZEOF_XMMWORD/4
@@ -524,7 +513,7 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
 	por	xmmA,xmmB
 	por	xmmE,xmmG
 .adj0:	; ----------------
-	maskmovdqu xmmA,xmmE			; movntdqu XMMWORD [edi], xmmA
+	movdqu	XMMWORD [edi], xmmA
 %endif ; STRICT_MEMORY_ACCESS ; ---------------
 
 %endif ; RGB_PIXELSIZE ; ---------------