I noticed that there are notes about removing data segment variables and
moving them into text-segment (in the form of push/pop). I don't think that
is faster than directly loading the values. Has anyone checked this out?
Patrick Baggett
On Nov 8, 2007 5:12 PM, Benno Schulenberg <[EMAIL PROTECTED]> wrote:
> Signed-off-by: Benno Schulenberg <[EMAIL PROTECTED]>
> ---
> src/mesa/x86/read_rgba_span_x86.S | 117
> ++----------------------------------
> 1 files changed, 7 insertions(+), 110 deletions(-)
>
> diff --git a/src/mesa/x86/read_rgba_span_x86.S
> b/src/mesa/x86/read_rgba_span_x86.S
> index 8509df6..2e5c3be 100644
> --- a/src/mesa/x86/read_rgba_span_x86.S
> +++ b/src/mesa/x86/read_rgba_span_x86.S
> @@ -33,20 +33,7 @@
> .file "read_rgba_span_x86.S"
> #if !defined(__DJGPP__) && !defined(__MINGW32__) /* this one cries for
> assyntax.h */
> /* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - .section .rodata
> - .align 16
> - .type mask, @object
> - .size mask, 32
> -mask:
> - .long 0xff00ff00
> - .long 0xff00ff00
> - .long 0xff00ff00
> - .long 0xff00ff00
> - .long 0x00ff0000
> - .long 0x00ff0000
> - .long 0x00ff0000
> - .long 0x00ff0000
> + * Replaced data segment constants with text-segment instructions.
> */
> #define LOAD_MASK(mvins,m1,m2) \
> pushl $0xff00ff00 ;\
> @@ -61,8 +48,7 @@ mask:
> mvins (%esp), m2 ;\
> addl $32, %esp
>
> -
> -/* I implemented these as macros because the appear in quite a few
> places,
> +/* I implemented these as macros because they appear in several places,
> * and I've tweaked them a number of times. I got tired of changing every
> * place they appear. :)
> */
> @@ -99,11 +85,6 @@ _generic_read_RGBA_span_BGRA8888_REV_MMX:
> #ifdef USE_INNER_EMMS
> emms
> #endif
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - movq mask, %mm1
> - movq mask+16, %mm2
> - */
> LOAD_MASK(movq,%mm1,%mm2)
>
> movl 8(%esp), %ebx /* source pointer */
> @@ -201,11 +182,7 @@ _generic_read_RGBA_span_BGRA8888_REV_SSE:
> #ifdef USE_INNER_EMMS
> emms
> #endif
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - movq mask, %mm1
> - movq mask+16, %mm2
> - */
> +
> LOAD_MASK(movq,%mm1,%mm2)
>
> movl 16(%esp), %ebx /* source pointer */
> @@ -364,11 +341,6 @@ _generic_read_RGBA_span_BGRA8888_REV_SSE2:
> pushl %esi
> pushl %ebx
>
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - movdqa mask, %xmm1
> - movdqa mask+16, %xmm2
> - */
> LOAD_MASK(movdqu,%xmm1,%xmm2)
>
> movl 12(%esp), %ebx /* source pointer */
> @@ -491,60 +463,12 @@ _generic_read_RGBA_span_BGRA8888_REV_SSE2:
>
>
>
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - */
> -#if 0
> - .section .rodata
> -
> - .align 16
> -mask_565:
> - .word 0xf800
> - .word 0x07e0
> - .word 0x001f
> - .word 0x0000
> -
> -/* Setting SCALE_ADJUST to 5 gives a perfect match with the classic C
> - * implementation in Mesa. Setting SCALE_ADJUST to 0 is slightly faster
> but
> - * at a small cost to accuracy.
> - */
> -
> -#define SCALE_ADJUST 5
> -#if SCALE_ADJUST == 5
> -prescale:
> - .word 0x0001
> - .word 0x0010
> - .word 0x0200
> - .word 0x0000
> -
> -scale:
> - .word 0x20e8 /* (0x00ff0000 / 0x000007c0) + 1 */
> - .word 0x40c5 /* (0x00ff0000 / 0x000003f0) + 1 */
> - .word 0x839d /* (0x00ff0000 / 0x000001f0) + 1 */
> - .word 0x0000
> -#elif SCALE_ADJUST == 0
> -prescale:
> - .word 0x0001
> - .word 0x0020
> - .word 0x0800
> - .word 0x0000
> -
> -scale:
> - .word 0x0108 /* (0x00ff0000 / 0x0000f800) + 1 */
> - .word 0x0104 /* (0x00ff0000 / 0x0000fc00) + 1 */
> - .word 0x0108 /* (0x00ff0000 / 0x0000f800) + 1 */
> - .word 0x0000
> -#else
> -#error SCALE_ADJUST must either be 5 or 0.
> -#endif
> -
> -
> -alpha: .long 0x00000000
> - .long 0x00ff0000
> -#endif
> -
> #define MASK_565_L 0x07e0f800
> #define MASK_565_H 0x0000001f
> +/* Setting SCALE_ADJUST to 5 gives a perfect match with the
> + * classic C implementation in Mesa. Setting SCALE_ADJUST
> + * to 0 is slightly faster but at a small cost to accuracy.
> + */
> #define SCALE_ADJUST 5
> #if SCALE_ADJUST == 5
> #define PRESCALE_L 0x00100001
> @@ -581,12 +505,6 @@ _generic_read_RGBA_span_RGB565_MMX:
> movl 8(%esp), %edx /* destination pointer */
> movl 12(%esp), %ecx /* number of pixels to copy */
>
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - movq mask_565, %mm5
> - movq prescale, %mm6
> - movq scale, %mm7
> - */
> pushl $MASK_565_H
> pushl $MASK_565_L
> movq (%esp), %mm5
> @@ -648,11 +566,6 @@ _generic_read_RGBA_span_RGB565_MMX:
> /* Always set the alpha value to 0xff.
> */
>
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - por alpha, %mm0
> - por alpha, %mm2
> - */
> por %mm3, %mm0
> por %mm3, %mm2
>
> @@ -665,8 +578,6 @@ _generic_read_RGBA_span_RGB565_MMX:
> movq %mm0, (%edx)
> addl $8, %edx
>
> -
> -
> pshufw $0xaa, %mm4, %mm0
> pshufw $0xff, %mm4, %mm2
>
> @@ -681,11 +592,6 @@ _generic_read_RGBA_span_RGB565_MMX:
> pmulhuw %mm7, %mm0
> pmulhuw %mm7, %mm2
>
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - por alpha, %mm0
> - por alpha, %mm2
> - */
> por %mm3, %mm0
> por %mm3, %mm2
>
> @@ -724,11 +630,6 @@ _generic_read_RGBA_span_RGB565_MMX:
> pmulhuw %mm7, %mm0
> pmulhuw %mm7, %mm2
>
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - por alpha, %mm0
> - por alpha, %mm2
> - */
> por %mm3, %mm0
> por %mm3, %mm2
>
> @@ -757,10 +658,6 @@ _generic_read_RGBA_span_RGB565_MMX:
> #endif
> pmulhuw %mm7, %mm0
>
> -/* Kevin F. Quinn 2nd July 2006
> - * Replace data segment constants with text-segment instructions
> - por alpha, %mm0
> - */
> por %mm3, %mm0
>
> packuswb %mm0, %mm0
> --
> 1.5.2.5
>
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc.
> Still grepping through log files to find problems? Stop.
> Now Search log events and configuration files using AJAX and a browser.
> Download your FREE copy of Splunk now >> http://get.splunk.com/
> _______________________________________________
> Mesa3d-dev mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
>
-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems? Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/
_______________________________________________
Mesa3d-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
