devs/foxbox
1
0
Fork 0
forked from len0rd/rockbox
Code Pull requests Activity

ARM asm LCD and ATA driver functions: Don't save r12 as it is a scratch reg. Saves a bit of stack and execution time.

Browse source 
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@21795 a1c6a512-1295-4272-9138-f99709370657
This commit is contained in:
Jens Arnold 2009-07-11 23:43:44 +00:00
parent fa59ed6ae7
commit 47d4c4739b
12 changed files with 256 additions and 261 deletions
Download patch file Download diff file Expand all files Collapse all files

26
firmware/target/arm/ipod/video/lcd-as-video.S
View file

@ -32,22 +32,22 @@
.type lcd_write_data, %function
/* r0 = addr, must be aligned */
lcd_write_data: /* r1 = pixel count, must be even */
stmfd sp!, {r4-r5, lr}
stmfd sp!, {r4, lr}
mov lr, #0x30000000 /* LCD data port */
subs r1, r1, #16
.loop16:
ldmgeia r0!, {r2-r5}
stmgeia lr, {r2-r5}
ldmgeia r0!, {r2-r5}
stmgeia lr, {r2-r5}
ldmgeia r0!, {r2-r4, r12}
stmgeia lr, {r2-r4, r12}
ldmgeia r0!, {r2-r4, r12}
stmgeia lr, {r2-r4, r12}
subges r1, r1, #16
bge .loop16
/* no need to correct the count, we're just checking bits from now */
tst r1, #8
ldmneia r0!, {r2-r5}
stmneia lr, {r2-r5}
ldmneia r0!, {r2-r4, r12}
stmneia lr, {r2-r4, r12}
tst r1, #4
ldmneia r0!, {r2-r3}
stmneia lr, {r2-r3}
@ -55,7 +55,7 @@ lcd_write_data: /* r1 = pixel count, must be even */
ldrne r3, [r0], #4
strne r3, [lr]
ldmfd sp!, {r4-r5, pc}
ldmfd sp!, {r4, pc}
/****************************************************************************
* extern void lcd_write_yuv420_lines(unsigned char const * const src[3],
@ -87,10 +87,10 @@ lcd_write_yuv420_lines:
/* r1 = dst = bcmaddr */
/* r2 = width */
/* r3 = stride */
stmfd sp!, { r4-r11, lr } /* save non-scratch */
ldmia r0, { r9-r11 } /* r9 = yuv_src[0] = Y'_p */
stmfd sp!, { r4-r10, lr } /* save non-scratch */
ldmia r0, { r9, r10, r12 } /* r9 = yuv_src[0] = Y'_p */
/* r10 = yuv_src[1] = Cb_p */
/* r11 = yuv_src[2] = Cr_p */
/* r12 = yuv_src[2] = Cr_p */
add r3, r9, r3 /* r3 = &ysrc[stride] */
add r4, r2, r2, asr #1 /* chroma buffer lenght = width/2 *3 */
mov r4, r4, asl #2 /* use words for str/ldm possibility */
@ -116,7 +116,7 @@ lcd_write_yuv420_lines:
10: /* loop start */
ldrb r0, [r10], #1 /* r0 = *usrc++ = *Cb_p++ */
ldrb r1, [r11], #1 /* r1 = *vsrc++ = *Cr_p++ */
ldrb r1, [r12], #1 /* r1 = *vsrc++ = *Cr_p++ */
sub r0, r0, #128 /* r0 = Cb-128 */
sub r1, r1, #128 /* r1 = Cr-128 */
@ -289,7 +289,7 @@ lcd_write_yuv420_lines:
ldr r3, [sp, #12]
add sp, sp, r3 /* deallocate buffer */
ldmfd sp!, { r4-r11, pc } /* restore registers */
ldmfd sp!, { r4-r10, pc } /* restore registers */
.ltorg
.size lcd_write_yuv420_lines, .-lcd_write_yuv420_lines