rockbox/firmware/target/arm/ipod/video/lcd-as-video.S
Jens Arnold 120fb4f319 Speed up YUV blitting on iPod Video by ~8% (FS #8075 by Andree Buschmann).
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@15820 a1c6a512-1295-4272-9138-f99709370657
2007-11-26 23:48:23 +00:00

295 lines
12 KiB
ArmAsm

/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2007 by Andree Buschmann
*
* All files in this archive are subject to the GNU General Public License.
* See the file COPYING in the source tree root for full license agreement.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
.section .icode, "ax", %progbits
/****************************************************************************
* void lcd_write_data(const fb_data *addr,
* int pixelcount);
*
* Writes pixelcount pixels from src-pointer (lcd_framebuffer) to BCM dataport.
*/
.align 2
.global lcd_write_data
.type lcd_write_data, %function
/* r0 = addr, must be aligned */
lcd_write_data: /* r1 = pixel count, must be even */
stmfd sp!, {r4-r6}
mov r2, #0x30000000 /* LCD data port */
subs r1, r1, #16
.loop16:
ldmgeia r0!, {r3-r6}
stmgeia r2, {r3-r6}
ldmgeia r0!, {r3-r6}
stmgeia r2, {r3-r6}
subges r1, r1, #16
bge .loop16
/* no need to correct the count, we're just checking bits from now */
tst r1, #8
ldmneia r0!, {r3-r6}
stmneia r2, {r3-r6}
tst r1, #4
ldmneia r0!, {r3-r4}
stmneia r2, {r3-r4}
tst r1, #2
ldrne r3, [r0], #4
strne r3, [r2]
ldmfd sp!, {r4-r6}
bx lr
/****************************************************************************
* extern void lcd_write_yuv420_lines(unsigned char const * const src[3],
* unsigned bcmaddr
* int width,
* int stride);
*
* Conversion from Motion JPEG and MPEG Y'PbPr to RGB is:
* |R| |1.164 0.000 1.596| |Y' - 16|
* |G| = |1.164 -0.391 -0.813| |Pb - 128|
* |B| |1.164 2.018 0.000| |Pr - 128|
*
* Scaled, normalized, rounded and tweaked to yield RGB 565:
* |R| |74 0 101| |Y' - 16| >> 9
* |G| = |74 -24 -51| |Cb - 128| >> 8
* |B| |74 128 0| |Cr - 128| >> 9
*
* Converts two lines from YUV to RGB565 and writes to BCM at once. First loop
* loads Cb/Cr, calculates the chroma offset and saves them to buffer. Within
* the second loop these chroma offset are reloaded from buffer.
* Within each loop two pixels are calculated and written to BCM. Before each
* loop the desired destination address is transmitted to BCM.
*/
.align 2
.global lcd_write_yuv420_lines
.type lcd_write_yuv420_lines, %function
lcd_write_yuv420_lines:
/* r0 = src = yuv_src */
/* r1 = dst = bcmaddr */
/* r2 = width */
/* r3 = stride */
stmfd sp!, { r4-r12 } /* save non-scratch */
ldmia r0, { r10-r12 } /* r10 = yuv_src[0] = Y'_p */
/* r11 = yuv_src[1] = Cb_p */
/* r12 = yuv_src[2] = Cr_p */
add r3, r10, 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 */
add r4, r4, #19 /* plus room for 4 additional words, */
bic r4, r4, #3 /* rounded up to multiples of 4 byte */
sub sp, sp, r4 /* and allocate on stack */
stmia sp, {r1-r4} /* bcmaddr, width, &ysrc[stride], stack_alloc */
mov r7, r2 /* r7 = loop count */
add r8, sp, #16 /* chroma buffer */
mov r9, #0x30000000 /* LCD data port */
/* The following writes dest address to BCM and waits for write ready */
orr r2, r9, #0x00010000 /* r2 = BCM_WR_ADDR32 */
orr r6, r9, #0x00030000 /* r6 = BCM_CONTROL */
str r1, [r2] /* BCM_WR_ADDR32 = bcmaddr */
.busy_1:
ldrh r1, [r6] /* while (!(BCM_CONTROL & 0x2)) */
tst r1, #0x2
beq .busy_1
/* 1st loop start */
10: /* loop start */
ldrb r0, [r11], #1 /* r0 = *usrc++ = *Cb_p++ */
ldrb r1, [r12], #1 /* r1 = *vsrc++ = *Cr_p++ */
sub r0, r0, #128 /* r0 = Cb-128 */
sub r1, r1, #128 /* r1 = Cr-128 */
add r2, r1, r1, asl #1 /* r2 = Cr*51 + Cb*24 */
add r2, r2, r2, asl #4
add r2, r2, r0, asl #3
add r2, r2, r0, asl #4
add r4, r1, r1, asl #2 /* r1 = Cr*101 */
add r4, r4, r1, asl #5
add r1, r4, r1, asl #6
add r1, r1, #256 /* r1 = rv = (r1 + 256) >> 9 */
mov r1, r1, asr #9
rsb r2, r2, #128 /* r2 = guv = (-r2 + 128) >> 8 */
mov r2, r2, asr #8
add r0, r0, #2 /* r0 = bu = (Cb*128 + 256) >> 9 */
mov r0, r0, asr #2
stmia r8!, {r0-r2} /* store r0, r1 and r2 to chroma buffer */
/* 1st loop, first pixel */
ldrb r5, [r10], #1 /* r5 = *ysrc++ = *Y'_p++ */
sub r5, r5, #16 /* r5 = (Y'-16) * 74 */
add r3, r5, r5, asl #2
add r5, r3, r5, asl #5
add r6, r1, r5, asr #8 /* r6 = r = (Y >> 9) + rv */
add r3, r2, r5, asr #7 /* r3 = g = (Y >> 8) + guv */
add r4, r0, r5, asr #8 /* r4 = b = (Y >> 9) + bu */
orr r5, r6, r4 /* check if clamping is needed... */
orr r5, r5, r3, asr #1 /* ...at all */
cmp r5, #31
bls 15f /* -> no clamp */
cmp r6, #31 /* clamp r */
mvnhi r6, r6, asr #31
andhi r6, r6, #31
cmp r3, #63 /* clamp g */
mvnhi r3, r3, asr #31
andhi r3, r3, #63
cmp r4, #31 /* clamp b */
mvnhi r4, r4, asr #31
andhi r4, r4, #31
15: /* no clamp */
/* calculate pixel_1 and save to r5 for later pixel packing */
orr r4, r4, r3, lsl #5 /* pixel_1 = r<<11 | g<<5 | b */
orr r5, r4, r6, lsl #11 /* r5 = pixel_1 */
/* 1st loop, second pixel */
ldrb r4, [r10], #1 /* r4 = *ysrc++ = *Y'_p++ */
sub r4, r4, #16 /* r4 = (Y'-16) * 74 */
add r3, r4, r4, asl #2
add r4, r3, r4, asl #5
add r6, r1, r4, asr #8 /* r6 = r = (Y >> 9) + rv */
add r3, r2, r4, asr #7 /* r3 = g = (Y >> 8) + guv */
add r4, r0, r4, asr #8 /* r4 = b = (Y >> 9) + bu */
orr r0, r6, r4 /* check if clamping is needed... */
orr r0, r0, r3, asr #1 /* ...at all */
cmp r0, #31
bls 15f /* -> no clamp */
cmp r6, #31 /* clamp r */
mvnhi r6, r6, asr #31
andhi r6, r6, #31
cmp r3, #63 /* clamp g */
mvnhi r3, r3, asr #31
andhi r3, r3, #63
cmp r4, #31 /* clamp b */
mvnhi r4, r4, asr #31
andhi r4, r4, #31
15: /* no clamp */
/* calculate pixel_2 and pack with pixel_1 before writing */
orr r4, r4, r3, lsl #5 /* pixel_2 = r<<11 | g<<5 | b */
orr r4, r4, r6, lsl #11 /* r4 = pixel_2 */
orr r4, r5, r4, lsl #16 /* r4 = pixel_2<<16 | pixel_1 */
str r4, [r9] /* write packed pixels */
subs r7, r7, #2 /* check for loop end */
bgt 10b /* back to beginning */
/* 1st loop end */
/* Reload several registers for pointer rewinding for next loop */
add r8, sp, #16 /* chroma buffer */
ldmia sp, { r1, r7, r10} /* r1 = bcmaddr */
/* r7 = loop count */
/* r10 = &ysrc[stride] */
/* The following writes dest address to BCM and waits for write ready */
orr r2, r9, #0x00010000 /* r2 = BCM_WR_ADDR32 */
orr r6, r9, #0x00030000 /* r6 = BCM_CONTROL */
add r1, r1, #640 /* dst += (LCD_WIDTH*2) */
str r1, [r2] /* BCM_WR_ADDR32 = dst */
.busy_2:
ldrh r1, [r6] /* while (!(BCM_CONTROL & 0x2)) */
tst r1, #0x2
beq .busy_2
/* 2nd loop start */
20: /* loop start */
/* restore r0 (bu), r1 (rv) and r2 (guv) from chroma buffer */
ldmia r8!, {r0-r2}
/* 2nd loop, first pixel */
ldrb r5, [r10], #1 /* r5 = *ysrc++ = *Y'_p++ */
sub r5, r5, #16 /* r5 = (Y'-16) * 74 */
add r3, r5, r5, asl #2
add r5, r3, r5, asl #5
add r6, r1, r5, asr #8 /* r6 = r = (Y >> 9) + rv */
add r3, r2, r5, asr #7 /* r3 = g = (Y >> 8) + guv */
add r4, r0, r5, asr #8 /* r4 = b = (Y >> 9) + bu */
orr r5, r6, r4 /* check if clamping is needed... */
orr r5, r5, r3, asr #1 /* ...at all */
cmp r5, #31
bls 15f /* -> no clamp */
cmp r6, #31 /* clamp r */
mvnhi r6, r6, asr #31
andhi r6, r6, #31
cmp r3, #63 /* clamp g */
mvnhi r3, r3, asr #31
andhi r3, r3, #63
cmp r4, #31 /* clamp b */
mvnhi r4, r4, asr #31
andhi r4, r4, #31
15: /* no clamp */
/* calculate pixel_1 and save to r5 for later pixel packing */
orr r4, r4, r3, lsl #5 /* pixel_1 = r<<11 | g<<5 | b */
orr r5, r4, r6, lsl #11 /* r5 = pixel_1 */
/* 2nd loop, second pixel */
ldrb r4, [r10], #1 /* r4 = *ysrc++ = *Y'_p++ */
sub r4, r4, #16 /* r4 = (Y'-16) * 74 */
add r3, r4, r4, asl #2
add r4, r3, r4, asl #5
add r6, r1, r4, asr #8 /* r6 = r = (Y >> 9) + rv */
add r3, r2, r4, asr #7 /* r3 = g = (Y >> 8) + guv */
add r4, r0, r4, asr #8 /* r4 = b = (Y >> 9) + bu */
orr r0, r6, r4 /* check if clamping is needed... */
orr r0, r0, r3, asr #1 /* ...at all */
cmp r0, #31
bls 15f /* -> no clamp */
cmp r6, #31 /* clamp r */
mvnhi r6, r6, asr #31
andhi r6, r6, #31
cmp r3, #63 /* clamp g */
mvnhi r3, r3, asr #31
andhi r3, r3, #63
cmp r4, #31 /* clamp b */
mvnhi r4, r4, asr #31
andhi r4, r4, #31
15: /* no clamp */
/* calculate pixel_2 and pack with pixel_1 before writing */
orr r4, r4, r3, lsl #5 /* pixel_2 = r<<11 | g<<5 | b */
orr r4, r4, r6, lsl #11 /* r4 = pixel_2 */
orr r4, r5, r4, lsl #16 /* r4 = pixel_2<<16 | pixel_1 */
str r4, [r9] /* write packed pixels */
subs r7, r7, #2 /* check for loop end */
bgt 20b /* back to beginning */
/* 2nd loop end */
ldr r3, [sp, #12]
add sp, sp, r3 /* deallocate buffer */
ldmfd sp!, { r4-r12 } /* restore registers */
bx lr
.ltorg
.size lcd_write_yuv420_lines, .-lcd_write_yuv420_lines