len0rd/rockbox
1
0
Fork 1
mirror of https://github.com/Rockbox/rockbox.git synced 2025-12-08 20:55:17 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

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

62
firmware/target/arm/as3525/lcd-as-e200v2-fuze.S
View file

@ -51,7 +51,7 @@ lcd_write_yuv420_lines:
@ r0 = yuv_src
@ r1 = width
@ r2 = stride
stmfd sp!, { r4-r12 } @ save non-scratch
stmfd sp!, { r4-r10, lr } @ save non-scratch
ldmia r0, { r4, r5, r6 } @ r4 = yuv_src[0] = Y'_p
@ r5 = yuv_src[1] = Cb_p
@ r6 = yuv_src[2] = Cr_p
@ -76,9 +76,9 @@ lcd_write_yuv420_lines:
add r10, r10, r8, asl #3 @
add r10, r10, r8, asl #4 @
@
add r11, r9, r9, asl #2 @ r9 = Cr*101
add r11, r11, r9, asl #5 @
add r9, r11, r9, asl #6 @
add lr, r9, r9, asl #2 @ r9 = Cr*101
add lr, lr, r9, asl #5 @
add r9, lr, r9, asl #6 @
@
add r8, r8, #2 @ r8 = bu = (Cb*128 + 128) >> 8
mov r8, r8, asr #2 @
@ -88,19 +88,19 @@ lcd_write_yuv420_lines:
mov r10, r10, asr #8 @
@ compute R, G, and B
add r0, r8, r7, asr #8 @ r0 = b = (Y >> 9) + bu
add r11, r9, r7, asr #8 @ r11 = r = (Y >> 9) + rv
add lr, r9, r7, asr #8 @ lr = r = (Y >> 9) + rv
add r7, r10, r7, asr #7 @ r7 = g = (Y >> 8) + guv
@
orr r12, r0, r11 @ check if clamping is needed...
orr r12, r0, lr @ check if clamping is needed...
orr r12, r12, r7, asr #1 @ ...at all
cmp r12, #31 @
bls 15f @ no clamp @
cmp r0, #31 @ clamp b
mvnhi r0, r0, asr #31 @
andhi r0, r0, #31 @
cmp r11, #31 @ clamp r
mvnhi r11, r11, asr #31 @
andhi r11, r11, #31 @
cmp lr, #31 @ clamp r
mvnhi lr, lr, asr #31 @
andhi lr, lr, #31 @
cmp r7, #63 @ clamp g
mvnhi r7, r7, asr #31 @
andhi r7, r7, #63 @
@ -108,7 +108,7 @@ lcd_write_yuv420_lines:
@
ldrb r12, [r4, r2] @ r12 = Y' = *(Y'_p + stride)
@
orr r0, r0, r11, lsl #11 @ r0 = (r << 11) | b
orr r0, r0, lr, lsl #11 @ r0 = (r << 11) | b
orr r0, r0, r7, lsl #5 @ r0 = (r << 11) | (g << 5) | b
strh r0, [r3, #0x10] @ write pixel
1: @ busy @
@ -121,19 +121,19 @@ lcd_write_yuv420_lines:
add r7, r12, r7, asl #5 @
@ compute R, G, and B
add r0, r8, r7, asr #8 @ r0 = b = (Y >> 9) + bu
add r11, r9, r7, asr #8 @ r11 = r = (Y >> 9) + rv
add lr, r9, r7, asr #8 @ lr = r = (Y >> 9) + rv
add r7, r10, r7, asr #7 @ r7 = g = (Y >> 8) + guv
@
orr r12, r0, r11 @ check if clamping is needed...
orr r12, r0, lr @ check if clamping is needed...
orr r12, r12, r7, asr #1 @ ...at all
cmp r12, #31 @
bls 15f @ no clamp @
cmp r0, #31 @ clamp b
mvnhi r0, r0, asr #31 @
andhi r0, r0, #31 @
cmp r11, #31 @ clamp r
mvnhi r11, r11, asr #31 @
andhi r11, r11, #31 @
cmp lr, #31 @ clamp r
mvnhi lr, lr, asr #31 @
andhi lr, lr, #31 @
cmp r7, #63 @ clamp g
mvnhi r7, r7, asr #31 @
andhi r7, r7, #63 @
@ -141,7 +141,7 @@ lcd_write_yuv420_lines:
@
ldrb r12, [r4], #1 @ r12 = Y' = *(Y'_p++)
@
orr r0, r0, r11, lsl #11 @ r0 = (r << 11) | b
orr r0, r0, lr, lsl #11 @ r0 = (r << 11) | b
orr r0, r0, r7, lsl #5 @ r0 = (r << 11) | (g << 5) | b
strh r0, [r3, #0x10] @ write pixel
1: @ busy @
@ -154,19 +154,19 @@ lcd_write_yuv420_lines:
add r7, r12, r7, asl #5 @
@ compute R, G, and B
add r0, r8, r7, asr #8 @ r0 = b = (Y >> 9) + bu
add r11, r9, r7, asr #8 @ r11 = r = (Y >> 9) + rv
add lr, r9, r7, asr #8 @ lr = r = (Y >> 9) + rv
add r7, r10, r7, asr #7 @ r7 = g = (Y >> 8) + guv
@
orr r12, r0, r11 @ check if clamping is needed...
orr r12, r0, lr @ check if clamping is needed...
orr r12, r12, r7, asr #1 @ ...at all
cmp r12, #31 @
bls 15f @ no clamp @
cmp r0, #31 @ clamp b
mvnhi r0, r0, asr #31 @
andhi r0, r0, #31 @
cmp r11, #31 @ clamp r
mvnhi r11, r11, asr #31 @
andhi r11, r11, #31 @
cmp lr, #31 @ clamp r
mvnhi lr, lr, asr #31 @
andhi lr, lr, #31 @
cmp r7, #63 @ clamp g
mvnhi r7, r7, asr #31 @
andhi r7, r7, #63 @
@ -175,7 +175,7 @@ lcd_write_yuv420_lines:
ldrb r12, [r4, r2] @ r12 = Y' = *(Y'_p + stride)
@
@
orr r0, r0, r11, lsl #11 @ r0 = (r << 11) | b
orr r0, r0, lr, lsl #11 @ r0 = (r << 11) | b
orr r0, r0, r7, lsl #5 @ r0 = (r << 11) | (g << 5) | b
strh r0, [r3, #0x10] @ write pixel
1: @ busy @
@ -188,25 +188,25 @@ lcd_write_yuv420_lines:
add r7, r12, r7, asl #5 @
@ compute R, G, and B
add r0, r8, r7, asr #8 @ r0 = b = (Y >> 9) + bu
add r11, r9, r7, asr #8 @ r11 = r = (Y >> 9) + rv
add lr, r9, r7, asr #8 @ lr = r = (Y >> 9) + rv
add r7, r10, r7, asr #7 @ r7 = g = (Y >> 8) + guv
@
orr r12, r0, r11 @ check if clamping is needed...
orr r12, r0, lr @ check if clamping is needed...
orr r12, r12, r7, asr #1 @ ...at all
cmp r12, #31 @
bls 15f @ no clamp @
cmp r0, #31 @ clamp b
mvnhi r0, r0, asr #31 @
andhi r0, r0, #31 @
cmp r11, #31 @ clamp r
mvnhi r11, r11, asr #31 @
andhi r11, r11, #31 @
cmp lr, #31 @ clamp r
mvnhi lr, lr, asr #31 @
andhi lr, lr, #31 @
cmp r7, #63 @ clamp g
mvnhi r7, r7, asr #31 @
andhi r7, r7, #63 @
15: @ no clamp @
@
orr r0, r0, r11, lsl #11 @ r0 = (r << 11) | b
orr r0, r0, lr, lsl #11 @ r0 = (r << 11) | b
orr r0, r0, r7, lsl #5 @ r0 = (r << 11) | (g << 5) | b
strh r0, [r3, #0x10] @ write pixel
1: @ busy @
@ -217,7 +217,7 @@ lcd_write_yuv420_lines:
subs r1, r1, #2 @ subtract block from width
bgt 10b @ loop line @
@
ldmfd sp!, { r4-r12 } @ restore registers and return
ldmfd sp!, { r4-r10, pc } @ restore registers and return
bx lr @
.ltorg @ dump constant pool
.size lcd_write_yuv420_lines, .-lcd_write_yuv420_lines
@ -258,7 +258,7 @@ lcd_write_yuv420_lines_odither:
@ r2 = stride
@ r3 = x_screen
@ [sp] = y_screen
stmfd sp!, { r4-r12, lr } @ save non-scratch
stmfd sp!, { r4-r11, lr } @ save non-scratch
ldmia r0, { r4, r5, r6 } @ r4 = yuv_src[0] = Y'_p
@ r5 = yuv_src[1] = Cb_p
@ r6 = yuv_src[2] = Cr_p
@ -496,6 +496,6 @@ lcd_write_yuv420_lines_odither:
subs r1, r1, #2 @ subtract block from width
bgt 10b @ loop line @
@
ldmfd sp!, { r4-r12, pc } @ restore registers and return
ldmfd sp!, { r4-r11, pc } @ restore registers and return
.ltorg @ dump constant pool
.size lcd_write_yuv420_lines_odither, .-lcd_write_yuv420_lines_odither

16
firmware/target/arm/as3525/sansa-clip/lcd-as-clip.S
View file

@ -39,15 +39,15 @@
* r3/r4 - current block of phases
* r5/r6 - current block of values
* r7 - lcd data accumulator
* r8 - phase signs mask
* r12 - phase signs mask
* lr - lcd bridge address
*/
lcd_grey_data:
stmfd sp!, {r4-r8, lr}
mov r8, #0x80
orr r8, r8, r8, lsl #8
orr r8, r8, r8, lsl #16
stmfd sp!, {r4-r7, lr}
mov r12, #0x80
orr r12, r12, r12, lsl #8
orr r12, r12, r12, lsl #16
ldr lr, =GPIOA_BASE
mov r3, #(1<<5)
@ -70,7 +70,7 @@ lcd_grey_data:
orrne r7, r7, #0x2000
tst r3, #0x80000000
orrne r7, r7, #0x1000
bic r3, r3, r8
bic r3, r3, r12
add r3, r3, r5
/* set bits 3..0 */
@ -82,7 +82,7 @@ lcd_grey_data:
orrne r7, r7, #0x02
tst r4, #0x80000000
orrne r7, r7, #0x01
bic r4, r4, r8
bic r4, r4, r12
add r4, r4, r6
stmia r1!, {r3-r4}
@ -102,5 +102,5 @@ lcd_grey_data:
ands r5, r5, #(1<<10) @ wait until push fifo empties
beq 1b
ldmfd sp!, {r4-r8, pc}
ldmfd sp!, {r4-r7, pc}
.size lcd_grey_data,.-lcd_grey_data