Speed up of iPod nano 1G and iPod color LCD. Use HDD6330 asm part for YUV blitting, introduce special handling for full width screen updates. Speed up is about +30% for YUV on both color/nano1G.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@28930 a1c6a512-1295-4272-9138-f99709370657

2 files changed, 182 insertions, 102 deletions

diff --git a/firmware/target/arm/ipod/lcd-as-color-nano.S b/firmware/target/arm/ipod/lcd-as-color-nano.S
new file mode 100755
index 0000000000..d4df4d496a
--- /dev/null
+++ b/firmware/target/arm/ipod/lcd-as-color-nano.S
@@ -0,0 +1,152 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id:$
+ *
+ * Copyright (C) 2010 by Andree Buschmann
+ *
+ * Generic asm helper function used by YUV blitting.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
+ * KIND, either express or implied.
+ *
+ ****************************************************************************/
+
+#include "config.h"
+#include "cpu.h"
+
+    .section    .icode, "ax", %progbits
+
+/****************************************************************************
+*  void lcd_yuv_write_inner_loop(unsigned char const * const ysrc,
+*                                unsigned char const * const usrc,
+*                                unsigned char const * const vsrc,
+*                                int width);
+*
+*   YUV- > RGB565 conversion
+*   |R|   |1.000000 -0.000001  1.402000| |Y'|
+*   |G| = |1.000000 -0.334136 -0.714136| |Pb|
+*   |B|   |1.000000  1.772000  0.000000| |Pr|
+*   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
+*
+*/
+    .align      2
+    .global     lcd_yuv_write_inner_loop
+    .type       lcd_yuv_write_inner_loop, %function
+
+lcd_yuv_write_inner_loop:
+                                        @ r0 = ysrc
+                                        @ r1 = usrc
+                                        @ r2 = vsrc
+                                        @ r3 = width
+    stmfd sp!, { r4-r11, lr }           @ save regs
+    mov r4, #0x70000000                 @ r4 = LCD2_BLOCK_CTRL - 0x20
+    add r4, r4, #0x8a00                 @
+    add r5, r4, #0x100                  @ r5 = LCD2_BLOCK_DATA
+10:                                     @ loop
+
+    ldrb r7, [r1], #1                   @ *usrc++
+    ldrb r8, [r2], #1                   @ *vsrc++
+
+    sub r7, r7, #128                    @ Cb -= 128
+    sub r8, r8, #128                    @ Cr -= 128
+
+    add r10, r8, r8, asl #2             @ Cr*101
+    add r10, r10, r8, asl #5
+    add r10, r10, r8, asl #6
+
+    add r11, r8, r8, asl #1             @ Cr*51 + Cb*24
+    add r11, r11, r11, asl #4
+    add r11, r11, r7, asl #3
+    add r11, r11, r7, asl #4
+
+    add r12, r7, #2                     @ r12 = bu = (Cb*128 + 256) >> 9
+    mov r12, r12, asr #2
+    add r10, r10, #256                  @ r10 = rv = (Cr*101 + 256) >> 9
+    mov r10, r10, asr #9
+    rsb r11, r11, #128                  @ r11 = guv = (-r11 + 128) >> 8
+    mov r11, r11, asr #8
+
+@ pixel_1
+    ldrb r7, [r0], #1                   @ *ysrc++
+    sub r7, r7, #16                     @ Y = (Y' - 16) * 37
+    add r8, r7, r7, asl #2
+    add r7, r8, r7, asl #5
+
+    add r9, r10, r7, asr #8             @ R = (Y >> 8) + rv
+    add r8, r11, r7, asr #7             @ G = (Y >> 7) + guv
+    add r7, r12, r7, asr #8             @ B = (Y >> 8) + bu
+
+    cmp r9, #31                         @ clamp R
+    mvnhi r9, r9, asr #31
+    andhi r9, r9, #31
+
+    cmp r8, #63                         @ clamp G
+    mvnhi r8, r8, asr #31
+    andhi r8, r8, #63
+
+    cmp r7, #31                         @ clamp B
+    mvnhi r7, r7, asr #31
+    andhi r7, r7, #31
+
+    orr r6, r7, r8, lsl #5              @ pack pixel
+    orr r6, r6, r9, lsl #11
+
+    mov r7, r6, lsl #8                  @ swap bytes
+    and r7, r7, #0xff00
+    add r6, r7, r6, lsr #8
+
+@ pixel_2
+    ldrb r7, [r0], #1                   @ *ysrc++
+    sub r7, r7, #16                     @ Y = (Y' - 16) * 37
+    add r8, r7, r7, asl #2
+    add r7, r8, r7, asl #5
+
+    add r9, r10, r7, asr #8             @ R = (Y >> 8) + rv
+    add r8, r11, r7, asr #7             @ G = (Y >> 7) + guv
+    add r7, r12, r7, asr #8             @ B = (Y >> 8) + bu
+
+    cmp r9, #31                         @ clamp R
+    mvnhi r9, r9, asr #31
+    andhi r9, r9, #31
+
+    cmp r8, #63                         @ clamp G
+    mvnhi r8, r8, asr #31
+    andhi r8, r8, #63
+
+    cmp r7, #31                         @ clamp B
+    mvnhi r7, r7, asr #31
+    andhi r7, r7, #31
+
+    orr r7, r7, r8, lsl #5              @ pack pixel
+    orr r7, r7, r9, lsl #11
+
+    orr r6, r6, r7, lsl #24             @ swap bytes and add pixels simultaneously
+    mov r7, r7, lsr #8
+    orr r6, r6, r7, lsl #16
+#if 1
+11:                                     @ while (!(LCD2_BLOCK_CTRL & LCD2_BLOCK_TXOK));
+    ldr r11, [r4, #0x20]                @
+    tst r11, #0x1000000                 @
+    beq 11b                             @
+#endif
+    str r6, [r5]                        @ send two pixels
+
+    subs r3, r3, #2                     @ decrease width
+    bgt 10b                             @ loop
+
+    ldmpc regs=r4-r11                   @ restore regs
+    .ltorg                              @ dump constant pool
+    .size   lcd_yuv_write_inner_loop, .-lcd_yuv_write_inner_loop
diff --git a/firmware/target/arm/ipod/lcd-color_nano.c b/firmware/target/arm/ipod/lcd-color_nano.c
index 7d004cb0f2..e3b9ea8eb6 100644
--- a/firmware/target/arm/ipod/lcd-color_nano.c
+++ b/firmware/target/arm/ipod/lcd-color_nano.c
@@ -121,38 +121,14 @@ void lcd_init_device(void)
 }
 
 /*** update functions ***/
+extern void lcd_yuv_write_inner_loop(unsigned char const * const ysrc,
+                                     unsigned char const * const usrc,
+                                     unsigned char const * const vsrc,
+                                     int width);
 
 #define CSUB_X 2
 #define CSUB_Y 2
 
-/*   YUV- > RGB565 conversion
- *   |R|   |1.000000 -0.000001  1.402000| |Y'|
- *   |G| = |1.000000 -0.334136 -0.714136| |Pb|
- *   |B|   |1.000000  1.772000  0.000000| |Pr|
- *   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
-*/
-
-#define RGBYFAC   74   /*  1.0      */
-#define RVFAC    101   /*  1.402    */
-#define GVFAC   (-51)  /* -0.714136 */
-#define GUFAC   (-24)  /* -0.334136 */
-#define BUFAC    128   /*  1.772    */
-
-/* ROUNDOFFS contain constant for correct round-offs as well as
-   constant parts of the conversion matrix (e.g. (Y'-16)*RGBYFAC
-   -> constant part = -16*RGBYFAC). Through extraction of these
-   constant parts we save at leat 4 substractions in the conversion
-   loop */
-#define ROUNDOFFSR (256 - 16*RGBYFAC - 128*RVFAC)
-#define ROUNDOFFSG (128 - 16*RGBYFAC - 128*GVFAC - 128*GUFAC)
-#define ROUNDOFFSB (256 - 16*RGBYFAC             - 128*BUFAC)
-
-#define MAX_5BIT 0x1f
-#define MAX_6BIT 0x3f
-
 /* Performance function to blit a YUV bitmap directly to the LCD */
 void lcd_blit_yuv(unsigned char * const src[3],
                   int src_x, int src_y, int stride,
@@ -222,7 +198,8 @@ void lcd_blit_yuv(unsigned char * const src[3],
     const int stride_div_csub_x = stride/CSUB_X;
 
     h=0;
-    while (1) {
+    while (1)
+    {
         /* upsampling, YUV->RGB conversion and reduction to RGB565 in one go */
         const unsigned char *ysrc = src[0] + stride * src_y + src_x;
 
@@ -231,17 +208,11 @@ void lcd_blit_yuv(unsigned char * const src[3],
 
         const unsigned char *usrc = src[1] + uvoffset;
         const unsigned char *vsrc = src[2] + uvoffset;
-        const unsigned char *row_end = ysrc + width;
-
-        int yp, up, vp;
-        int red1, green1, blue1;
-        int red2, green2, blue2;
 
-        int rc, gc, bc;
         int pixels_to_write;
-        fb_data pixel1,pixel2;
 
-        if (h==0) {
+        if (h==0)
+        {
             while (!(LCD2_BLOCK_CTRL & LCD2_BLOCK_READY));
             LCD2_BLOCK_CONFIG = 0;
 
@@ -251,7 +222,8 @@ void lcd_blit_yuv(unsigned char * const src[3],
             h = height;
 
             /* calculate how much we can do in one go */
-            if (pixels_to_write > 0x10000) {
+            if (pixels_to_write > 0x10000)
+            {
                 h = (0x10000/2) / width;
                 pixels_to_write = (width * h) * 2;
             }
@@ -262,61 +234,7 @@ void lcd_blit_yuv(unsigned char * const src[3],
             LCD2_BLOCK_CTRL = 0x34000000;
         }
 
-        do
-        {
-            up = *usrc++;
-            vp = *vsrc++;
-            rc = RVFAC * vp              + ROUNDOFFSR;
-            gc = GVFAC * vp + GUFAC * up + ROUNDOFFSG;
-            bc =              BUFAC * up + ROUNDOFFSB;
-            
-            /* Pixel 1 -> RGB565 */
-            yp = *ysrc++ * RGBYFAC;
-            red1   = (yp + rc) >> 9;
-            green1 = (yp + gc) >> 8;
-            blue1  = (yp + bc) >> 9;
-
-            /* Pixel 2 -> RGB565 */
-            yp = *ysrc++ * RGBYFAC;
-            red2   = (yp + rc) >> 9;
-            green2 = (yp + gc) >> 8;
-            blue2  = (yp + bc) >> 9;
-
-            /* Since out of bounds errors are relatively rare, we check two
-               pixels at once to see if any components are out of bounds, and
-               then fix whichever is broken. This works due to high values and
-               negative values both being !=0 when bitmasking them.
-               We first check for red and blue components (5bit range). */
-            if ((red1 | blue1 | red2 | blue2) & ~MAX_5BIT)
-            {
-                if (red1  & ~MAX_5BIT)
-                    red1  = (red1  >> 31) ? 0 : MAX_5BIT;
-                if (blue1 & ~MAX_5BIT)
-                    blue1 = (blue1 >> 31) ? 0 : MAX_5BIT;
-                if (red2  & ~MAX_5BIT)
-                    red2  = (red2  >> 31) ? 0 : MAX_5BIT;
-                if (blue2 & ~MAX_5BIT)
-                    blue2 = (blue2 >> 31) ? 0 : MAX_5BIT;
-            }
-            /* We second check for green component (6bit range) */
-            if ((green1 | green2) & ~MAX_6BIT)
-            {
-                if (green1 & ~MAX_6BIT)
-                    green1 = (green1 >> 31) ? 0 : MAX_6BIT;
-                if (green2 & ~MAX_6BIT)
-                    green2 = (green2 >> 31) ? 0 : MAX_6BIT;
-            }
-
-            pixel1 = swap16((red1 << 11) | (green1 << 5) | blue1);
-
-            pixel2 = swap16((red2 << 11) | (green2 << 5) | blue2);
-
-            while (!(LCD2_BLOCK_CTRL & LCD2_BLOCK_TXOK));
-
-            /* output 2 pixels */
-            LCD2_BLOCK_DATA = (pixel2 << 16) | pixel1;
-        }
-        while (ysrc < row_end);
+        lcd_yuv_write_inner_loop(ysrc,usrc,vsrc,width);
 
         src_y++;
         h--;
@@ -415,16 +333,26 @@ void lcd_update_rect(int x, int y, int width, int height)
         LCD2_BLOCK_CONFIG = 0xc0010000 | (pixels_to_write - 1);
         LCD2_BLOCK_CTRL = 0x34000000;
 
-        /* for each row */
-        for (r = 0; r < h; r++) {
-            /* for each column */
-            for (c = 0; c < width; c += 2) {
-                while (!(LCD2_BLOCK_CTRL & LCD2_BLOCK_TXOK));
-
-                /* output 2 pixels */
-                LCD2_BLOCK_DATA = *addr++;
+        if (LCD_WIDTH == width) {
+            /* for each row and column in a single loop */
+            for (r = 0; r < h*width; r += 2) {
+                    while (!(LCD2_BLOCK_CTRL & LCD2_BLOCK_TXOK));
+    
+                    /* output 2 pixels */
+                    LCD2_BLOCK_DATA = *addr++;
+            }
+        } else {
+            /* for each row */
+            for (r = 0; r < h; r++) {
+                /* for each column */
+                for (c = 0; c < width; c += 2) {
+                    while (!(LCD2_BLOCK_CTRL & LCD2_BLOCK_TXOK));
+    
+                    /* output 2 pixels */
+                    LCD2_BLOCK_DATA = *addr++;
+                }
+                addr += (LCD_WIDTH - width)/2;
             }
-            addr += (LCD_WIDTH - width)/2;
         }
 
         while (!(LCD2_BLOCK_CTRL & LCD2_BLOCK_READY));