Gigabeat: Add an assembly lcd_yuv_blit. Seems to be good for a modest speedup. ;)

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12998 a1c6a512-1295-4272-9138-f99709370657

3 files changed, 255 insertions, 94 deletions

diff --git a/firmware/SOURCES b/firmware/SOURCES
index 9ec97c191e..8b1d3cf896 100644
--- a/firmware/SOURCES
+++ b/firmware/SOURCES
@@ -494,6 +494,7 @@ target/arm/gigabeat/meg-fx/ata-meg-fx.c
 target/arm/gigabeat/meg-fx/backlight-meg-fx.c
 target/arm/gigabeat/meg-fx/button-meg-fx.c
 target/arm/gigabeat/meg-fx/i2c-meg-fx.c
+target/arm/gigabeat/meg-fx/lcd-as-meg-fx.S
 target/arm/gigabeat/meg-fx/lcd-meg-fx.c
 target/arm/gigabeat/meg-fx/power-meg-fx.c
 target/arm/gigabeat/meg-fx/sc606-meg-fx.c
diff --git a/firmware/target/arm/gigabeat/meg-fx/lcd-as-meg-fx.S b/firmware/target/arm/gigabeat/meg-fx/lcd-as-meg-fx.S
new file mode 100644
index 0000000000..414f26d3c5
--- /dev/null
+++ b/firmware/target/arm/gigabeat/meg-fx/lcd-as-meg-fx.S
@@ -0,0 +1,222 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id:$
+ *
+ * Copyright (C) 2007 by Michael Sevakis
+ *
+ * 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.
+ *
+ ****************************************************************************/
+
+#include "config.h"
+#include "cpu.h"
+
+/****************************************************************************
+ * void lcd_write_yuv_420_lines(fb_data *dst,
+ *                              unsigned char chroma_buf[3][LCD_HEIGHT/2],
+                                unsigned char const * const src[3],
+ *                              int width,
+ *                              int stride);
+ *
+ *   |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
+ */
+    .section    .icode, "ax", %progbits
+    .align      2
+    .global     lcd_write_yuv420_lines
+    .type       lcd_write_yuv420_lines, %function
+lcd_write_yuv420_lines:
+                                        @ r0   = dst
+                                        @ r1   = chroma_buf
+                                        @ r2   = yuv_src
+                                        @ r3   = width
+                                        @ [sp] = stride
+    stmdb       sp!, { r4-r12, lr }     @ save non-scratch
+    stmdb       sp!, { r0, r3 }         @ save dst and width
+    mov         r14, #74                @ r14 = Y factor
+    ldmia       r2, { r4, r5, r6 }      @ r4 = yuv_src[0] = Y'_p
+                                        @ r5 = yuv_src[1] = Cb_p
+                                        @ r6 = yuv_src[2] = Cr_p
+10: @ loop line 1                       @
+    ldrb        r2, [r4], #1            @ r2  = *Y'_p++;
+    ldrb        r8, [r5], #1            @ r8  = *Cb_p++;
+    ldrb        r11, [r6], #1           @ r11 = *Cr_p++;
+                                        @
+                                        @ compute Y
+    sub         r2, r2, #16             @ r7 = Y = (Y' - 16)*74
+    mul         r7, r2, r14             @
+                                        @
+    sub         r8, r8, #128            @ Cb -= 128
+    sub         r11, r11, #128          @ Cr -= 128
+                                        @
+    mvn         r2, #24                 @ compute guv
+    mul         r10, r2, r8             @ r10 = Cb*-24
+    mvn         r2, #51                 @
+    mla         r10, r2, r11, r10       @ r10 = r10 + Cr*-51
+                                        @
+    mov         r2, #101                @ compute rv
+    mul         r9, r11, r2             @ r9 = rv = Cr*101
+                                        @
+                                        @ store chromas in line buffer
+    add         r8, r8, #2              @ bu = (Cb + 2) >> 2
+    mov         r8, r8, asr #2          @ 
+    strb        r8, [r1], #1            @
+    add         r9, r9, #256            @ rv = (Cr + 256) >> 9
+    mov         r9, r9, asr #9          @
+    strb        r9, [r1], #1            @
+    mov         r10, r10, asr #8        @ guv >>= 8
+    strb        r10, [r1], #1           @
+                                        @ compute R, G, and B
+    add         r2, r8, r7, asr #9      @ r2  = b = (Y >> 9) + bu
+    add         r11, r9, r7, asr #9     @ r11 = r = (Y >> 9) + rv
+    add         r7, r10, r7, asr #8     @ r7  = g = (Y >> 8) + guv
+                                        @
+    orr         r12, r2, r11            @ check if clamping is needed...
+    orr         r12, r12, r7, asr #1    @ ...at all
+    cmp         r12, #31                @
+    bls         15f @ no clamp          @
+    mov         r12, #31                @
+    cmp         r12, r2                 @ clamp b
+    andlo       r2, r12, r2, asr #31    @
+    eorlo       r2, r2, r12             @
+    cmp         r12, r11                @ clamp r
+    andlo       r11, r12, r11, asr #31  @
+    eorlo       r11, r11, r12           @
+    cmp         r12, r7, asr #1         @ clamp g
+    andlo       r7, r12, r7, asr #31    @
+    eorlo       r7, r7, r12             @
+    orrlo       r7, r7, r7, asl #1      @
+15: @ no clamp                          @
+                                        @
+    orr         r12, r2, r7, lsl #5     @ r4 |= (g << 5)
+    ldrb        r2, [r4], #1            @ r2 = Y' = *Y'_p++
+    orr         r12, r12, r11, lsl #11  @ r4 = b | (r << 11)
+    strh        r12, [r0], #240         @ store pixel
+                                        @
+    sub         r2, r2, #16             @ r7 = Y = (Y' - 16)*74
+    mul         r7, r2, r14             @ next Y
+                                        @ compute R, G, and B
+    add         r2, r8, r7, asr #9      @ r2  = b = (Y >> 9) + bu
+    add         r11, r9, r7, asr #9     @ r11 = r = (Y >> 9) + rv
+    add         r7, r10, r7, asr #8     @ r7  = g = (Y >> 8) + guv
+                                        @
+    orr         r12, r2, r11            @ check if clamping is needed...
+    orr         r12, r12, r7, asr #1    @ ...at all
+    cmp         r12, #31                @
+    bls         15f @ no clamp          @
+    mov         r12, #31                @
+    cmp         r12, r2                 @ clamp b
+    andlo       r2, r12, r2, asr #31    @
+    eorlo       r2, r2, r12             @
+    cmp         r12, r11                @ clamp r
+    andlo       r11, r12, r11, asr #31  @
+    eorlo       r11, r11, r12           @
+    cmp         r12, r7, asr #1         @ clamp g
+    andlo       r7, r12, r7, asr #31    @
+    eorlo       r7, r7, r12             @
+    orrlo       r7, r7, r7, asl #1      @
+15: @ no clamp                          @
+                                        @
+    orr         r12, r2, r11, lsl #11   @ r4 = b | (r << 11)
+    orr         r12, r12, r7, lsl #5    @ r4 |= (g << 5)
+    strh        r12, [r0, #240]!        @ store pixel
+    add         r0, r0, #2*240          @
+                                        @
+    subs        r3, r3, #2              @
+    bgt         10b @ loop line 1       @
+                                        @ do second line
+                                        @
+    ldmia       sp!, { r0, r3 }         @ pop dst and width
+    sub         r0, r0, #2              @ set dst to start of next line                                        @
+    sub         r1, r1, r3, asl #1      @ rewind chroma pointer...
+    ldr         r2, [sp, #40]           @ r2 = stride
+    add         r1, r1, r3, asr #1      @ ... (r1 -= width/2*3)
+                                        @ move sources to start of next line
+    sub         r2, r2, r3              @ r2 = skip = stride - width
+    add         r4, r4, r2              @ r4 = Y'_p + skip
+                                        @
+20: @ loop line 2                       @
+    ldrb        r2, [r4], #1            @ r7 = Y' = *Y'_p++
+    ldrsb       r8, [r1], #1            @ reload saved chromas
+    ldrsb       r9, [r1], #1            @
+    ldrsb       r10, [r1], #1           @
+                                        @
+    sub         r2, r2, #16             @ r2 = Y = (Y' - 16)*74
+    mul         r7, r2, r14             @
+                                        @ compute R, G, and B
+    add         r2, r8, r7, asr #9      @ r2  = b = (Y >> 9) + bu
+    add         r11, r9, r7, asr #9     @ r11 = r = (Y >> 9) + rv
+    add         r7, r10, r7, asr #8     @ r7  = g = (Y >> 8) + guv
+                                        @
+    orr         r12, r2, r11            @ check if clamping is needed...
+    orr         r12, r12, r7, asr #1    @ ...at all
+    cmp         r12, #31                @
+    bls         25f @ no clamp          @
+    mov         r12, #31                @
+    cmp         r12, r2                 @ clamp b
+    andlo       r2, r12, r2, asr #31    @
+    eorlo       r2, r2, r12             @
+    cmp         r12, r11                @ clamp r
+    andlo       r11, r12, r11, asr #31  @
+    eorlo       r11, r11, r12           @
+    cmp         r12, r7, asr #1         @ clamp g
+    andlo       r7, r12, r7, asr #31    @
+    eorlo       r7, r7, r12             @
+    orrlo       r7, r7, r7, asl #1      @
+25: @ no clamp                          @
+                                        @
+    orr         r12, r2, r11, lsl #11   @ r4 = b | (r << 11)
+    ldrb        r2, [r4], #1            @ r2 = Y' = *Y'_p++
+    orr         r12, r12, r7, lsl #5    @ r4 |= (g << 5)
+    strh        r12, [r0], #240         @ store pixel
+                                        @
+                                        @ do second pixel
+                                        @
+    sub         r2, r2, #16             @ r2 = Y = (Y' - 16)*74
+    mul         r7, r2, r14             @
+                                        @ compute R, G, and B
+    add         r2, r8, r7, asr #9      @ r2  = b = (Y >> 9) + bu
+    add         r11, r9, r7, asr #9     @ r11 = r = (Y >> 9) + rv
+    add         r7, r10, r7, asr #8     @ r7  = g = (Y >> 8) + guv
+                                        @
+    orr         r12, r2, r11            @ check if clamping is needed...
+    orr         r12, r12, r7, asr #1    @ ...at all
+    cmp         r12, #31                @
+    bls         25f @ no clamp          @
+    mov         r12, #31                @
+    cmp         r12, r2                 @ clamp b
+    andlo       r2, r12, r2, asr #31    @
+    eorlo       r2, r2, r12             @
+    cmp         r12, r11                @ clamp r
+    andlo       r11, r12, r11, asr #31  @
+    eorlo       r11, r11, r12           @
+    cmp         r12, r7, asr #1         @ clamp g
+    andlo       r7, r12, r7, asr #31    @
+    eorlo       r7, r7, r12             @
+    orrlo       r7, r7, r7, asl #1      @
+25: @ no clamp                          @
+                                        @
+    orr         r12, r2, r11, lsl #11   @ r4 = b | (r << 11)
+    orr         r12, r12, r7, lsl #5    @ r4 |= (g << 5)
+    strh        r12, [r0, #240]!        @ store pixel
+    add         r0, r0, #2*240          @
+                                        @
+    subs        r3, r3, #2              @
+    bgt         20b @ loop line 2       @
+                                        @
+    ldmia       sp!, { r4-r12, pc }     @ restore registers and return
+    .size   lcd_write_yuv420_lines, .-lcd_write_yuv420_lines
diff --git a/firmware/target/arm/gigabeat/meg-fx/lcd-meg-fx.c b/firmware/target/arm/gigabeat/meg-fx/lcd-meg-fx.c
index 2d2f32069a..1bb68f9686 100644
--- a/firmware/target/arm/gigabeat/meg-fx/lcd-meg-fx.c
+++ b/firmware/target/arm/gigabeat/meg-fx/lcd-meg-fx.c
@@ -293,22 +293,12 @@ void lcd_bitmap_transparent_part(const fb_data *src, int src_x, int src_y,
         : : "r" (src), "r" (dst), "r" (width), "r" (dst_end), "r" (stride), "r" (transcolor) : "r0", "r1" );
 }
 
-#define CSUB_X 2
-#define CSUB_Y 2
-
-#define RFULL (31*257)
-#define GFULL (63*257)
-#define BFULL (31*257)
-#define RYFAC 9277     /* 31 * 257 *  1.0      * (255/219) */
-#define GYFAC 18853    /* 63 * 257 *  1.0      * (255/219) */
-#define BYFAC 9277     /* 31 * 257 *  1.0      * (255/219) */
-#define RVFAC 12716    /* 31 * 257 *  1.402    * (255/224) */
-#define GVFAC (-13163) /* 63 * 257 * -0.714136 * (255/224) */
-#define GUFAC (-6343)  /* 63 * 257 * -0.344136 * (255/224) */
-#define BUFAC 16071    /* 31 * 257 *  1.772    * (255/224) */
-
-#define ROUNDOFFS (127*257)
-
+/* Line write helper function for lcd_yuv_blit. Write two lines of yuv420. */
+extern void lcd_write_yuv420_lines(fb_data *dst,
+                                   unsigned char chroma_buf[LCD_HEIGHT/2*3],
+                                   unsigned char const * const src[3],
+                                   int width,
+                                   int stride);
 /* Performance function to blit a YUV bitmap directly to the LCD */
 /* For the Gigabeat - show it rotated */
 /* So the LCD_WIDTH is now the height */
@@ -316,90 +306,38 @@ void lcd_yuv_blit(unsigned char * const src[3],
                   int src_x, int src_y, int stride,
                   int x, int y, int width, int height)
 {
-    width = (width + 1) & ~1;
-    fb_data *dst = (fb_data*)FRAME + x * LCD_WIDTH + (LCD_WIDTH - y) - 1;
-    fb_data *dst_last = dst - (height - 1);
+    /* Caches for chroma data so it only need be recaculated every other
+       line */
+    unsigned char chroma_buf[LCD_HEIGHT/2*3]; /* 480 bytes */
+    unsigned char const * yuv_src[3];
+    off_t z;
 
-    for (;;)
-    {
-        fb_data *dst_row = dst;
-        int count = width;
-        const unsigned char *ysrc = src[0] + stride * src_y + src_x;
-        int y, u, v;
-        int red, green, blue;
-        unsigned rbits, gbits, bbits;
-
-        /* upsampling, YUV->RGB conversion and reduction to RGB565 in one go */
-        const unsigned char *usrc = src[1] + (stride/CSUB_X) * (src_y/CSUB_Y)
-                                           + (src_x/CSUB_X);
-        const unsigned char *vsrc = src[2] + (stride/CSUB_X) * (src_y/CSUB_Y)
-                                           + (src_x/CSUB_X);
-        int xphase = src_x % CSUB_X;
-        int rc, gc, bc;
-
-        u = *usrc++ - 128;
-        v = *vsrc++ - 128;
-        rc = RVFAC * v + ROUNDOFFS;
-        gc = GVFAC * v + GUFAC * u + ROUNDOFFS;
-        bc = BUFAC * u + ROUNDOFFS;
-
-        do
-        {
-            y = *ysrc++ - 16;
-            red   = RYFAC * y + rc;
-            green = GYFAC * y + gc;
-            blue  = BYFAC * y + bc;
-
-            if ((unsigned)red > (RFULL*255+ROUNDOFFS))
-            {
-                if (red < 0)
-                    red = 0;
-                else
-                    red = (RFULL*255+ROUNDOFFS);
-            }
-            if ((unsigned)green > (GFULL*255+ROUNDOFFS))
-            {
-                if (green < 0)
-                    green = 0;
-                else
-                    green = (GFULL*255+ROUNDOFFS);
-            }
-            if ((unsigned)blue > (BFULL*255+ROUNDOFFS))
-            {
-                if (blue < 0)
-                    blue = 0;
-                else
-                    blue = (BFULL*255+ROUNDOFFS);
-            }
-            rbits = ((unsigned)red) >> 16 ;
-            gbits = ((unsigned)green) >> 16 ;
-            bbits = ((unsigned)blue) >> 16 ;
-
-            *dst_row = (rbits << 11) | (gbits << 5) | bbits;
-
-            /* next pixel - since rotated, add WIDTH */
-            dst_row += LCD_WIDTH;
-
-            if (++xphase >= CSUB_X)
-            {
-                u = *usrc++ - 128;
-                v = *vsrc++ - 128;
-                rc = RVFAC * v + ROUNDOFFS;
-                gc = GVFAC * v + GUFAC * u + ROUNDOFFS;
-                bc = BUFAC * u + ROUNDOFFS;
-                xphase = 0;
-            }
-        }
-        while (--count);
+    if (!lcd_on)
+        return;
 
-        if (dst == dst_last) break;
+    /* Sorry, but width and height must be >= 2 or else */
+    width &= ~1;
+    height >>= 1;
 
-        dst--;
-        src_y++;
+    fb_data *dst = (fb_data*)FRAME + x * LCD_WIDTH + (LCD_WIDTH - y) - 1;
+
+    z = stride*src_y;
+    yuv_src[0] = src[0] + z + src_x;
+    yuv_src[1] = src[1] + (z >> 2) + (src_x >> 1);
+    yuv_src[2] = src[2] + (yuv_src[1] - src[1]);
+
+    do
+    {
+        lcd_write_yuv420_lines(dst, chroma_buf, yuv_src, width,
+                               stride);
+        yuv_src[0] += stride << 1; /* Skip down two luma lines */
+        yuv_src[1] += stride >> 1; /* Skip down one chroma line */
+        yuv_src[2] += stride >> 1;
+        dst -= 2;
     }
+    while (--height > 0);
 }
 
-
 void lcd_set_contrast(int val) {
   (void) val;
   // TODO: