iPod Classic: Use DMA (and double buffering) for LCD updates

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@29474 a1c6a512-1295-4272-9138-f99709370657

2 files changed, 103 insertions, 134 deletions

diff --git a/firmware/target/arm/s5l8702/ipod6g/lcd-asm-ipod6g.S b/firmware/target/arm/s5l8702/ipod6g/lcd-asm-ipod6g.S
index 7d9e8274d5..ec8a24c4e5 100644
--- a/firmware/target/arm/s5l8702/ipod6g/lcd-asm-ipod6g.S
+++ b/firmware/target/arm/s5l8702/ipod6g/lcd-asm-ipod6g.S
@@ -19,57 +19,15 @@
  *
  ****************************************************************************/
 
-#define FORCE_FIFO_WAIT
-/**************************************************************************** 
- * #define FORCE_FIFO_WAIT
- *
- * This is not needed in YUV blitting when the LCD IF is fast enough. In this
- * case YUV-to-RGB conversion per pixel needs longer than the transfer of a 
- * pixel via the LCD IF. For iPod nano 2G this is true if the LCD IF is 
- * configured to use LCD_PHTIME = 0x00 (see lcd-nano2g.c).
- ****************************************************************************/
 
 #include "config.h"
 
     .section .icode, "ax", %progbits
     
-/****************************************************************************
- * void lcd_write_line(const fb_data *addr, 
- *                     int pixelcount, 
- *                     const unsigned int lcd_base_addr);
- * 
- * Writes pixelcount pixels from src-pointer (lcd_framebuffer) to LCD dataport.
- */
-    .align  2
-    .global lcd_write_line
-    .type   lcd_write_line, %function
-                                      /* r0 = addr, must be aligned */
-                                      /* r1 = pixel count, must be even */
-lcd_write_line:                       /* r2 = LCD_BASE */
-    stmfd   sp!, {r4-r6, lr}          /* save non-scratch registers */
-    add     r12, r2, #0x40            /* LCD_WDATA = LCD data port */
-
-.loop:
-    ldmia r0!, {r3, r5}               /* read 2 pixel (=8 byte) */
-
-    /* wait for FIFO half full */
-.fifo_wait:
-    ldr     lr, [r2, #0x1C]           /* while (LCD_STATUS & 0x08); */
-    tst     lr, #0x8
-    bgt     .fifo_wait
-    
-    mov     r4, r3, asr #16           /* r3 = 1st pixel, r4 = 2nd pixel */
-    mov     r6, r5, asr #16           /* r5 = 3rd pixel, r6 = 4th pixel */
-    stmia   r12, {r3-r6}              /* write pixels (lowest 16 bit used) */
-
-    subs    r1, r1, #4
-    bgt     .loop
-
-    ldmpc   regs=r4-r6
 
 /****************************************************************************
  * extern void lcd_write_yuv420_lines(unsigned char const * const src[3],
- *                                    const unsigned LCD_BASE,
+ *                                    uint16_t* out,
  *                                    int width,
  *                                    int stride);
  *
@@ -93,7 +51,7 @@ lcd_write_line:                       /* r2 = LCD_BASE */
     .type       lcd_write_yuv420_lines, %function
 lcd_write_yuv420_lines:
                                       /* r0 = src = yuv_src */
-                                      /* r1 = dst = LCD_BASE */
+                                      /* r1 = dst = out */
                                       /* r2 = width */
                                       /* r3 = stride */                
     stmfd       sp!, { r4-r10, lr }   /* save non-scratch */
@@ -103,14 +61,14 @@ lcd_write_yuv420_lines:
     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 */
-    add         r4, r4, #19           /*   plus room for 4 additional words, */
+    add         r4, r4, #15           /*   plus room for 3 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}           /* LCD_BASE, width, &ysrc[stride], stack_alloc */
+    stmia       sp, {r2-r4}           /* width, &ysrc[stride], stack_alloc */
 
     mov         r7, r2                /* r7 = loop count */
-    add         r8, sp, #16           /* chroma buffer */
-    add         lr, r1, #0x40         /* LCD data port = LCD_BASE + 0x40 */
+    add         r8, sp, #12           /* chroma buffer */
+    mov         lr, r1                /* RGB565 data destination buffer */
 
     /* 1st loop start */
 10:                                   /* loop start */
@@ -195,23 +153,16 @@ lcd_write_yuv420_lines:
     /* calculate pixel_2 and pack with pixel_1 before writing */
     orr         r5, r5, r3, lsl #5    /* pixel_2 = r<<11 | g<<5 | b */
     orr         r5, r5, r6, lsl #11   /* r5 = pixel_2 */
-#ifdef FORCE_FIFO_WAIT
-    /* wait for FIFO half full */
-.fifo_wait1:
-    ldr         r3, [lr, #-0x24]      /* while (LCD_STATUS & 0x08); */
-    tst         r3, #0x8
-    bgt         .fifo_wait1
-#endif
-    stmia       lr, {r4,r5}           /* write pixel_1 and pixel_2 */
+    orr         r4, r4, r5, lsl #16
+    str         r4, [lr], #4          /* write pixel_1 and pixel_2 */
 
     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, r9}     /* r1  = LCD_BASE */
-                                      /* r7  = loop count */
+    add         r8, sp, #12           /* chroma buffer */
+    ldmia       sp, {r7, r9}          /* r7  = loop count */
                                       /* r9 = &ysrc[stride] */   
 
     /* 2nd loop start */
@@ -275,20 +226,14 @@ lcd_write_yuv420_lines:
     /* calculate pixel_2 and pack with pixel_1 before writing */
     orr         r5, r5, r3, lsl #5    /* pixel_2 = r<<11 | g<<5 | b */
     orr         r5, r5, r6, lsl #11   /* r5 = pixel_2 */
-#ifdef FORCE_FIFO_WAIT
-    /* wait for FIFO half full */
-.fifo_wait2:
-    ldr         r3, [lr, #-0x24]      /* while (LCD_STATUS & 0x08); */
-    tst         r3, #0x8
-    bgt         .fifo_wait2
-#endif
-    stmia       lr, {r4,r5}           /* write pixel_1 and pixel_2 */
+    orr         r4, r4, r5, lsl #16
+    str         r4, [lr], #4          /* write pixel_1 and pixel_2 */
 
     subs        r7, r7, #2            /* check for loop end */
     bgt         20b                   /* back to beginning  */
     /* 2nd loop end */
 
-    ldr         r3, [sp, #12]
+    ldr         r3, [sp, #8]
     add         sp, sp, r3            /* deallocate buffer */
     ldmpc       regs=r4-r10           /* restore registers */
 
diff --git a/firmware/target/arm/s5l8702/ipod6g/lcd-ipod6g.c b/firmware/target/arm/s5l8702/ipod6g/lcd-ipod6g.c
index 7b3e9319d7..de731a91b8 100644
--- a/firmware/target/arm/s5l8702/ipod6g/lcd-ipod6g.c
+++ b/firmware/target/arm/s5l8702/ipod6g/lcd-ipod6g.c
@@ -27,6 +27,7 @@
 #include "cpu.h"
 #include "pmu-target.h"
 #include "power.h"
+#include "string.h"
 
 
 #define R_HORIZ_GRAM_ADDR_SET     0x200
@@ -48,6 +49,10 @@
 /** globals **/
 
 int lcd_type; /* also needed in debug-s5l8702.c */
+static struct dma_lli lcd_lli[(LCD_WIDTH * LCD_HEIGHT - 1) / 0xfff] CACHEALIGN_ATTR;
+static struct wakeup lcd_wakeup;
+static struct mutex lcd_mutex;
+static uint16_t lcd_dblbuf[LCD_HEIGHT][LCD_WIDTH];
 
 
 static inline void s5l_lcd_write_cmd_data(int cmd, int data)
@@ -100,6 +105,7 @@ bool lcd_active(void)
 
 void lcd_shutdown(void)
 {
+    mutex_lock(&lcd_mutex);
     pmu_write(0x2b, 0);  /* Kill the backlight, instantly. */
     pmu_write(0x29, 0);
 
@@ -131,6 +137,7 @@ void lcd_shutdown(void)
         s5l_lcd_write_cmd(0x10);
         sleep(HZ / 20);
     }
+    mutex_unlock(&lcd_mutex);
 }
 
 void lcd_sleep(void)
@@ -142,6 +149,8 @@ void lcd_sleep(void)
 void lcd_init_device(void)
 {
     /* Detect lcd type */
+    wakeup_init(&lcd_wakeup);
+    mutex_init(&lcd_mutex);
     lcd_type = (PDAT6 & 0x30) >> 4;
 }
 
@@ -149,7 +158,9 @@ void lcd_init_device(void)
 
 static inline void lcd_write_pixel(fb_data pixel)
 {
+    mutex_lock(&lcd_mutex);
     LCD_WDATA = pixel;
+    mutex_unlock(&lcd_mutex);
 }
 
 /* Update the display.
@@ -165,121 +176,134 @@ extern void lcd_write_line(const fb_data *addr,
                            int pixelcount,
                            const unsigned int lcd_base_addr);
 
-/* Update a fraction of the display. */
-void lcd_update_rect(int, int, int, int) ICODE_ATTR;
-void lcd_update_rect(int x, int y, int width, int height)
+static void displaylcd_setup(int x, int y, int width, int height) ICODE_ATTR;
+static void displaylcd_setup(int x, int y, int width, int height)
 {
-    int y0, x0, y1, x1;
-    fb_data* p;
-    
-    /* Both x and width need to be preprocessed due to asm optimizations */
-    x     = x & ~1;                 /* ensure x is even */
-    width = (width + 3) & ~3;       /* ensure width is a multiple of 4 */
+    mutex_lock(&lcd_mutex);
+    while (DMAC0C4CONFIG & 1) wakeup_wait(&lcd_wakeup, HZ / 10);
 
-    x0 = x;                         /* start horiz */
-    y0 = y;                         /* start vert */
-    x1 = (x + width) - 1;           /* max horiz */
-    y1 = (y + height) - 1;          /* max vert */
+    int xe = (x + width) - 1;           /* max horiz */
+    int ye = (y + height) - 1;          /* max vert */
 
     if (lcd_type & 2) {
-        s5l_lcd_write_cmd_data(R_HORIZ_ADDR_START_POS, x0);
-        s5l_lcd_write_cmd_data(R_HORIZ_ADDR_END_POS,   x1);
-        s5l_lcd_write_cmd_data(R_VERT_ADDR_START_POS,  y0);
-        s5l_lcd_write_cmd_data(R_VERT_ADDR_END_POS,    y1);
+        s5l_lcd_write_cmd_data(R_HORIZ_ADDR_START_POS, x);
+        s5l_lcd_write_cmd_data(R_HORIZ_ADDR_END_POS,   xe);
+        s5l_lcd_write_cmd_data(R_VERT_ADDR_START_POS,  y);
+        s5l_lcd_write_cmd_data(R_VERT_ADDR_END_POS,    ye);
 
-        s5l_lcd_write_cmd_data(R_HORIZ_GRAM_ADDR_SET,  x0);
-        s5l_lcd_write_cmd_data(R_VERT_GRAM_ADDR_SET,   y0);
+        s5l_lcd_write_cmd_data(R_HORIZ_GRAM_ADDR_SET,  x);
+        s5l_lcd_write_cmd_data(R_VERT_GRAM_ADDR_SET,   y);
 
         s5l_lcd_write_cmd(R_WRITE_DATA_TO_GRAM);
     } else {
         s5l_lcd_write_cmd(R_COLUMN_ADDR_SET);
-        s5l_lcd_write_data(x0 >> 8);
-        s5l_lcd_write_data(x0 & 0xff);
-        s5l_lcd_write_data(x1 >> 8);
-        s5l_lcd_write_data(x1 & 0xff);
+        s5l_lcd_write_data(x >> 8);
+        s5l_lcd_write_data(x & 0xff);
+        s5l_lcd_write_data(xe >> 8);
+        s5l_lcd_write_data(xe & 0xff);
 
         s5l_lcd_write_cmd(R_ROW_ADDR_SET);
-        s5l_lcd_write_data(y0 >> 8);
-        s5l_lcd_write_data(y0 & 0xff);
-        s5l_lcd_write_data(y1 >> 8);
-        s5l_lcd_write_data(y1 & 0xff);
+        s5l_lcd_write_data(y >> 8);
+        s5l_lcd_write_data(y & 0xff);
+        s5l_lcd_write_data(ye >> 8);
+        s5l_lcd_write_data(ye & 0xff);
 
         s5l_lcd_write_cmd(R_MEMORY_WRITE);
     }
+}
+
+static void displaylcd_dma(int pixels) ICODE_ATTR;
+static void displaylcd_dma(int pixels)
+{
+    int i;
+    void* data = lcd_dblbuf;
+    for (i = -1; i < (int)ARRAYLEN(lcd_lli) && pixels > 0; i++, pixels -= 0xfff)
+    {
+        bool last = i + 1 >= (int)ARRAYLEN(lcd_lli) || pixels <= 0xfff;
+        struct dma_lli* lli = i < 0 ? (struct dma_lli*)((int)&DMAC0C4LLI) : &lcd_lli[i];
+        lli->srcaddr = data;
+        lli->dstaddr = (void*)((int)&LCD_WDATA);
+        lli->nextlli = last ? NULL : &lcd_lli[i + 1];
+        lli->control = 0x70240000 | (last ? pixels : 0xfff)
+                     | (last ? 0x80000000 : 0) | 0x4000000;
+        data += 0x1ffe;
+    }
+    clean_dcache();
+    DMAC0C4CONFIG = 0x88c1;
+    mutex_unlock(&lcd_mutex);
+}
+
+void INT_DMAC0C4(void) ICODE_ATTR;
+void INT_DMAC0C4(void)
+{
+    DMAC0INTTCCLR = 0x10;
+    wakeup_signal(&lcd_wakeup);
+}
+
+/* Update a fraction of the display. */
+void lcd_update_rect(int, int, int, int) ICODE_ATTR;
+void lcd_update_rect(int x, int y, int width, int height)
+{
+    int pixels = width * height;
+    fb_data* p = &lcd_framebuffer[y][x];
+    uint16_t* out = lcd_dblbuf[0];
+    
+    displaylcd_setup(x, y, width, height);
 
     /* Copy display bitmap to hardware */
-    p = &lcd_framebuffer[y0][x0];
     if (LCD_WIDTH == width) {
         /* Write all lines at once */
-        lcd_write_line(p, height*LCD_WIDTH, LCD_BASE);
+        memcpy(out, p, pixels * 2);
     } else {
-        y1 = height;
         do {
             /* Write a single line */
-            lcd_write_line(p, width, LCD_BASE);
+            memcpy(out, p, width * 2);
             p += LCD_WIDTH;
-        } while (--y1 > 0 );
+            out += width;
+        } while (--height);
     }
+
+    displaylcd_dma(pixels);
 }
 
 /* Line write helper function for lcd_yuv_blit. Writes two lines of yuv420. */
 extern void lcd_write_yuv420_lines(unsigned char const * const src[3],
-                                   const unsigned int lcd_baseadress,
+                                   uint16_t* outbuf,
                                    int width,
                                    int stride);
 
 /* 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,
+                  int x, int y, int width, int height) ICODE_ATTR;
+void lcd_blit_yuv(unsigned char * const src[3],
+                  int src_x, int src_y, int stride,
                   int x, int y, int width, int height)
 {
-    unsigned int z, y0, x0, y1, x1;;
+    unsigned int z;
     unsigned char const * yuv_src[3];
     
     width = (width + 1) & ~1;       /* ensure width is even */
 
-    x0 = x;                         /* start horiz */
-    y0 = y;                         /* start vert */
-    x1 = (x + width) - 1;           /* max horiz */
-    y1 = (y + height) - 1;          /* max vert */
-
-    if (lcd_type & 2) {
-        s5l_lcd_write_cmd_data(R_HORIZ_ADDR_START_POS, x0);
-        s5l_lcd_write_cmd_data(R_HORIZ_ADDR_END_POS,   x1);
-        s5l_lcd_write_cmd_data(R_VERT_ADDR_START_POS,  y0);
-        s5l_lcd_write_cmd_data(R_VERT_ADDR_END_POS,    y1);
-
-        s5l_lcd_write_cmd_data(R_HORIZ_GRAM_ADDR_SET,  x0);
-        s5l_lcd_write_cmd_data(R_VERT_GRAM_ADDR_SET,   y0);
-
-        s5l_lcd_write_cmd(R_WRITE_DATA_TO_GRAM);
-    } else {
-        s5l_lcd_write_cmd(R_COLUMN_ADDR_SET);
-        s5l_lcd_write_data(x0 >> 8);
-        s5l_lcd_write_data(x0 & 0xff);
-        s5l_lcd_write_data(x1 >> 8);
-        s5l_lcd_write_data(x1 & 0xff);
-
-        s5l_lcd_write_cmd(R_ROW_ADDR_SET);
-        s5l_lcd_write_data(y0 >> 8);
-        s5l_lcd_write_data(y0 & 0xff);
-        s5l_lcd_write_data(y1 >> 8);
-        s5l_lcd_write_data(y1 & 0xff);
-
-        s5l_lcd_write_cmd(R_MEMORY_WRITE);
-    }
+    int pixels = width * height;
+    uint16_t* out = lcd_dblbuf[0];
 
     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]);
 
+    displaylcd_setup(x, y, width, height);
+
     height >>= 1;
 
     do {
-        lcd_write_yuv420_lines(yuv_src, LCD_BASE, width, stride);
+        lcd_write_yuv420_lines(yuv_src, out, width, stride);
         yuv_src[0] += stride << 1;
         yuv_src[1] += stride >> 1; /* Skip down one chroma line */
         yuv_src[2] += stride >> 1;
-    } while (--height > 0);
+        out += width << 1;
+    } while (--height);
+
+    displaylcd_dma(pixels);
 }