rk27xx lcd code rework

Use DMA engine for fullscreen updates and bypass mode for partial
updates. This gives major boost on rk27generic:
default ARM:AHB:APB 200:100:50
HEAD    1/1:  26.3fps 1/4: 105.0fps
patched 1/1: 116.5fps 1/4: 249.5fps

with freq scalling NORMAL mode ARM:AHB:APB 50:50:50
HEAD    1/1:  13.1fps 1/4: 52.5fps
patched 1/1:  54.5fps 1/4: 119.0fps

Tested on rk27generic noname DAP and on Hifimans.

Change-Id: Id9dd4d2d61542c7ea6b5c6336b170d6357cefde9

1 files changed, 178 insertions, 74 deletions

diff --git a/firmware/target/arm/rk27xx/lcdif-rk27xx.c b/firmware/target/arm/rk27xx/lcdif-rk27xx.c
index affc49b213..01b19603fd 100644
--- a/firmware/target/arm/rk27xx/lcdif-rk27xx.c
+++ b/firmware/target/arm/rk27xx/lcdif-rk27xx.c
@@ -5,9 +5,9 @@
  *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
  *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
  *                     \/            \/     \/    \/            \/
- * $Id$
  *
  * Copyright (C) 2011 Marcin Bukat
+ * Copyright (C) 2012 Andrew Ryabinin
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
@@ -25,9 +25,28 @@
 #include "system.h"
 #include "cpu.h"
 #include "lcdif-rk27xx.h"
+#include "lcd-target.h"
+#include <string.h>
 
+/* dwdma linked list struct - hw defined
+ * we don't use __attribute__((packed)) just because
+ * all fields are 32bits and so are aligned
+ */
+struct llp_t {
+    uint32_t sar;
+    uint32_t dar;
+    struct llp_t *llp;
+    uint32_t ctl_l;
+    uint32_t ctl_h;
+    uint32_t dstat;
+};
+
+/* array of structs which describes full screen update
+ * each struct describes dma transfer for single line
+ */
+static struct llp_t scr_llp[LCD_HEIGHT];
 
-unsigned int lcd_data_transform(unsigned int data)
+static uint32_t lcd_data_transform(uint32_t data)
 {
     unsigned int r, g, b;
 
@@ -49,38 +68,10 @@ unsigned int lcd_data_transform(unsigned int data)
     return (r | g | b);
 }
 
-void lcd_cmd(unsigned int cmd)
-{
-    LCD_COMMAND = lcd_data_transform(cmd);
-}
-
-void lcd_data(unsigned int data)
-{
-    LCD_DATA = lcd_data_transform(data);
-}
-
-void lcd_write_reg(unsigned int reg, unsigned int val)
-{
-    lcd_cmd(reg);
-    lcd_data(val);
-}
-
-static void lcdctrl_bypass(unsigned int on_off)
-{
-    while (!(LCDC_STA & LCDC_MCU_IDLE));
-
-    if (on_off)
-        MCU_CTRL |= MCU_CTRL_BYPASS;
-    else
-        MCU_CTRL &= ~MCU_CTRL_BYPASS;
-}
-
-/* This part is unclear. I am unable to use buffered/FIFO based writes
- * to lcd. Depending on settings of IF I get various patterns on display
- * but not what I want to display apparently.
- */
 static void lcdctrl_init(void)
 {
+    int i;
+
     /* alpha b111
      * stop at current frame complete
      * MCU mode
@@ -89,37 +80,43 @@ static void lcdctrl_init(void)
     LCDC_CTRL = ALPHA(7) | LCDC_STOP | LCDC_MCU | RGB24B;
     MCU_CTRL = ALPHA_BASE(0x3f) | MCU_CTRL_BYPASS;
 
-    HOR_ACT = LCD_WIDTH + 3;    /* define horizonatal active region */
-    VERT_ACT = LCD_HEIGHT;       /* define vertical active region */
-    VERT_PERIOD = 0xfff;  /* CSn/WEn/RDn signal timings */
-
-    LINE0_YADDR = LINE_ALPHA_EN | 0x7fe;
-    LINE1_YADDR = LINE_ALPHA_EN | ((1 * LCD_WIDTH) - 2);
-    LINE2_YADDR = LINE_ALPHA_EN | ((2 * LCD_WIDTH) - 2);
-    LINE3_YADDR = LINE_ALPHA_EN | ((3 * LCD_WIDTH) - 2);
+    /* Warning: datasheet addresses and OF addresses
+     * don't match for HOR_ACT and VERT_ACT
+     */
+    HOR_ACT = LCD_WIDTH + 3;        /* define horizonatal active region */
+    VERT_ACT = LCD_HEIGHT;          /* define vertical active region */
+    VERT_PERIOD = (1<<7)|(1<<5)|1;  /* CSn/WEn/RDn signal timings */
 
-    LINE0_UVADDR = 0x7fe + 1;
-    LINE1_UVADDR = ((1 * LCD_WIDTH) - 2 + 1);
-    LINE2_UVADDR = ((2 * LCD_WIDTH) - 2 + 1);
-    LINE3_UVADDR = ((3 * LCD_WIDTH) - 2 + 1);
+    lcd_display_init();
+    lcdctrl_bypass(0);
 
-#if 0
+    /* This lines define layout of data in lcdif internal buffer
+     * LINEx_UVADDR = LINEx_YADDR + 1
+     * buffer is organized as 2048 x 32bit
+     * we use RGB565 (16 bits per pixel) so we pack 2 pixels
+     * in every lcdbuffer mem cell
+     */
     LINE0_YADDR = 0;
-    LINE1_YADDR = (1 * LCD_WIDTH);
-    LINE2_YADDR = (2 * LCD_WIDTH);
-    LINE3_YADDR = (3 * LCD_WIDTH);
+    LINE1_YADDR = 1 * LCD_WIDTH/2;
+    LINE2_YADDR = 2 * LCD_WIDTH/2;
+    LINE3_YADDR = 3 * LCD_WIDTH/2;
 
     LINE0_UVADDR = 1;
-    LINE1_UVADDR = (1 * LCD_WIDTH) + 1;
-    LINE2_UVADDR = (2 * LCD_WIDTH) + 1;
-    LINE3_UVADDR = (3 * LCD_WIDTH) + 1;
-
-    START_X = 0;
-    START_Y = 0;
-    DELTA_X = 0x200; /* no scaling */
-    DELTA_Y = 0x200; /* no scaling */
-#endif
-    LCDC_INTR_MASK = INTR_MASK_LINE; /* INTR_MASK_EVENLINE; */
+    LINE1_UVADDR = (1 * LCD_WIDTH/2) + 1;
+    LINE2_UVADDR = (2 * LCD_WIDTH/2) + 1;
+    LINE3_UVADDR = (3 * LCD_WIDTH/2) + 1;
+
+    LCDC_INTR_MASK = INTR_MASK_EVENLINE;
+
+    /* This loop seems to fix strange glitch where
+     * whole lcd content was shifted ~4 lines verticaly
+     * on second lcd_update call
+     */
+    for (i=0; i<2048; i++)
+        *((uint32_t *)LCD_BUFF + i) = 0;
+
+    /* Setup buffered writes to lcd controler */
+    MCU_CTRL = MCU_CTRL_RS_HIGH|MCU_CTRL_BUFF_WRITE|MCU_CTRL_BUFF_START;
 }
 
 /* configure pins to drive lcd in 18bit mode (16bit mode for HiFiMAN's) */
@@ -138,26 +135,133 @@ static void iomux_lcd(enum lcdif_mode_t mode)
     SCU_IOMUXB_CON |= IOMUX_LCD_D815;
 }
 
-void lcdif_init(enum lcdif_mode_t mode)
+static void dwdma_init(void)
 {
-    iomux_lcd(mode);   /* setup pins for lcd interface */
-    lcdctrl_init();    /* basic lcdc module configuration */
-    lcdctrl_bypass(1); /* run in bypass mode - all writes goes directly to lcd controller */
+    DWDMA_DMA_CHEN = 0xf00;
+    DWDMA_CLEAR_BLOCK = 0x0f;
+    DWDMA_DMA_CFG = 1;        /* global enable */
 }
 
-/* This is ugly hack. We drive lcd in bypass mode
- * where all writes goes directly to lcd controller.
- * This is suboptimal at best. IF module povides
- * FIFO, internal sram buffer, hardware scaller,
- * DMA signals, hardware alpha blending and more.
- * BUT the fact is that I have no idea how to use
- * this modes. Datasheet floating around is very
- * unclean in this regard and OF uses ackward
- * lcd update routines which are hard to understand.
- * Moreover OF sets some bits in IF module registers
- * which are referred as reseved in datasheet.
+static void llp_setup(void *src, void *dst, struct llp_t *llp, uint32_t size)
+{
+    llp->sar = (uint32_t)src;
+    llp->dar = (uint32_t)dst;
+    llp->llp = llp + 1;
+    llp->ctl_h = size;
+    llp->ctl_l = (1<<20) |
+                 (1<<23) |
+                 (1<<17) |
+                 (2<<1)  |
+                 (2<<4)  |
+                 (3<<11) |
+                 (3<<14) |
+                 (1<<27) |
+                 (1<<28) ;
+}
+
+static void llp_end(struct llp_t *llp)
+{
+    llp->ctl_l &= ~((1<<27)|(1<<28));
+}
+
+
+static void dwdma_start(uint8_t ch, struct llp_t *llp, uint8_t handshake)
+{
+    DWDMA_SAR(ch) = 0;
+    DWDMA_DAR(ch) = 0;
+    DWDMA_LLP(ch) = (uint32_t)llp;
+    DWDMA_CTL_L(ch) = (1<<20) |
+                      (1<<23) |
+                      (1<<17) |
+                      (2<<1)  |
+                      (2<<4)  |
+                      (3<<11) |
+                      (3<<14) |
+                      (1<<27) |
+                      (1<<28) ;
+              
+   DWDMA_CTL_H(ch) = 1;
+   DWDMA_CFG_L(ch) = (7<<5);
+   DWDMA_CFG_H(ch) = (handshake<<11)|(1<<2);
+   DWDMA_SGR(ch) = (13<<20);
+   DWDMA_DMA_CHEN = (0x101<<ch);
+}
+
+static void create_llp(void)
+{
+    int i;
+
+    /* build LLPs */
+    for (i=0; i<LCD_HEIGHT; i++)
+        llp_setup((void *)FBADDR(0,i),
+                  (void*)(LCD_BUFF+((i%4)*4*LCD_WIDTH/2)),
+                  &(scr_llp[i]),
+                  LCD_WIDTH/2);
+
+    llp_end(&scr_llp[LCD_HEIGHT-1]);
+}
+
+/* Public functions */
+
+/* Switch between lcdif bypass mode and buffered mode
+ * used in private implementations of lcd_set_gram_area()
+ */
+void lcdctrl_bypass(unsigned int on_off)
+{
+    while (!(LCDC_STA & LCDC_MCU_IDLE));
+
+    if (on_off)
+        MCU_CTRL |= MCU_CTRL_BYPASS;
+    else
+        MCU_CTRL &= ~MCU_CTRL_BYPASS;
+}
+
+/* Helper functions used to write commands
+ * to lcd controler in bypass mode
+ * used in controler specific implementations of:
+ * lcd_sleep()
+ * lcd_display_init()
  */
+void lcd_cmd(unsigned int cmd)
+{
+    LCD_COMMAND = lcd_data_transform(cmd);
+}
+
+void lcd_data(unsigned int data)
+{
+    LCD_DATA = lcd_data_transform(data);
+}
+
+void lcd_write_reg(unsigned int reg, unsigned int val)
+{
+    lcd_cmd(reg);
+    lcd_data(val);
+}
+
+/* rockbox API functions */
+void lcd_init_device(void)
+{
+    iomux_lcd(LCD_DATABUS_WIDTH);   /* setup pins for lcd interface */
+    dwdma_init();                   /* init dwdma module */
+    create_llp();                   /* build LLPs for screen update dma */
+    lcdctrl_init();                 /* basic lcdc module configuration */
+}
+
 void lcd_update()
 {
-    lcd_update_rect(0, 0, LCD_WIDTH, LCD_HEIGHT);
+    lcd_set_gram_area(0, 0, LCD_WIDTH, LCD_HEIGHT);
+    lcdctrl_bypass(0);
+
+    commit_discard_dcache_range(FBADDR(0,0), 2*LCD_WIDTH*LCD_HEIGHT);
+
+    while (!(LCDC_STA & LCDC_MCU_IDLE));
+
+    dwdma_start(0, scr_llp, 6);
+    udelay(10);
+
+    /* Setup buffered writes to lcd controler */
+    MCU_CTRL = MCU_CTRL_RS_HIGH|MCU_CTRL_BUFF_WRITE|MCU_CTRL_BUFF_START;
+
+    /* Wait for DMA transfer to finish */
+    while (DWDMA_CTL_L(0) & (1<<27));
 }