Merge pull request #43 from fbergama/develop

Develop

Makefile

@@ -3,13 +3,13 @@ RPI ?= 1
 
 ARMGNU ?= arm-none-eabi
 ifeq ($(strip $(RPI)),1)
-CFLAGS = -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -mno-unaligned-access -fsigned-char -march=armv6j -mtune=arm1176jzf-s -DRPI=1
+CFLAGS = -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -fsigned-char -march=armv6j -mtune=arm1176jzf-s -DRPI=1
 else ifeq ($(strip $(RPI)),2)
-CFLAGS = -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -mno-unaligned-access -fsigned-char -march=armv7-a -mtune=cortex-a7 -DRPI=2
+CFLAGS = -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -fsigned-char -march=armv7-a -mtune=cortex-a7 -DRPI=2
 else ifeq ($(strip $(RPI)),3)
-CFLAGS = -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -mno-unaligned-access -fsigned-char -march=armv8-a -mtune=cortex-a53 -DRPI=3
+CFLAGS = -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -fsigned-char -march=armv8-a -mtune=cortex-a53 -DRPI=3
 else
-CFLAGS = -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -mno-unaligned-access -fsigned-char -march=armv8-a -mtune=cortex-a53 -DRPI=4
+CFLAGS = -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -fsigned-char -march=armv8-a -mtune=cortex-a53 -DRPI=4
 endif
 
 ## Important!!! asm.o must be the first object to be linked!

README.md

@@ -193,26 +193,26 @@ There's a [example](shapes/sample/anim_dinos.txt) for a moving dinosaur (after f
 
 ## Graphics performance
 
-If you want to do graphics, get a Pi Zero. Multi core Pi's like the models 2 - 4 are much slower if only one core is used like PiGFX does.
+Since we have cache enabled now on the Pi, the speed is more or less the same on every model. In most cases it's probably faster to not enable DMA.
 
-Graphics which are drawn on a multiple of 4 pixels x-position draw much faster than on a multiple of 2 or 1. You will also get more speed if your image has a width of a multiple of 4. Drawing solid is faster than drawing transparent.
+Graphics which are drawn on a multiple of 4 pixels x-position draw faster than on a multiple of 2 or 1. You will also get more speed if your image has a width of a multiple of 4. Drawing solid is faster than drawing transparent.
 
-This is a table that shows the time for drawing a 400x300 pixel image on different x-positions.
+This is a table that shows the time for drawing a 400x300 pixel image on different x-positions. This is an example for getting an idea.
 
 |Pi Gen.  |Aligned 4 px |Aligned 2 px |unaligned |DMA enabled
 |-------  |------------ |------------ |--------- |-----------
-|1 (zero) |20.8ms       |40.4ms       |53.0ms    |12.5ms
-|2        |46.9ms       |90.8ms       |114.5ms   |14.2ms
-|3        |50.1ms       |93.9ms       |122.1ms   |14.1ms
-|4        |33.4ms       |64.9ms       |144.4ms   |5.4ms
+|1 (zero) |4.7ms        |9.3ms        |10.5ms    |3.9ms
+|2        |1.9ms        |2.1ms        |2.6ms     |5.2ms
+|3        |2.2ms        |2.7ms        |3.8ms     |5.2ms
+|4        |0.9ms        |1.6ms        |3.3ms     |4.9ms
 
 ## Collision detection
 
 At the time a sprite is defined, the borders of the sprite are used for detecting collisions with other sprites, so this is always a rectangle. A collision is reported if 2 sprites touch or overlap. Collisions with lower sprite indexes are detected first.
 
 Collision detection takes place at the time a sprite is defined or moved. This sprite is checked against all other active sprites. All detected collisions are reported.
 
-A collision is reported as a keyboard input with this syntax: \ESC[#<idx1>;<idx2>c where idx1 is the sprite index you just defined or moved and idx2 is the sprite it collides with.
+A collision is reported as a keyboard input with this syntax: \ESC[#-idx1-;-idx2-c where -idx1- is the sprite index you just defined or moved and -idx2- is the sprite it collides with.
 
 ## Compiling on Mac / Linux
 

bin/pigfx.txt

@@ -14,5 +14,5 @@ swapDelWithBackspace = 1    ; Substitute DEL (0x7F) with BACKSPACE (0x08)
 
 [General]
 showRC2014Logo = 0          ; Show the RC2014 logo at startup
-disableGfxDMA = 0           ; Disable fast DMA memory access for graphics (if 1)
+disableGfxDMA = 1           ; Disable fast DMA memory access for graphics (if 1)
 disableCollision = 0        ; Disable sprite collision detection

doc/6800_8080_IF.txt

@@ -1,7 +1,21 @@
+  //////////////////////////////////////////
+  // PiGFX parallel interface description //
+  //////////////////////////////////////////
+  Based on the datasheet of RA8835
+  See https://www.newhavendisplay.com/app_notes/RA8835.pdf
+
+  ----------------
+  Hardware Signals
+  ----------------
+  The Pi must use 3.3V high level signals,
+  use a bidirectional level shifter if you use 5v signals.
+
+  The behaviour of the interface (8080 or 6800 style) is configurable in config.txt.
+
   8080 Signal  |  6800 Signal  |  Raspberry Pin    |  Description
 --------------------------------------------------------------------------
   /CS          |  /CS          |  GPIO 4 (Pin 7)   |  Chip Select (low active)
-  /RS          |  /RS          |  GPIO 18 (Pin 12) |  Register Select (1 = Data, 0 = command)
+  /RS          |  /RS          |  GPIO 18 (Pin 12) |  Register Select (1 = data, 0 = command)
   /RD          |  E            |  GPIO 25 (Pin 22) |  8080 Read (low active) / 6800 Enable (high active)
   /WR          |  R/W          |  GPIO 8 (Pin 24)  |  8080 Write (low active) Input latched at rising edge / 6800 1=read 0=write  Input latched at falling edge of E
   D0           |  D0           |  GPIO 17 (Pin 11) |  Data Bit 0
@@ -11,4 +25,33 @@
   D4           |  D4           |  GPIO 9 (Pin 21)  |  Data Bit 4
   D5           |  D5           |  GPIO 11 (Pin 23) |  Data Bit 5
   D6           |  D6           |  GPIO 23 (Pin 16) |  Data Bit 6
-  D7           |  D7           |  GPIO 24 (Pin 18) |  Data Bit 7
+  D7           |  D7           |  GPIO 24 (Pin 18) |  Data Bit 7
+
+  ---------------------------------
+  Discription of the 8080 interface
+  ---------------------------------
+  If /CS is low, all data lines are in input / high impedance state.
+
+  Reading from the Pi:
+  On the falling edge of /RD the Pi starts to set its datalines,
+  considering the state of /RS and /CS. Data is read by the CPU on the
+  rising edge of /RD.
+
+  Writing to the Pi:
+  On the rising edge of /WR, the Pi latches the state of the databus,
+  considerung the state of /RS and /CS.
+
+
+  ---------------------------------
+  Discription of the 6800 interface
+  ---------------------------------
+  If /CS is low, all data lines are in input / high impedance state.
+
+  Reading from the Pi:
+  On the rising edge of E the Pi starts to set its datalines,
+  considering the state of /RS, /CS and R/W. Data is read by the CPU
+  on the falling edge of E.
+
+  Writing to the Pi
+  On the falling edfe of E, the Pi latches the state of the databus,
+  considering the state of /RS, /CS and R/W.

doc/terminal_codes.txt

@@ -1,6 +1,7 @@
 Cursor Control
 
 <ESC>[?25l          Cursor invisible
+<ESC>[?25b          Cursor blinking
 <ESC>[?25h          Cursor visible
 <ESC>[H             Move to 0-0
 <ESC>[f             Move to 0-0

make.bat

@@ -12,7 +12,7 @@ call makeuspi %RPI%
 set OOB= asm.o exceptionstub.o synchronize.o mmu.o pigfx.o uart.o irq.o utils.o gpio.o mbox.o prop.o board.o actled.o framebuffer.o console.o gfx.o dma.o nmalloc.o uspios_wrapper.o ee_printf.o stupid_timer.o block.o emmc.o c_utils.o mbr.o fat.o config.o ini.o ps2.o keyboard.o binary_assets.o
 set LIBUSPI= uspi/lib/libuspi.a
 
-set CFLAGS= -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -mno-unaligned-access -fsigned-char
+set CFLAGS= -Wall -Wextra -O2 -g -nostdlib -nostartfiles -fno-stack-limit -ffreestanding -fsigned-char
 if "%RPI%"=="1" (
   set CFLAGS= %CFLAGS% -march=armv6j -mtune=arm1176jzf-s -DRPI=1
 )

makeall

@@ -1,9 +1,8 @@
 #!/bin/sh 
-# We need to add the compiler option -mno-unaligned-access in uspi. If not, the compiler will generate halb-word accesses on odd memory addresses. RPI1 could handle this by enabling unaligned access in a CPU config register. However RPI2&3 have it already enabled, but need a enabled MMU with pagetables. They otherwise just crash.
 for i in 1 2 3
 do
     cd uspi
-    echo -e "RASPPI = $i\nCFLAGS += -mno-unaligned-access" > Config.mk
+    echo "RASPPI = $i" > Config.mk
     ./makeall clean
     ./makeall
     cd ..

makeuspi.bat

@@ -6,7 +6,7 @@ set RPI=%1
 
 set OOBUSPI= uspilibrary.o dwhcidevice.o dwhciregister.o dwhcixferstagedata.o usbconfigparser.o usbdevice.o usbdevicefactory.o usbendpoint.o usbrequest.o usbstandardhub.o devicenameservice.o macaddress.o usbfunction.o smsc951x.o lan7800.o string.o util.o usbmassdevice.o dwhciframeschednper.o dwhciframeschedper.o keymap.o usbkeyboard.o dwhcirootport.o usbmouse.o dwhciframeschednsplit.o usbgamepad.o synchronize.o usbstring.o usbmidi.o
 
-set CFLAGS= -Wall -Wno-psabi -fsigned-char -fno-builtin -nostdinc -nostdlib -std=gnu99 -undef -mno-unaligned-access -DRASPPI=%RPI% -I ../include -O2
+set CFLAGS= -Wall -Wno-psabi -fsigned-char -fno-builtin -nostdinc -nostdlib -std=gnu99 -undef -DRASPPI=%RPI% -I ../include -O2
 if "%RPI%"=="1" (
   set CFLAGS= %CFLAGS% -march=armv6j -mtune=arm1176jzf-s
 )

pigfx_config.h.in

@@ -9,7 +9,7 @@
 
 #define PIGFX_MAJVERSION        1               /* Major version number */
 #define PIGFX_MINVERSION        8               /* Minor version number */
-#define PIGFX_BUILDVERSION      0               /* Build version. */
+#define PIGFX_BUILDVERSION      1               /* Build version. */
 
 /** Versions:
 

src/actled.c

@@ -46,7 +46,7 @@ static void set_130(const int on) {
 
   msg->header.size = sizeof(*msg);
   msg->header.code = 0;
-  msg->tag.id = UINT32_C(0x00038041);
+  msg->tag.id = MAILBOX_TAG_SET_GPIO_STATE;
   msg->tag.size = sizeof(msg->value);
   msg->tag.code = 0;
   msg->value.request.pin = 130;

src/asm.S

@@ -22,7 +22,7 @@ _reset_h:                        .word   _reset_
     _data_abort_h:               .word   DataAbortStub
     _unused_handler_h:           .word   hang
     _interrupt_h:                .word   irq_handler_
-    _fast_interrupt_h:           .word   hang
+    _fast_interrupt_h:           .word   fiq_handler_
 
 /* The bootloader starts, loads are executable, and enters */
 /* execution at 0x8000 with the following values set.      */

src/config.c

@@ -93,7 +93,7 @@ void setDefaultConfig()
     PiGfxConfig.skipBackspaceEcho = 0;
     PiGfxConfig.swapDelWithBackspace = 1;
     PiGfxConfig.showRC2014Logo = 0;
-    PiGfxConfig.disableGfxDMA = 0;
+    PiGfxConfig.disableGfxDMA = 1;
     PiGfxConfig.disableCollision = 0;
     pigfx_strcpy(PiGfxConfig.keyboardLayout, "us");
 }

src/dma.c

@@ -11,6 +11,7 @@
 #include "utils.h"
 #include "mbox.h"
 #include "console.h"
+#include "memory.h"
 
 
 #define DMA_CS_OFFSET        0x00
@@ -32,13 +33,17 @@ typedef struct _DMA_Ctrl_Block
 } DMA_Control_Block;
 
 
-DMA_Control_Block __attribute__((aligned(0x100))) ctr_blocks[16];
+// DMA Control block needs to be in a coherent section. We choose Coherent start + 2048 bytes
+// Coherent start is already used by the mailbox
+// control block needs to be aligned 256
+DMA_Control_Block* ctr_blocks;
 unsigned int curr_blk;
 
 unsigned int channel;
 
 void dma_init()
 {
+    ctr_blocks = (DMA_Control_Block*)(MEM_COHERENT_REGION+0x800);
     curr_blk = 0;
     channel = 0;
     // Enable DMA on used channel

src/framebuffer.c

@@ -92,14 +92,6 @@ static const unsigned int xterm_colors[NB_PALETTE_ELE] = {
 FB_RETURN_TYPE fb_init( unsigned int ph_w, unsigned int ph_h, unsigned int vrt_w, unsigned int vrt_h,
                         unsigned int bpp, void** pp_fb, unsigned int* pfbsize, unsigned int* pPitch )
 {
-#if ENABLED(FRAMEBUFFER_DEBUG)
-    unsigned int getPhysRes_w, getPhysRes_h;
-
-    // Get physical display size
-    if (fb_get_phys_res(&getPhysRes_w, &getPhysRes_h) != 0) return FB_GET_DISPLAY_SIZE_FAIL;
-    cout("Display size: ");cout_d(getPhysRes_w);cout("x");cout_d(getPhysRes_h);cout_endl();
-#endif
-
     // Set up the framebuffer and get one
     // Set display size
     typedef struct
@@ -194,7 +186,7 @@ FB_RETURN_TYPE fb_init( unsigned int ph_w, unsigned int ph_h, unsigned int vrt_w
     msg->tag_virt_size.size = sizeof(msg->value_virt_size);
     msg->tag_virt_size.code = 0;
     msg->value_virt_size.request.display_w = vrt_w;
-    msg->value_virt_size.request.display_h = vrt_h;
+    msg->value_virt_size.request.display_h = vrt_h*2;   // our virtual display is double the y size for implementing double buffering
 
     msg->tag_colour_depth.id = MAILBOX_TAG_SET_COLOUR_DEPTH; // Set colour depth
     msg->tag_colour_depth.size = sizeof(msg->value_colour_depth);
@@ -223,24 +215,9 @@ FB_RETURN_TYPE fb_init( unsigned int ph_w, unsigned int ph_h, unsigned int vrt_w
     *pp_fb = (void*)mem_vc2arm(msg->value_get_buf.response.bufferaddr);
     *pfbsize = msg->value_get_buf.response.buffersize;
 
-#if ENABLED(FRAMEBUFFER_DEBUG)
-    cout("Set physical display size: ");cout_d(ph_w);cout("x");cout_d(ph_h);cout_endl();
-    cout("Response: ");cout_d(msg->value_phys_size.response.act_display_w);cout("x");cout_d(msg->value_phys_size.response.act_display_h);cout_endl();
-    cout("Set virtual display size: ");cout_d(vrt_w);cout("x");cout_d(vrt_h);cout_endl();
-    cout("Response: ");cout_d(msg->value_virt_size.response.act_display_w);cout("x");cout_d(msg->value_virt_size.response.act_display_h);cout_endl();
-    cout("Set colour depth: ");cout_d(bpp);cout(" bpp");cout_endl();
-    cout("Response: ");cout_d(msg->value_colour_depth.response.act_depth);cout(" bpp");cout_endl();
-    cout("Screen addr: ");cout_h((unsigned int)*pp_fb); cout_endl();
-    cout("Screen size: ");cout_d(*pfbsize); cout_endl();
-#endif
-
     // Get pitch (bytes per line)
     if (fb_get_pitch(pPitch) != 0) return FB_INVALID_PITCH;
 
-#if ENABLED(FRAMEBUFFER_DEBUG)
-    cout("Pitch: "); cout_d(*pPitch); cout_endl();
-#endif
-
     return FB_SUCCESS;
 }
 
@@ -463,3 +440,40 @@ FB_RETURN_TYPE fb_get_pitch( unsigned int* pPitch )
     return FB_SUCCESS;
 }
 
+FB_RETURN_TYPE fb_switch_framebuffer(unsigned int yOffset)
+{
+    typedef struct {
+        mbox_msgheader_t header;
+        mbox_tagheader_t tag;
+
+        union {
+            struct {
+                uint32_t xOffset;
+                uint32_t yOffset;
+            }
+            request;
+            // No response.
+        }
+        value;
+
+        mbox_msgfooter_t footer;
+    }
+    message_t;
+
+    message_t* msg = (message_t*)MEM_COHERENT_REGION;
+
+    msg->header.size = sizeof(*msg);
+    msg->header.code = 0;
+    msg->tag.id = MAILBOX_TAG_SET_VIRTUAL_OFFSET;
+    msg->tag.size = sizeof(msg->value);
+    msg->tag.code = 0;
+    msg->value.request.xOffset = 0;
+    msg->value.request.yOffset = yOffset;
+    msg->footer.end = 0;
+
+    if (mbox_send(msg) != 0) {
+        return FB_ERROR;
+    }
+
+    return FB_SUCCESS;
+}

src/framebuffer.h

@@ -31,6 +31,7 @@ FB_RETURN_TYPE fb_get_phys_res(unsigned int* pRes_w, unsigned int* pRes_h);
 extern FB_RETURN_TYPE fb_set_grayscale_palette();
 extern FB_RETURN_TYPE fb_set_xterm_palette();
 FB_RETURN_TYPE fb_get_pitch( unsigned int* pPitch );
+FB_RETURN_TYPE fb_switch_framebuffer(unsigned int yOffset);
 
 
 #endif