make gp2x mp3 playback functional (need to unpack and compile helix decoder separately in platform/common/helix)

Makefile

@@ -159,8 +159,10 @@ endif
 
 endif # USE_FRONTEND
 
-OBJS += platform/common/mp3.o
+OBJS += platform/common/mp3.o platform/common/mp3_sync.o
 ifeq "$(PLATFORM_MP3)" "1"
+platform/common/mp3_helix.o: CFLAGS += -Iplatform/libpicofe
+OBJS += platform/common/mp3_helix.o
 else ifeq "$(HAVE_LIBAVCODEC)" "1"
 OBJS += platform/common/mp3_libavcodec.o
 else

platform/common/helix/Makefile

@@ -0,0 +1,42 @@
+CROSS ?= arm-linux-gnueabi-
+
+CC = $(CROSS)gcc
+AS = $(CROSS)as
+AR = $(CROSS)ar
+TOOLCHAIN = $(notdir $(CROSS))
+
+CFLAGS += -Ipub -O2 -Wall -fstrict-aliasing -ffast-math
+ifneq ($(findstring arm-,$(TOOLCHAIN)),)
+CFLAGS += -mcpu=arm940t -mtune=arm940t -mfloat-abi=soft -mfpu=fpa -mabi=apcs-gnu -mno-thumb-interwork
+ASFLAGS = -mcpu=arm940t -mfloat-abi=soft -mfpu=fpa -mabi=apcs-gnu
+OBJS += real/arm/asmpoly_gcc.o
+else
+CFLAGS += -m32
+ASFLAGS += -m32
+OBJS += real/polyphase.o
+endif
+
+LIB = $(TOOLCHAIN)helix_mp3.a
+SHLIB = $(TOOLCHAIN)helix_mp3.so
+
+all: $(LIB) $(SHLIB)
+
+
+OBJS += mp3dec.o mp3tabs.o
+#OBJS += ipp/bitstream.o ipp/buffers.o ipp/dequant.o ipp/huffman.o ipp/imdct.o ipp/subband.o
+OBJS += real/bitstream.o real/buffers.o real/dct32.o real/dequant.o real/dqchan.o real/huffman.o
+OBJS += real/hufftabs.o real/imdct.o real/scalfact.o real/stproc.o real/subband.o real/trigtabs.o
+
+OBJS += lib.o
+
+real/arm/asmpoly_gcc.o: real/arm/asmpoly_gcc.s
+	$(CC) -o $@ $(ASFLAGS) -c $<
+
+$(LIB) : $(OBJS)
+	$(AR) r $@ $^
+$(SHLIB) : $(OBJS) /home/build/opt/open2x/gcc-4.1.1-glibc-2.3.6/lib/gcc/arm-open2x-linux/4.1.1/libgcc.a
+	$(CC) -o $@ -nostdlib -shared $(CFLAGS) $^
+
+clean:
+	$(RM) -f $(OBJS)
+

platform/common/helix/lib.c

@@ -0,0 +1,122 @@
+#include <stdlib.h>
+#include <stdint.h>
+
+// libgcc has this with gcc 4.x
+void raise(int sig)
+{
+}
+
+// very limited heap functions for helix decoder
+
+static char heap[65000] __attribute__((aligned(16)));
+static long heap_offs;
+
+void __malloc_init(void)
+{
+	heap_offs = 0;
+}
+
+void *malloc(size_t size)
+{
+        void *chunk = heap + heap_offs;
+        size = (size+15) & ~15;
+        if (heap_offs + size > sizeof(heap))
+                return NULL;
+        else {
+                heap_offs += size;
+                return chunk;
+        }
+}
+
+void free(void *chunk)
+{
+	if (chunk == heap)
+		heap_offs = 0;
+}
+
+#if 0
+void *memcpy (void *dest, const void *src, size_t n)
+{
+	char       *_dest = dest;
+        const char *_src = src;
+	while (n--) *_dest++ = *_src++;
+	return dest;
+}
+
+void *memmove (void *dest, const void *src, size_t n)
+{
+	char       *_dest = dest+n;
+        const char *_src = src+n;
+	if (dest <= src || dest >= _src)
+		return memcpy(dest, src, n);
+	while (n--) *--_dest = *--_src;
+	return dest;
+}
+#else
+/* memcpy/memmove in C with some simple optimizations.
+ * ATTN does dirty aliasing tricks with undefined behaviour by standard.
+ * (this works fine with gcc, though...)
+ */
+void *memcpy(void *dest, const void *src, size_t n)
+{
+	struct _16 { uint32_t a[4]; };
+	union { const void *v; char *c; uint64_t *l; struct _16 *s; }
+		ss = { src }, ds = { dest };
+	const int lm = sizeof(uint32_t)-1;
+
+	if ((((unsigned)ss.c ^ (unsigned)ds.c) & lm) == 0) {
+		/* fast copy if pointers have the same aligment */
+		while (((unsigned)ss.c & lm) && n > 0)	/* align to word */
+			*ds.c++ = *ss.c++, n--;
+		while (n >= sizeof(struct _16))	/* copy 16 bytes blocks */
+			*ds.s++ = *ss.s++, n -= sizeof(struct _16);
+		if (n >= sizeof(uint64_t))	/* copy leftover 8 byte block */
+			*ds.l++ = *ss.l++, n -= sizeof(uint64_t);
+	} else {
+		/* byte copy if pointers are unaligned */
+		while (n >= 8) {		/* copy 8 byte blocks */
+			*ds.c++ = *ss.c++, n--; *ds.c++ = *ss.c++, n--;
+			*ds.c++ = *ss.c++, n--; *ds.c++ = *ss.c++, n--;
+			*ds.c++ = *ss.c++, n--; *ds.c++ = *ss.c++, n--;
+			*ds.c++ = *ss.c++, n--; *ds.c++ = *ss.c++, n--;
+		}
+	}
+	/* copy max. 8 leftover bytes */
+	while (n > 0)
+		*ds.c++ = *ss.c++, n--;
+	return dest;
+}
+
+void *memmove (void *dest, const void *src, size_t n)
+{
+	struct _16 { uint32_t a[4]; };
+	union { const void *v; char *c; uint64_t *l; struct _16 *s; }
+		ss = { src+n }, ds = { dest+n };
+	const int lm = sizeof(uint32_t)-1;
+
+	if (dest <= src || dest >= src+n)
+		return memcpy(dest, src, n);
+
+	if ((((unsigned)ss.c ^ (unsigned)ds.c) & lm) == 0) {
+		/* fast copy if pointers have the same aligment */
+		while (((unsigned)ss.c & lm) && n > 0)
+			*--ds.c = *--ss.c, n--;
+		while (n >= sizeof(struct _16))
+			*--ds.s = *--ss.s, n -= sizeof(struct _16);
+		if (n >= sizeof(uint64_t))
+			*--ds.l = *--ss.l, n -= sizeof(uint64_t);
+	} else {
+		/* byte copy if pointers are unaligned */
+		while (n >= 8) {
+			*--ds.c = *--ss.c, n--; *--ds.c = *--ss.c, n--;
+			*--ds.c = *--ss.c, n--; *--ds.c = *--ss.c, n--;
+			*--ds.c = *--ss.c, n--; *--ds.c = *--ss.c, n--;
+			*--ds.c = *--ss.c, n--; *--ds.c = *--ss.c, n--;
+		}
+	}
+	/* copy max. 8 leftover bytes */
+	while (n > 0)
+		*--ds.c = *--ss.c, n--;
+	return dest;
+}
+#endif

platform/common/mp3.c

@@ -21,33 +21,6 @@ unsigned short mpeg1_l3_bitrates[16] = {
 	0, 32, 40, 48, 56, 64, 80, 96, 112, 128, 160, 192, 224, 256, 320
 };
 
-int mp3_find_sync_word(const unsigned char *buf, int size)
-{
-	const unsigned char *p, *pe;
-
-	/* find byte-aligned syncword - need 12 (MPEG 1,2) or 11 (MPEG 2.5) matching bits */
-	for (p = buf, pe = buf + size - 3; p <= pe; p++)
-	{
-		int pn;
-		if (p[0] != 0xff)
-			continue;
-		pn = p[1];
-		if ((pn & 0xf8) != 0xf8 || // currently must be MPEG1
-		    (pn & 6) == 0) {       // invalid layer
-			p++; continue;
-		}
-		pn = p[2];
-		if ((pn & 0xf0) < 0x20 || (pn & 0xf0) == 0xf0 || // bitrates
-		    (pn & 0x0c) != 0) { // not 44kHz
-			continue;
-		}
-
-		return p - buf;
-	}
-
-	return -1;
-}
-
 static int try_get_bitrate(unsigned char *buf, int buf_size)
 {
 	int offs1, offs = 0;

platform/common/mp3.h

@@ -12,8 +12,8 @@ int mp3dec_decode(FILE *f, int *file_pos, int file_len);
 extern unsigned short mpeg1_l3_bitrates[16];
 
 #ifdef __GP2X__
-void mp3_update_local(int *buffer, int length, int stereo);
+int _mp3dec_start(FILE *f, int fpos_start);
-void mp3_start_play_local(void *f, int pos);
+int _mp3dec_decode(FILE *f, int *file_pos, int file_len);
 #endif
 
 #endif // __COMMON_MP3_H__

platform/common/mp3_helix.c

@@ -9,6 +9,7 @@
 
 #include <stdio.h>
 #include <string.h>
+#include <dlfcn.h>
 
 #include <pico/pico_int.h>
 #include <pico/sound/mix.h>
@@ -20,10 +21,15 @@ static HMP3Decoder mp3dec;
 static unsigned char mp3_input_buffer[2 * 1024];
 
 #ifdef __GP2X__
-#define mp3_update mp3_update_local
+#define mp3dec_decode _mp3dec_decode
-#define mp3_start_play mp3_start_play_local
+#define mp3dec_start _mp3dec_start
 #endif
 
+static void *libhelix;
+HMP3Decoder (*p_MP3InitDecoder)(void);
+void (*p_MP3FreeDecoder)(HMP3Decoder);
+int (*p_MP3Decode)(HMP3Decoder, unsigned char **, int *, short *, int);
+
 int mp3dec_decode(FILE *f, int *file_pos, int file_len)
 {
 	unsigned char *readPtr;
@@ -51,7 +57,7 @@ int mp3dec_decode(FILE *f, int *file_pos, int file_len)
 		bytesLeft -= offset;
 
 		had_err = err;
-		err = MP3Decode(mp3dec, &readPtr, &bytesLeft, cdda_out_buffer, 0);
+		err = p_MP3Decode(mp3dec, &readPtr, &bytesLeft, cdda_out_buffer, 0);
 		if (err) {
 			if (err == ERR_MP3_MAINDATA_UNDERFLOW && !had_err) {
 				// just need another frame
@@ -86,10 +92,31 @@ int mp3dec_decode(FILE *f, int *file_pos, int file_len)
 
 int mp3dec_start(FILE *f, int fpos_start)
 {
+	if (libhelix == NULL) {
+		libhelix = dlopen("./libhelix.so", RTLD_NOW);
+		if (libhelix == NULL) {
+			lprintf("mp3dec: load libhelix.so: %s\n", dlerror());
+			return -1;
+		}
+
+		p_MP3InitDecoder = dlsym(libhelix, "MP3InitDecoder");
+		p_MP3FreeDecoder = dlsym(libhelix, "MP3FreeDecoder");
+		p_MP3Decode = dlsym(libhelix, "MP3Decode");
+
+		if (p_MP3InitDecoder == NULL || p_MP3FreeDecoder == NULL
+		    || p_MP3Decode == NULL)
+		{
+			lprintf("mp3dec: missing symbol(s) in libhelix.so\n");
+			dlclose(libhelix);
+			libhelix = NULL;
+			return -1;
+		}
+	}
+
 	// must re-init decoder for new track
 	if (mp3dec)
-		MP3FreeDecoder(mp3dec);
+		p_MP3FreeDecoder(mp3dec);
-	mp3dec = MP3InitDecoder();
+	mp3dec = p_MP3InitDecoder();
 
 	return (mp3dec == 0) ? -1 : 0;
 }

platform/common/mp3_sync.c

@@ -0,0 +1,27 @@
+
+int mp3_find_sync_word(const unsigned char *buf, int size)
+{
+	const unsigned char *p, *pe;
+
+	/* find byte-aligned syncword - need 12 (MPEG 1,2) or 11 (MPEG 2.5) matching bits */
+	for (p = buf, pe = buf + size - 3; p <= pe; p++)
+	{
+		int pn;
+		if (p[0] != 0xff)
+			continue;
+		pn = p[1];
+		if ((pn & 0xf8) != 0xf8 || // currently must be MPEG1
+		    (pn & 6) == 0) {       // invalid layer
+			p++; continue;
+		}
+		pn = p[2];
+		if ((pn & 0xf0) < 0x20 || (pn & 0xf0) == 0xf0 || // bitrates
+		    (pn & 0x0c) != 0) { // not 44kHz
+			continue;
+		}
+
+		return p - buf;
+	}
+
+	return -1;
+}

platform/gp2x/940ctl.c

@@ -425,8 +425,7 @@ int YM2612UpdateOne_940(int *buffer, int length, int stereo, int is_buf_empty)
 int mp3dec_decode(FILE *f, int *file_pos, int file_len)
 {
 	if (!(PicoIn.opt & POPT_EXT_FM)) {
-		//mp3_update_local(buffer, length, stereo);
+		return _mp3dec_decode(f, file_pos, file_len);
-		return 0;
 	}
 
 	// check if playback was started, track not ended
@@ -457,8 +456,7 @@ int mp3dec_decode(FILE *f, int *file_pos, int file_len)
 int mp3dec_start(FILE *f, int fpos_start)
 {
 	if (!(PicoIn.opt & POPT_EXT_FM)) {
-		//mp3_start_play_local(f, pos);
+		return _mp3dec_start(f, fpos_start);
-		return -1;
 	}
 
 	if (loaded_mp3 != f)

platform/gp2x/code940/940.c

@@ -2,7 +2,7 @@
 // (c) Copyright 2006-2007, Grazvydas "notaz" Ignotas
 
 #include "940shared.h"
-#include "../../common/mp3.h"
+#include "../../common/helix/pub/mp3dec.h"
 
 static _940_data_t *shared_data = (_940_data_t *)   0x00100000;
 static _940_ctl_t  *shared_ctl  = (_940_ctl_t *)    0x00200000;
@@ -19,7 +19,7 @@ void drain_wb(void);
 // is changed by other core just before we update it
 void set_if_not_changed(int *val, int oldval, int newval);
 
-void _memcpy(void *dst, const void *src, int count);
+extern void *memcpy(void *dest, const void *src, unsigned long n);
 
 //	asm volatile ("mov r0, #0" ::: "r0");
 //	asm volatile ("mcr p15, 0, r0, c7, c6,  0" ::: "r0"); /* flush dcache */
@@ -153,6 +153,8 @@ void Main940(void)
 	int job = 0;
 	ym2612_940 = &shared_data->ym2612;
 
+//	extern unsigned __bss_start__, __bss_end__;
+//	memset(&__bss_start__, 0, &__bss_end__ - &__bss_start__);
 
 	for (;;)
 	{
@@ -167,6 +169,7 @@ void Main940(void)
 				shared_ctl->writebuff0[0] = shared_ctl->writebuff1[0] = 0xffff;
 				YM2612Init_(shared_ctl->baseclock, shared_ctl->rate);
 				/* Helix mp3 decoder */
+				__malloc_init();
 				shared_data->mp3dec = MP3InitDecoder();
 				break;
 
@@ -185,7 +188,7 @@ void Main940(void)
 
 			case JOB940_PICOSTATESAVE2:
 				YM2612PicoStateSave2(0, 0);
-				_memcpy(shared_ctl->writebuff0, ym2612_940->REGS, 0x200);
+				memcpy(shared_ctl->writebuff0, ym2612_940->REGS, 0x200);
 				break;
 
 			case JOB940_PICOSTATELOAD2_PREP:
@@ -193,7 +196,7 @@ void Main940(void)
 				break;
 
 			case JOB940_PICOSTATELOAD2:
-				_memcpy(ym2612_940->REGS, shared_ctl->writebuff0, 0x200);
+				memcpy(ym2612_940->REGS, shared_ctl->writebuff0, 0x200);
 				YM2612PicoStateLoad2(0, 0);
 				break;
 
@@ -207,6 +210,7 @@ void Main940(void)
 
 			case JOB940_MP3RESET:
 				if (shared_data->mp3dec) MP3FreeDecoder(shared_data->mp3dec);
+				__malloc_init();
 				shared_data->mp3dec = MP3InitDecoder();
 				break;
 		}
@@ -215,4 +219,3 @@ void Main940(void)
 		dcache_clean();
 	}
 }
-

platform/gp2x/code940/Makefile

@@ -1,17 +1,23 @@
 # you may or may not need to change this
-#devkit_path = x:/stuff/dev/devkitgp2x/
+#devkit_path ?= $(HOME)/opt/devkitGP2X/
-devkit_path ?= $(HOME)/opt/devkitGP2X/
+#lgcc_path = $(devkit_path)lib/gcc/arm-linux/4.0.3/
-lgcc_path = $(devkit_path)lib/gcc/arm-linux/4.0.3/
-CROSS = arm-linux-
 #CROSS = $(devkit_path)bin/arm-linux-
+#devkit_path ?= $(HOME)/opt/open2x
+#lgcc_path = $(devkit_path)/gcc-4.1.1-glibc-2.3.6/lib/gcc/arm-open2x-linux/4.1.1/
+#CROSS ?= $(devkit_path)/gcc-4.1.1-glibc-2.3.6/bin/arm-open2x-linux-
+#devkit_path ?= $(HOME)/opt/arm-unknown-linux-gnu
+#lgcc_path = $(HOME)/opt/open2x/gcc-4.1.1-glibc-2.3.6/lib/gcc/arm-open2x-linux/4.1.1/
+#CROSS ?= $(devkit_path)/bin/arm-unknown-linux-gnu-
+lgcc_path = $(HOME)/opt/open2x/gcc-4.1.1-glibc-2.3.6/lib/gcc/arm-open2x-linux/4.1.1/
+CROSS ?= arm-linux-gnueabi-
 
 # settings
 #up = 1
 
-CFLAGS += -O2 -Wall -fomit-frame-pointer -fstrict-aliasing -ffast-math
+CFLAGS += -O2 -Wall -mno-thumb-interwork -fstrict-aliasing -ffast-math
-CFLAGS += -I../.. -I. -D__GP2X__ -DARM
+CFLAGS += -I../../common/helix/pub -I../../.. -I. -D__GP2X__ -DARM
-CFLAGS += -mcpu=arm940t -mtune=arm940t
+CFLAGS += -mcpu=arm940t -mtune=arm940t -mabi=apcs-gnu -mfloat-abi=soft -mfpu=fpa
-LDFLAGS = -static -s -e code940 -Ttext 0x0 -L$(lgcc_path) -lgcc
+LDFLAGS = -static -e code940 -Ttext 0x0 -L$(lgcc_path) -lgcc
 
 GCC = $(CROSS)gcc
 STRIP = $(CROSS)strip
@@ -36,7 +42,9 @@ all: $(BIN)
 # stuff for 940 core
 
 # init, emu_control, emu
-OBJS940 += 940init.o 940.o 940ym2612.o memcpy.o misc_arm.o mp3.o
+OBJS940 += 940init.o 940.o 940ym2612.o misc_arm.o mp3_sync.o
+# the asm memcpy code crashes job LOAD2 on 940. Possibly a globbered reg?
+# OBJS940 += memcpy.o
 # the asm code seems to be faster when run on 920, but not on 940 for some reason
 # OBJS940 += ../../Pico/sound/ym2612_asm.o
 
@@ -44,12 +52,13 @@ OBJS940 += 940init.o 940.o 940ym2612.o memcpy.o misc_arm.o mp3.o
 OBJS940 += uClibc/memset.o uClibc/s_floor.o uClibc/e_pow.o uClibc/e_sqrt.o uClibc/s_fabs.o
 OBJS940 += uClibc/s_scalbn.o uClibc/s_copysign.o uClibc/k_sin.o uClibc/k_cos.o uClibc/s_sin.o
 OBJS940 += uClibc/e_rem_pio2.o uClibc/k_rem_pio2.o uClibc/e_log.o uClibc/wrappers.o
+LIBHELIX ?= ../../common/helix/$(notdir $(CROSS))helix_mp3.a
 
 $(BIN) : code940.elf
 	@echo ">>>" $@
 	$(OBJCOPY) -O binary $< $@
 
-code940.elf : $(OBJS940) ../../common/helix/$(CROSS)helix-mp3.a
+code940.elf : $(OBJS940) $(LIBHELIX)
 	@echo ">>>" $@
 	$(LD) $^ $(LDFLAGS) -o $@ -Map code940.map
 
@@ -64,8 +73,12 @@ misc_arm.o : ../../../pico/misc_arm.s
 	@echo ">>>" $@
 	$(GCC) $(CFLAGS) -DEXTERNAL_YM2612 -c $< -o $@
 
-../../common/helix/helix_mp3.a:
+mp3_sync.o: ../../common/mp3_sync.c
-	@make -C ../../common/helix/
+	@echo ">>>" $@
+	$(GCC) $(CFLAGS) -Os -DCODE940 -c $< -o $@
+
+$(LIBHELIX):
+	@$(MAKE) -C ../../common/helix/ CROSS=$(CROSS)
 
 
 up: $(BIN)
@@ -82,7 +95,7 @@ tidy:
 ##
 OBJSMP3T = mp3test.o ../gp2x.o ../asmutils.o ../usbjoy.o
 
-mp3test.gpe : $(OBJSMP3T) ../helix/helix_mp3.a
+mp3test.gpe : $(OBJSMP3T) $(LIBHELIX)
 	$(GCC) -static -o $@ $^
 	$(STRIP) $@
 	@cp -v $@ /mnt/gp2x/mnt/sd

platform/gp2x/code940/mp3test.c

@@ -13,7 +13,7 @@
 //#include "emu.h"
 //#include "menu.h"
 #include "../asmutils.h"
-#include "../helix/pub/mp3dec.h"
+#include "../../helix/pub/mp3dec.h"
 
 /* we will need some gp2x internals here */
 extern volatile unsigned short *gp2x_memregs; /* from minimal library rlyeh */

platform/gp2x/code940/uClibc/memset.s

@@ -22,7 +22,7 @@
 		.text
 		.global memset
 		.type memset,%function
-		.align 4
+		.align 2
 
 memset:
 	mov	a4, a1

platform/gp2x/code940/uClibc/wrappers.c

@@ -4,9 +4,17 @@ double pow(double x, double y)
 {
 	return __ieee754_pow(x, y);
 }
+double __pow_finite(double x, double y)
+{
+	return __ieee754_pow(x, y);
+}
 
 
 double log(double x)
 {
 	return __ieee754_log(x);
 }
+double __log_finite(double x)
+{
+	return __ieee754_log(x);
+}