Integrate ARM 64bit fixedpoint patch submitted by Tim Deegan in pkg/16479.

This feature is enabled by MPG123_ARM_FIXED64.

mpg123/Makefile.common contains common stuff of mpg123 and mpg123-esound.

2 files changed, 115 insertions, 15 deletions

diff --git a/audio/mpg123/patches/patch-aa b/audio/mpg123/patches/patch-aa
index 0b3523482a3..558a3765590 100644
--- a/audio/mpg123/patches/patch-aa
+++ b/audio/mpg123/patches/patch-aa
@@ -1,24 +1,26 @@
-$NetBSD: patch-aa,v 1.20 2002/06/22 19:03:11 kent Exp $
+$NetBSD: patch-aa,v 1.21 2002/06/23 08:45:09 kent Exp $
 
---- Makefile.orig	Sun Jun 23 03:50:48 2002
+--- Makefile.orig	Sun Jun 23 16:56:17 2002
 +++ Makefile
-@@ -54,6 +54,15 @@
+@@ -54,6 +54,17 @@
  	@echo "make aix-tk3play    IBM AIX"
  	@echo "make os2            IBM OS/2"
  	@echo "make netbsd         NetBSD"
-+	@echo "make netbsd-arm     NetBSD for strongarm"
++	@echo "make netbsd-arm     NetBSD for strongarm (32bit fixedpoint)"
++	@echo "make netbsd-arm-64  NetBSD for strongarm (64bit fixedpoint)"
 +	@echo "make netbsd-i386    NetBSD optimized for i386"
 +	@echo "make netbsd-i386-elf NetBSD optimized for i386, ELF format"
 +	@echo "make netbsd-m68k    NetBSD optimized for m68k"
 +	@echo "make netbsd-esd     NetBSD with EsounD"
-+	@echo "make netbsd-arm-esd NetBSD for strongarm with EsounD"
++	@echo "make netbsd-arm-esd NetBSD for strongarm (32bit fixedpoint) with EsounD"
++	@echo "make netbsd-arm-64-esd NetBSD for strongarm (64bit fixedpoint) with EsounD"
 +	@echo "make netbsd-i386-esd NetBSD optimized for i386 with EsounD"
 +	@echo "make netbsd-i386-elf-esd NetBSD optimized for i386, ELF format with EsounD"
-+	@echo "make netbsd-m68k-esd    NetBSD optimized for m68k with EsounD"
++	@echo "make netbsd-m68k-esd NetBSD optimized for m68k with EsounD"
  	@echo "make bsdos          BSDI BSD/OS"
  	@echo "make bsdos4         BSDI BSD/OS 4.0"
  	@echo "make bsdos-nas      BSDI BSD/OS with NAS support"
-@@ -496,18 +505,100 @@
+@@ -496,18 +507,120 @@
  
  netbsd:
  	$(MAKE) CC=cc LDFLAGS= \
@@ -46,8 +48,7 @@ $NetBSD: patch-aa,v 1.20 2002/06/22 19:03:11 kent Exp $
 +		OBJECTS='decode_i386.o dct64_i386.o decode_i586.o \
 +			audio_sun.o term.o' \
 +		CFLAGS='$(CFLAGS) -Wall -ansi -pedantic -O4 -fomit-frame-pointer \
- 			-funroll-all-loops -ffast-math -DROT_I386 \
--			-DI386_ASSEM -DREAL_IS_FLOAT -DUSE_MMAP -DNETBSD' \
++			-funroll-all-loops -ffast-math -DROT_I386 \
 +			-DI386_ASSEM -DPENTIUM_OPT -DREAL_IS_FLOAT -DUSE_MMAP \
 +			-DREAD_MMAP -DNETBSD -DTERM_CONTROL' \
 +		mpg123-make
@@ -70,6 +71,16 @@ $NetBSD: patch-aa,v 1.20 2002/06/22 19:03:11 kent Exp $
 +			-DNETBSD' \
 +		mpg123-make
 +
++netbsd-arm-64:
++	$(MAKE) CC=cc LDFLAGS= \
++		OBJECTS='decode.o dct64.o audio_sun.o term.o' \
++		CFLAGS='$(CFLAGS) -Wall -O6 -fomit-frame-pointer \
++			-funroll-all-loops -finline-functions -ffast-math \
++			-march=armv4 -mtune=strongarm110 -mcpu=strongarm110 \
++			-DREAL_IS_FIXED_ARMV4 -DUSE_MMAP -DREAD_MMAP \
++			-DTERM_CONTROL -DNETBSD' \
++		mpg123-make
++
 +netbsd-esd:
 +	$(MAKE) CC=cc LDFLAGS= \
 +		AUDIO_LIB="`$(ESD_CONFIG) --libs`" \
@@ -91,13 +102,13 @@ $NetBSD: patch-aa,v 1.20 2002/06/22 19:03:11 kent Exp $
 +		mpg123-make
 +
 +netbsd-i386-elf-esd:
-+	echo GETBITS: $(GETBITS)
 +	$(MAKE) CC=cc LDFLAGS= \
 +		AUDIO_LIB="`$(ESD_CONFIG) --libs`" \
 +		OBJECTS='decode_i386.o dct64_i386.o decode_i586.o \
 +			audio_esd.o term.o' \
 +		CFLAGS='$(CFLAGS) -Wall -ansi -pedantic -O4 -fomit-frame-pointer \
-+			-funroll-all-loops -ffast-math -DROT_I386 \
+ 			-funroll-all-loops -ffast-math -DROT_I386 \
+-			-DI386_ASSEM -DREAL_IS_FLOAT -DUSE_MMAP -DNETBSD' \
 +			-DI386_ASSEM -DPENTIUM_OPT -DREAL_IS_FLOAT -DUSE_MMAP \
 +			-DREAD_MMAP -DNETBSD -DTERM_CONTROL \
 +			`$(ESD_CONFIG) --cflags`' \
@@ -122,6 +133,17 @@ $NetBSD: patch-aa,v 1.20 2002/06/22 19:03:11 kent Exp $
 +			-march=armv4 -mtune=strongarm -mcpu=strongarm \
 +			-DREAL_IS_FIXED -DUSE_MMAP -DREAD_MMAP -DTERM_CONTROL \
 +			-DNETBSD `$(ESD_CONFIG) --cflags`' \
++		mpg123-make
++
++netbsd-arm-64-esd:
++	$(MAKE) CC=cc LDFLAGS= \
++		AUDIO_LIB="`$(ESD_CONFIG) --libs`" \
++		OBJECTS='decode.o dct64.o audio_esd.o term.o' \
++		CFLAGS='$(CFLAGS) -Wall -O6 -fomit-frame-pointer \
++			-funroll-all-loops -finline-functions -ffast-math \
++			-march=armv4 -mtune=strongarm110 -mcpu=strongarm110 \
++			-DREAL_IS_FIXED_ARMV4 -DUSE_MMAP -DREAD_MMAP \
++			-DTERM_CONTROL -DNETBSD `$(ESD_CONFIG) --cflags`' \
  		mpg123-make
  
  bsdos:
diff --git a/audio/mpg123/patches/patch-aj b/audio/mpg123/patches/patch-aj
index 58a38bbefbb..15eb00dd41c 100644
--- a/audio/mpg123/patches/patch-aj
+++ b/audio/mpg123/patches/patch-aj
@@ -1,8 +1,8 @@
-$NetBSD: patch-aj,v 1.1 1999/10/12 04:43:15 simonb Exp $
+$NetBSD: patch-aj,v 1.2 2002/06/23 08:45:09 kent Exp $
 
---- mpg123.h.orig	Sat May 29 00:51:58 1999
-+++ mpg123.h	Tue Oct 12 01:17:16 1999
-@@ -56,8 +56,37 @@
+--- mpg123.h.orig	Fri May 28 23:51:58 1999
++++ mpg123.h
+@@ -56,8 +56,115 @@
  #  define real float
  #elif defined(REAL_IS_LONG_DOUBLE)
  #  define real long double
@@ -19,6 +19,84 @@ $NetBSD: patch-aj,v 1.1 1999/10/12 04:43:15 simonb Exp $
 +# define REAL_TO_SHORT(x)	((x) >> REAL_RADIX)
 +# define REAL_MUL(x, y)		(((long long)(x) * (long long)(y)) >> REAL_RADIX)
 +
++#elif defined(REAL_IS_FIXED_ARMV4)
++# define REAL_IS_FIXED
++# define real long long
++
++# define REAL_RADIX		32
++# define REAL_FACTOR		(4294967296.0)
++
++# define REAL_PLUS_32767	( 32767LL << REAL_RADIX )
++# define REAL_MINUS_32768	( -32768LL << REAL_RADIX )
++
++# define DOUBLE_TO_REAL(x)	((long long)((x) * REAL_FACTOR))
++# define REAL_TO_SHORT(x)	((x) >> REAL_RADIX)
++
++/*
++ * "Reals" are signed, 15 bits before the point, 32 bits after.
++ * Multiplication is done like so:
++ *
++ *   if |x| < 2^8, shift x >> 7, else shift x >>15
++ *   if |y| < 2^8, shift y >> 7, else shift y >>15
++ *   multiply x * y (just avoids clipping MSB)
++ *   shift the result >> 2, 10 or 18 bits, to make a total shift of 32 bits
++ *
++ * I'd do it in C, but gcc optimises it rather poorly (2K stack
++ * frames, anyone?).  Hence the bit of (ARMv4) asm magic below.
++ * (Unfortunately, the CLZ opcode, which would make for a much neater
++ * solution to this, doesn't appear until ARMv5)
++ *
++ * (tjd April 2002)
++ */
++
++# define REAL_MUL(x, y)		real_mul((x),(y))
++
++static __inline__ long long real_mul(long long x, long long y) {
++    const register long long _x asm("r0") = (x);
++    const register long long _y asm("r2") = (y);
++    register long long rv asm("r0");
++    __asm__ __volatile__ (
++"@ we'll need to shift the result at least >> 2\n"
++"	mov	r4, #2\n"
++"@ both args >> 7,\n"
++"	mov	r0, r0, lsr #7\n"
++"	mov	r2, r2, lsr #7\n"
++"	orr	r0, r0, r1, lsl #25\n"
++"	orr	r2, r2, r3, lsl #25\n"
++"	mov	r1, r1, asr #7\n"
++"	mov	r3, r3, asr #7\n"
++"@ check if x needs further shiftage\n"
++"	cmp	r1, #0\n"
++"	cmnne	r1, #1\n"
++"	addeq	r4, r4, #8\n"
++"@ shift x if needed\n"
++"	movne	r0, r0, lsr #8\n"
++"	orrne	r0, r0, r1, lsl #24\n"
++"	movne	r1, r1, asr #8\n"
++"@ check if y needs further shiftage\n"
++"	cmp	r3, #0\n"
++"	cmnne	r3, #1\n"
++"	addeq	r4, r4, #8\n"
++"@ shift y if needed\n"
++"	movne	r2, r2, lsr #8\n"
++"	orrne	r2, r2, r3, lsl #24\n"
++"	movne	r3, r3, asr #8\n"
++"@ multiply (ignoring overflow)\n"
++"	mul	r3, r0, r3\n"
++"	mla	r3, r1, r2, r3\n"
++"	umull	r0, r1, r2, r0\n"
++"	add	r1, r1, r3\n"
++"@ shift the result the appropriate amount\n"
++"	rsb	r3, r4, #32\n"
++"	mov	r0, r0, lsr r4\n"
++"	orr	r0, r0, r1, lsl r3\n"
++"	mov	r1, r1, asr r4\n"
++	: "=r" (rv)
++	: "0" (_x), "r" (_y)
++	: "r1", "r3", "r4", "cc");
++    return rv;
++}
++
  #else
  #  define real double
 +#endif