* ALL: support for ARM-specific optimizations, code by Koji Agawa.

AUTHORS

@@ -9,8 +9,10 @@ Sam Hocevar <sam@zoy.org> and Christophe Massiot <massiot@via.ecp.fr>
 are the current maintainers.
 
 Other contributors include:
+	Koji Agawa <i (AT) atty (DOT) jp> - ARM code
 	Bruno Barreyra <barreyra@ufl.edu> - build fixes
 	Gildas Bazin <gbazin@netcourrier.com> - mingw32 port
+	Diego Biurrun <diego@biurrun.de> - portability fixes
 	Alexander W. Chin <alexc@newt.phys.unsw.edu.au> - progressive_seq fix
 	Stephen Crowley <stephenc@dns2.digitalpassage.com> - build fixes
 	Didier Gautheron <dgautheron@magic.fr> - bug fixes

configure.ac

@@ -17,6 +17,7 @@ AM_MAINTAINER_MODE
 dnl Checks for compiler
 AC_PROG_CC
 AC_PROG_GCC_TRADITIONAL
+AM_PROG_AS
 
 dnl Checks for headers. We do this before the CC-specific section because
 dnl autoconf generates tests for generic headers before the first header test.
@@ -58,6 +59,7 @@ elif test x"$GCC" = x"yes"; then
     AC_TRY_CFLAGS([$TRY_CFLAGS $CFLAGS],[OPT_CFLAGS="$TRY_CFLAGS"])
 
     dnl arch-specific flags
+    arm_conditional=false
     case "$host" in
     i?86-* | k?-* | x86_64-* | amd64-*)
 	AC_DEFINE([ARCH_X86],,[x86 architecture])
@@ -99,6 +101,9 @@ elif test x"$GCC" = x"yes"; then
 	AC_TRY_CFLAGS([$TRY_CFLAGS $CFLAGS],[OPT_CFLAGS="$TRY_CFLAGS"]);;
     alpha*)
 	AC_DEFINE([ARCH_ALPHA],,[alpha architecture]);;
+    arm*)
+	arm_conditional=:
+	AC_DEFINE([ARCH_ARM],,[ARM architecture]);;
     esac
 elif test x"$CC" = x"tendracc"; then
     dnl TenDRA portability checking compiler
@@ -118,6 +123,8 @@ else
     esac
 fi
 
+AM_CONDITIONAL(ARCH_ARM, ${arm_conditional})
+
 dnl Checks for libtool - this must be done after we set cflags
 AC_DISABLE_SHARED
 AC_LIBTOOL_WIN32_DLL

include/mpeg2.h

@@ -162,6 +162,7 @@ void mpeg2_custom_fbuf (mpeg2dec_t * mpeg2dec, int custom_fbuf);
 #define MPEG2_ACCEL_ALPHA_MVI 2
 #define MPEG2_ACCEL_SPARC_VIS 1
 #define MPEG2_ACCEL_SPARC_VIS2 2
+#define MPEG2_ACCEL_ARM 1
 #define MPEG2_ACCEL_DETECT 0x80000000
 
 uint32_t mpeg2_accel (uint32_t accel);

libmpeg2/Makefile.am

@@ -11,8 +11,11 @@ noinst_LTLIBRARIES = libmpeg2arch.la
 libmpeg2arch_la_SOURCES = motion_comp_mmx.c idct_mmx.c \
 			  motion_comp_altivec.c idct_altivec.c \
 			  motion_comp_alpha.c idct_alpha.c \
-			  motion_comp_vis.c \
+			  motion_comp_vis.c motion_comp_arm.c \
 			  cpu_accel.c cpu_state.c
+if ARCH_ARM
+libmpeg2arch_la_SOURCES += motion_comp_arm_s.S
+endif
 libmpeg2arch_la_CFLAGS = $(OPT_CFLAGS) $(ARCH_OPT_CFLAGS) $(LIBMPEG2_CFLAGS)
 
 pkgconfigdir = $(libdir)/pkgconfig

libmpeg2/motion_comp.c

@@ -56,6 +56,11 @@ void mpeg2_mc_init (uint32_t accel)
     if (accel & MPEG2_ACCEL_SPARC_VIS)
 	mpeg2_mc = mpeg2_mc_vis;
     else
+#endif
+#ifdef ARCH_ARM
+    if (accel & MPEG2_ACCEL_ARM) {
+	mpeg2_mc = mpeg2_mc_arm;
+    } else
 #endif
 	mpeg2_mc = mpeg2_mc_c;
 }

libmpeg2/motion_comp_arm.c

+/*
+ * motion_comp_arm.c
+ * Copyright (C) 2004 AGAWA Koji <i (AT) atty (DOT) jp>
+ *
+ * This file is part of mpeg2dec, a free MPEG-2 video stream decoder.
+ * See http://libmpeg2.sourceforge.net/ for updates.
+ *
+ * mpeg2dec is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * mpeg2dec is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with mpeg2dec; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "config.h"
+
+#ifdef ARCH_ARM
+
+#include <inttypes.h>
+
+#include "mpeg2.h"
+#include "attributes.h"
+#include "mpeg2_internal.h"
+
+#define avg2(a,b)     ((a+b+1)>>1)
+#define avg4(a,b,c,d) ((a+b+c+d+2)>>2)
+
+#define predict_o(i)  (ref[i])
+#define predict_x(i)  (avg2 (ref[i], ref[i+1]))
+#define predict_y(i)  (avg2 (ref[i], (ref+stride)[i]))
+#define predict_xy(i) (avg4 (ref[i], ref[i+1], \
+			     (ref+stride)[i], (ref+stride)[i+1]))
+
+#define put(predictor,i) dest[i] = predictor (i)
+#define avg(predictor,i) dest[i] = avg2 (predictor (i), dest[i])
+
+/* mc function template */
+
+#define MC_FUNC(op,xy)							\
+static void inline MC_##op##_##xy##_16_c (uint8_t * dest, const uint8_t * ref,	\
+				   const int stride, int height)	\
+{									\
+    do {								\
+	op (predict_##xy, 0);						\
+	op (predict_##xy, 1);						\
+	op (predict_##xy, 2);						\
+	op (predict_##xy, 3);						\
+	op (predict_##xy, 4);						\
+	op (predict_##xy, 5);						\
+	op (predict_##xy, 6);						\
+	op (predict_##xy, 7);						\
+	op (predict_##xy, 8);						\
+	op (predict_##xy, 9);						\
+	op (predict_##xy, 10);						\
+	op (predict_##xy, 11);						\
+	op (predict_##xy, 12);						\
+	op (predict_##xy, 13);						\
+	op (predict_##xy, 14);						\
+	op (predict_##xy, 15);						\
+	ref += stride;							\
+	dest += stride;							\
+    } while (--height);						\
+}									\
+static void MC_##op##_##xy##_8_c (uint8_t * dest, const uint8_t * ref,	\
+				  const int stride, int height)		\
+{									\
+    do {								\
+	op (predict_##xy, 0);						\
+	op (predict_##xy, 1);						\
+	op (predict_##xy, 2);						\
+	op (predict_##xy, 3);						\
+	op (predict_##xy, 4);						\
+	op (predict_##xy, 5);						\
+	op (predict_##xy, 6);						\
+	op (predict_##xy, 7);						\
+	ref += stride;							\
+	dest += stride;							\
+    } while (--height);						\
+}									\
+/* definitions of the actual mc functions */
+
+MC_FUNC (avg,o)
+MC_FUNC (avg,x)
+MC_FUNC (put,y)
+MC_FUNC (avg,y)
+MC_FUNC (put,xy)
+MC_FUNC (avg,xy)
+
+
+extern void MC_put_o_16_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height);
+
+extern void MC_put_x_16_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height);
+
+
+static void MC_put_y_16_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_put_y_16_c(dest, ref, stride, height);
+}
+
+static void MC_put_xy_16_arm (uint8_t * dest, const uint8_t * ref,
+			       int stride, int height)
+{
+    MC_put_xy_16_c(dest, ref, stride, height);
+}
+
+extern void MC_put_o_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height);
+
+extern void MC_put_x_8_arm (uint8_t * dest, const uint8_t * ref,
+			    int stride, int height);
+
+static void MC_put_y_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    MC_put_y_8_c(dest, ref, stride, height);
+}
+
+static void MC_put_xy_8_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_put_xy_8_c(dest, ref, stride, height);
+}
+
+static void MC_avg_o_16_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_avg_o_16_c(dest, ref, stride, height);
+}
+
+static void MC_avg_x_16_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_avg_x_16_c(dest, ref, stride, height);
+}
+
+static void MC_avg_y_16_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_avg_y_16_c(dest, ref, stride, height);
+}
+
+static void MC_avg_xy_16_arm (uint8_t * dest, const uint8_t * ref,
+			       int stride, int height)
+{
+    MC_avg_xy_16_c(dest, ref, stride, height);
+}
+
+static void MC_avg_o_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    MC_avg_o_8_c(dest, ref, stride, height);
+}
+
+static void MC_avg_x_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    MC_avg_x_8_c(dest, ref, stride, height);
+}
+
+static void MC_avg_y_8_arm (uint8_t * dest, const uint8_t * ref,
+			     int stride, int height)
+{
+    MC_avg_y_8_c(dest, ref, stride, height);
+}
+
+static void MC_avg_xy_8_arm (uint8_t * dest, const uint8_t * ref,
+			      int stride, int height)
+{
+    MC_avg_xy_8_c(dest, ref, stride, height);
+}
+
+MPEG2_MC_EXTERN (arm)
+
+#endif

libmpeg2/motion_comp_arm_s.S

+@ motion_comp_arm_s.S
+@ Copyright (C) 2004 AGAWA Koji <i (AT) atty (DOT) jp>
+@
+@ This file is part of mpeg2dec, a free MPEG-2 video stream decoder.
+@ See http://libmpeg2.sourceforge.net/ for updates.
+@
+@ mpeg2dec is free software; you can redistribute it and/or modify
+@ it under the terms of the GNU General Public License as published by
+@ the Free Software Foundation; either version 2 of the License, or
+@ (at your option) any later version.
+@
+@ mpeg2dec is distributed in the hope that it will be useful,
+@ but WITHOUT ANY WARRANTY; without even the implied warranty of
+@ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+@ GNU General Public License for more details.
+@
+@ You should have received a copy of the GNU General Public License
+@ along with mpeg2dec; if not, write to the Free Software
+@ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+
+	.text
+
+@ ----------------------------------------------------------------
+	.align
+	.global MC_put_o_16_arm
+MC_put_o_16_arm:
+	@@ void func(uint8_t * dest, const uint8_t * ref, int stride, int height)
+	pld [r1]
+        stmfd sp!, {r4-r11, lr} @ R14 is also called LR
+	and r4, r1, #3
+	adr r5, MC_put_o_16_arm_align_jt
+	add r5, r5, r4, lsl #2
+	ldr pc, [r5]
+
+MC_put_o_16_arm_align0:
+	ldmia r1, {r4-r7}
+	add r1, r1, r2
+	pld [r1]
+	stmia r0, {r4-r7}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne MC_put_o_16_arm_align0
+        ldmfd sp!, {r4-r11, pc} @@ update PC with LR content.
+
+.macro	PROC shift
+	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	mov r9, r4, lsr #(\shift)
+	pld [r1]
+	mov r10, r5, lsr #(\shift)
+	orr r9, r9, r5, lsl #(32-\shift)
+	mov r11, r6, lsr #(\shift)
+	orr r10, r10, r6, lsl #(32-\shift)
+	mov r12, r7, lsr #(\shift)
+	orr r11, r11, r7, lsl #(32-\shift)
+	orr r12, r12, r8, lsl #(32-\shift)
+	stmia r0, {r9-r12}
+	subs r3, r3, #1
+	add r0, r0, r2
+.endm
+
+MC_put_o_16_arm_align1:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC(8)
+	bne 1b
+        ldmfd sp!, {r4-r11, pc} @@ update PC with LR content.
+MC_put_o_16_arm_align2:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC(16)
+	bne 1b
+        ldmfd sp!, {r4-r11, pc} @@ update PC with LR content.
+MC_put_o_16_arm_align3:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC(24)
+	bne 1b
+        ldmfd sp!, {r4-r11, pc} @@ update PC with LR content.
+MC_put_o_16_arm_align_jt:
+	.word MC_put_o_16_arm_align0
+	.word MC_put_o_16_arm_align1
+	.word MC_put_o_16_arm_align2
+	.word MC_put_o_16_arm_align3
+
+@ ----------------------------------------------------------------
+	.align
+	.global MC_put_o_8_arm
+MC_put_o_8_arm:
+	@@ void func(uint8_t * dest, const uint8_t * ref, int stride, int height)
+	pld [r1]
+        stmfd sp!, {r4-r10, lr} @ R14 is also called LR
+	and r4, r1, #3
+	adr r5, MC_put_o_8_arm_align_jt
+	add r5, r5, r4, lsl #2
+	ldr pc, [r5]
+MC_put_o_8_arm_align0:
+	ldmia r1, {r4-r5}
+	add r1, r1, r2
+	pld [r1]
+	stmia r0, {r4-r5}
+	add r0, r0, r2
+	subs r3, r3, #1
+	bne MC_put_o_8_arm_align0
+        ldmfd sp!, {r4-r10, pc} @@ update PC with LR content.
+
+.macro	PROC8 shift
+	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	mov r9, r4, lsr #(\shift)
+	pld [r1]
+	mov r10, r5, lsr #(\shift)
+	orr r9, r9, r5, lsl #(32-\shift)
+	orr r10, r10, r6, lsl #(32-\shift)
+	stmia r0, {r9-r10}
+	subs r3, r3, #1
+	add r0, r0, r2
+.endm
+
+MC_put_o_8_arm_align1:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC8(8)
+	bne 1b
+        ldmfd sp!, {r4-r10, pc} @@ update PC with LR content.
+
+MC_put_o_8_arm_align2:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC8(16)
+	bne 1b
+        ldmfd sp!, {r4-r10, pc} @@ update PC with LR content.
+
+MC_put_o_8_arm_align3:
+	and r1, r1, #0xFFFFFFFC
+1:	PROC8(24)
+	bne 1b
+        ldmfd sp!, {r4-r10, pc} @@ update PC with LR content.
+
+MC_put_o_8_arm_align_jt:
+	.word MC_put_o_8_arm_align0
+	.word MC_put_o_8_arm_align1
+	.word MC_put_o_8_arm_align2
+	.word MC_put_o_8_arm_align3
+
+@ ----------------------------------------------------------------
+.macro	AVG_PW rW1, rW2
+	mov \rW2, \rW2, lsl #24
+	orr \rW2, \rW2, \rW1, lsr #8
+	eor r9, \rW1, \rW2
+	and \rW2, \rW1, \rW2
+	and r10, r9, r12
+	add \rW2, \rW2, r10, lsr #1
+	and r10, r9, r11
+	add \rW2, \rW2, r10
+.endm
+
+	.align
+	.global MC_put_x_16_arm
+MC_put_x_16_arm:
+	@@ void func(uint8_t * dest, const uint8_t * ref, int stride, int height)
+	pld [r1]
+        stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+	and r4, r1, #3
+	adr r5, MC_put_x_16_arm_align_jt
+	ldr r11, [r5]
+	mvn r12, r11
+	add r5, r5, r4, lsl #2
+	ldr pc, [r5, #4]
+
+.macro	ADJ_ALIGN_QW shift, R0, R1, R2, R3, R4
+	mov \R0, \R0, lsr #(\shift)
+	orr \R0, \R0, \R1, lsl #(32 - \shift)
+	mov \R1, \R1, lsr #(\shift)
+	orr \R1, \R1, \R2, lsl #(32 - \shift)
+	mov \R2, \R2, lsr #(\shift)
+	orr \R2, \R2, \R3, lsl #(32 - \shift)
+	mov \R3, \R3, lsr #(\shift)
+	orr \R3, \R3, \R4, lsl #(32 - \shift)
+	mov \R4, \R4, lsr #(\shift)
+@	and \R4, \R4, #0xFF
+.endm
+
+MC_put_x_16_arm_align0:
+	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	pld [r1]
+	AVG_PW r7, r8
+	AVG_PW r6, r7
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r8}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne MC_put_x_16_arm_align0
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_16_arm_align1:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_QW 8, r4, r5, r6, r7, r8
+	AVG_PW r7, r8
+	AVG_PW r6, r7
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r8}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_16_arm_align2:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_QW 16, r4, r5, r6, r7, r8
+	AVG_PW r7, r8
+	AVG_PW r6, r7
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r8}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_16_arm_align3:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r8}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_QW 24, r4, r5, r6, r7, r8
+	AVG_PW r7, r8
+	AVG_PW r6, r7
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r8}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_16_arm_align_jt:
+	.word 0x01010101
+	.word MC_put_x_16_arm_align0
+	.word MC_put_x_16_arm_align1
+	.word MC_put_x_16_arm_align2
+	.word MC_put_x_16_arm_align3
+
+@ ----------------------------------------------------------------
+	.align
+	.global MC_put_x_8_arm
+MC_put_x_8_arm:
+	@@ void func(uint8_t * dest, const uint8_t * ref, int stride, int height)
+	pld [r1]
+        stmfd sp!, {r4-r11,lr} @ R14 is also called LR
+	and r4, r1, #3
+	adr r5, MC_put_x_8_arm_align_jt
+	ldr r11, [r5]
+	mvn r12, r11
+	add r5, r5, r4, lsl #2
+	ldr pc, [r5, #4]
+
+.macro	ADJ_ALIGN_DW shift, R0, R1, R2
+	mov \R0, \R0, lsr #(\shift)
+	orr \R0, \R0, \R1, lsl #(32 - \shift)
+	mov \R1, \R1, lsr #(\shift)
+	orr \R1, \R1, \R2, lsl #(32 - \shift)
+	mov \R2, \R2, lsr #(\shift)
+@	and \R4, \R4, #0xFF
+.endm
+
+MC_put_x_8_arm_align0:
+	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	pld [r1]
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r6}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne MC_put_x_8_arm_align0
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_8_arm_align1:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_DW 8, r4, r5, r6
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r6}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_8_arm_align2:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_DW 16, r4, r5, r6
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r6}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_8_arm_align3:
+	and r1, r1, #0xFFFFFFFC
+1:	ldmia r1, {r4-r6}
+	add r1, r1, r2
+	pld [r1]
+	ADJ_ALIGN_DW 24, r4, r5, r6
+	AVG_PW r5, r6
+	AVG_PW r4, r5
+	stmia r0, {r5-r6}
+	subs r3, r3, #1
+	add r0, r0, r2
+	bne 1b
+        ldmfd sp!, {r4-r11,pc} @@ update PC with LR content.
+MC_put_x_8_arm_align_jt:
+	.word 0x01010101
+	.word MC_put_x_8_arm_align0
+	.word MC_put_x_8_arm_align1
+	.word MC_put_x_8_arm_align2
+	.word MC_put_x_8_arm_align3

libmpeg2/mpeg2_internal.h

@@ -307,5 +307,6 @@ extern mpeg2_mc_t mpeg2_mc_3dnow;
 extern mpeg2_mc_t mpeg2_mc_altivec;
 extern mpeg2_mc_t mpeg2_mc_alpha;
 extern mpeg2_mc_t mpeg2_mc_vis;
+extern mpeg2_mc_t mpeg2_mc_arm;
 
 #endif /* LIBMPEG2_MPEG2_INTERNAL_H */