From 42d32cf53cd0aa0da7cf7a89c8b46adaf761936c Mon Sep 17 00:00:00 2001
From: Janne Grunau <janne-libav@jannau.net>
Date: Sat, 24 Sep 2011 13:05:55 +0200
Subject: [PATCH 1/5] rv34: NEON optimised inverse transform functions
Signed-off-by: Mans Rullgard <mans@mansr.com>
---
libavcodec/arm/Makefile | 6 ++
libavcodec/arm/rv34dsp_init_neon.c | 33 +++++++++
libavcodec/arm/rv34dsp_neon.S | 109 +++++++++++++++++++++++++++++
libavcodec/rv34dsp.c | 3 +
libavcodec/rv34dsp.h | 2 +
5 files changed, 153 insertions(+)
create mode 100644 libavcodec/arm/rv34dsp_init_neon.c
create mode 100644 libavcodec/arm/rv34dsp_neon.S
diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index 3374f0e2bd..9199faea3b 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -62,6 +62,12 @@ NEON-OBJS-$(CONFIG_AC3DSP) += arm/ac3dsp_neon.o
NEON-OBJS-$(CONFIG_DCA_DECODER) += arm/dcadsp_neon.o \
arm/synth_filter_neon.o \
+NEON-OBJS-$(CONFIG_RV30_DECODER) += arm/rv34dsp_init_neon.o \
+ arm/rv34dsp_neon.o \
+
+NEON-OBJS-$(CONFIG_RV40_DECODER) += arm/rv34dsp_init_neon.o \
+ arm/rv34dsp_neon.o \
+
NEON-OBJS-$(CONFIG_VP3_DECODER) += arm/vp3dsp_neon.o
NEON-OBJS-$(CONFIG_VP5_DECODER) += arm/vp56dsp_neon.o \
diff --git a/libavcodec/arm/rv34dsp_init_neon.c b/libavcodec/arm/rv34dsp_init_neon.c
new file mode 100644
index 0000000000..9a09fde7a9
--- /dev/null
+++ b/libavcodec/arm/rv34dsp_init_neon.c
@@ -0,0 +1,33 @@
+/*
+ * Copyright (c) 2011 Janne Grunau <janne-libav@jannau.net>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "libavcodec/avcodec.h"
+#include "libavcodec/rv34dsp.h"
+
+void ff_rv34_inv_transform_neon(DCTELEM *block);
+void ff_rv34_inv_transform_noround_neon(DCTELEM *block);
+
+void ff_rv34dsp_init_neon(RV34DSPContext *c, DSPContext* dsp)
+{
+ c->rv34_inv_transform_tab[0] = ff_rv34_inv_transform_neon;
+ c->rv34_inv_transform_tab[1] = ff_rv34_inv_transform_noround_neon;
+}
diff --git a/libavcodec/arm/rv34dsp_neon.S b/libavcodec/arm/rv34dsp_neon.S
new file mode 100644
index 0000000000..f700f5c321
--- /dev/null
+++ b/libavcodec/arm/rv34dsp_neon.S
@@ -0,0 +1,109 @@
+/*
+ * Copyright (c) 2011 Janne Grunau <janne-libav@jannau.net>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "asm.S"
+
+.macro rv34_inv_transform
+ mov r1, #16
+ vld1.16 {d28}, [r0,:64], r1 @ block[i+8*0]
+ vld1.16 {d29}, [r0,:64], r1 @ block[i+8*1]
+ vld1.16 {d30}, [r0,:64], r1 @ block[i+8*2]
+ vld1.16 {d31}, [r0,:64], r1 @ block[i+8*3]
+ vmov.s16 d0, #13
+ vshll.s16 q12, d29, #3
+ vshll.s16 q13, d29, #4
+ vshll.s16 q9, d31, #3
+ vshll.s16 q1, d31, #4
+ vmull.s16 q10, d28, d0
+ vmlal.s16 q10, d30, d0
+ vmull.s16 q11, d28, d0
+ vmlsl.s16 q11, d30, d0
+ vsubw.s16 q12, q12, d29 @ z2 = block[i+8*1]*7
+ vaddw.s16 q13, q13, d29 @ z3 = block[i+8*1]*17
+ vsubw.s16 q9, q9, d31
+ vaddw.s16 q1, q1, d31
+ vadd.s32 q13, q13, q9 @ z3 = 17*block[i+8*1] + 7*block[i+8*3]
+ vsub.s32 q12, q12, q1 @ z2 = 7*block[i+8*1] - 17*block[i+8*3]
+ vadd.s32 q1, q10, q13 @ z0 + z3
+ vadd.s32 q2, q11, q12 @ z1 + z2
+ vsub.s32 q8, q10, q13 @ z0 - z3
+ vsub.s32 q3, q11, q12 @ z1 - z2
+ vtrn.32 q1, q2
+ vtrn.32 q3, q8
+ vswp d3, d6
+ vswp d5, d16
+ vmov.s32 d0, #13
+ vadd.s32 q10, q1, q3
+ vsub.s32 q11, q1, q3
+ vshl.s32 q12, q2, #3
+ vshl.s32 q9, q2, #4
+ vmul.s32 q13, q11, d0[0]
+ vshl.s32 q11, q8, #4
+ vadd.s32 q9, q9, q2
+ vshl.s32 q15, q8, #3
+ vsub.s32 q12, q12, q2
+ vadd.s32 q11, q11, q8
+ vmul.s32 q14, q10, d0[0]
+ vsub.s32 q8, q15, q8
+ vsub.s32 q12, q12, q11
+ vadd.s32 q9, q9, q8
+ vadd.s32 q2, q13, q12 @ z1 + z2
+ vadd.s32 q1, q14, q9 @ z0 + z3
+ vsub.s32 q3, q13, q12 @ z1 - z2
+ vsub.s32 q15, q14, q9 @ z0 - z3
+.endm
+
+/* void ff_rv34_inv_transform_neon(DCTELEM *block); */
+function ff_rv34_inv_transform_neon, export=1
+ mov r2, r0
+ rv34_inv_transform
+ vrshrn.s32 d1, q2, #10 @ (z1 + z2) >> 10
+ vrshrn.s32 d0, q1, #10 @ (z0 + z3) >> 10
+ vrshrn.s32 d2, q3, #10 @ (z1 - z2) >> 10
+ vrshrn.s32 d3, q15, #10 @ (z0 - z3) >> 10
+ vst4.16 {d0[0], d1[0], d2[0], d3[0]}, [r2,:64], r1
+ vst4.16 {d0[1], d1[1], d2[1], d3[1]}, [r2,:64], r1
+ vst4.16 {d0[2], d1[2], d2[2], d3[2]}, [r2,:64], r1
+ vst4.16 {d0[3], d1[3], d2[3], d3[3]}, [r2,:64], r1
+ bx lr
+endfunc
+
+/* void rv34_inv_transform_noround_neon(DCTELEM *block); */
+function ff_rv34_inv_transform_noround_neon, export=1
+ mov r2, r0
+ rv34_inv_transform
+ vshl.s32 q11, q2, #1
+ vshl.s32 q10, q1, #1
+ vshl.s32 q12, q3, #1
+ vshl.s32 q13, q15, #1
+ vadd.s32 q11, q11, q2
+ vadd.s32 q10, q10, q1
+ vadd.s32 q12, q12, q3
+ vadd.s32 q13, q13, q15
+ vshrn.s32 d0, q10, #11 @ (z0 + z3)*3 >> 11
+ vshrn.s32 d1, q11, #11 @ (z1 + z2)*3 >> 11
+ vshrn.s32 d2, q12, #11 @ (z1 - z2)*3 >> 11
+ vshrn.s32 d3, q13, #11 @ (z0 - z3)*3 >> 11
+ vst4.16 {d0[0], d1[0], d2[0], d3[0]}, [r2,:64], r1
+ vst4.16 {d0[1], d1[1], d2[1], d3[1]}, [r2,:64], r1
+ vst4.16 {d0[2], d1[2], d2[2], d3[2]}, [r2,:64], r1
+ vst4.16 {d0[3], d1[3], d2[3], d3[3]}, [r2,:64], r1
+ bx lr
+endfunc
diff --git a/libavcodec/rv34dsp.c b/libavcodec/rv34dsp.c
index 59038a7a31..1f4cea8544 100644
--- a/libavcodec/rv34dsp.c
+++ b/libavcodec/rv34dsp.c
@@ -103,4 +103,7 @@ static void rv34_inv_transform_noround_c(DCTELEM *block){
av_cold void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp) {
c->rv34_inv_transform_tab[0] = rv34_inv_transform_c;
c->rv34_inv_transform_tab[1] = rv34_inv_transform_noround_c;
+
+ if (HAVE_NEON)
+ ff_rv34dsp_init_neon(c, dsp);
}
diff --git a/libavcodec/rv34dsp.h b/libavcodec/rv34dsp.h
index 4ade05060f..a1636e6eb5 100644
--- a/libavcodec/rv34dsp.h
+++ b/libavcodec/rv34dsp.h
@@ -56,6 +56,8 @@ void ff_rv30dsp_init(RV34DSPContext *c, DSPContext* dsp);
void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp);
void ff_rv40dsp_init(RV34DSPContext *c, DSPContext* dsp);
+void ff_rv34dsp_init_neon(RV34DSPContext *c, DSPContext *dsp);
+
void ff_rv40dsp_init_x86(RV34DSPContext *c, DSPContext *dsp);
#endif /* AVCODEC_RV34DSP_H */
From f054a82727728e813861851648e109cd24574178 Mon Sep 17 00:00:00 2001
From: Mans Rullgard <mans@mansr.com>
Date: Tue, 6 Dec 2011 12:44:05 +0000
Subject: [PATCH 2/5] ARM: move NEON H264 chroma mc to a separate file
This allows sharing code with the rv40 version of these functions.
Signed-off-by: Mans Rullgard <mans@mansr.com>
---
libavcodec/arm/Makefile | 1 +
libavcodec/arm/h264cmc_neon.S | 360 ++++++++++++++++++++++++++++++++++
libavcodec/arm/h264dsp_neon.S | 339 --------------------------------
3 files changed, 361 insertions(+), 339 deletions(-)
create mode 100644 libavcodec/arm/h264cmc_neon.S
diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index 9199faea3b..c125a59078 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -54,6 +54,7 @@ NEON-OBJS-$(CONFIG_RDFT) += arm/rdft_neon.o \
NEON-OBJS-$(CONFIG_H264DSP) += arm/h264dsp_neon.o \
arm/h264idct_neon.o \
+ arm/h264cmc_neon.o \
NEON-OBJS-$(CONFIG_H264PRED) += arm/h264pred_neon.o \
diff --git a/libavcodec/arm/h264cmc_neon.S b/libavcodec/arm/h264cmc_neon.S
new file mode 100644
index 0000000000..e10adaca10
--- /dev/null
+++ b/libavcodec/arm/h264cmc_neon.S
@@ -0,0 +1,360 @@
+/*
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "asm.S"
+
+/* chroma_mc8(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) */
+.macro h264_chroma_mc8 type
+function ff_\type\()_h264_chroma_mc8_neon, export=1
+ push {r4-r7, lr}
+ ldrd r4, [sp, #20]
+ .ifc \type,avg
+ mov lr, r0
+ .endif
+ pld [r1]
+ pld [r1, r2]
+
+A muls r7, r4, r5
+T mul r7, r4, r5
+T cmp r7, #0
+ rsb r6, r7, r5, lsl #3
+ rsb r12, r7, r4, lsl #3
+ sub r4, r7, r4, lsl #3
+ sub r4, r4, r5, lsl #3
+ add r4, r4, #64
+
+ beq 2f
+
+ add r5, r1, r2
+
+ vdup.8 d0, r4
+ lsl r4, r2, #1
+ vdup.8 d1, r12
+ vld1.8 {d4, d5}, [r1], r4
+ vdup.8 d2, r6
+ vld1.8 {d6, d7}, [r5], r4
+ vdup.8 d3, r7
+
+ vext.8 d5, d4, d5, #1
+ vext.8 d7, d6, d7, #1
+
+1: pld [r5]
+ vmull.u8 q8, d4, d0
+ vmlal.u8 q8, d5, d1
+ vld1.8 {d4, d5}, [r1], r4
+ vmlal.u8 q8, d6, d2
+ vext.8 d5, d4, d5, #1
+ vmlal.u8 q8, d7, d3
+ vmull.u8 q9, d6, d0
+ subs r3, r3, #2
+ vmlal.u8 q9, d7, d1
+ vmlal.u8 q9, d4, d2
+ vmlal.u8 q9, d5, d3
+ vrshrn.u16 d16, q8, #6
+ vld1.8 {d6, d7}, [r5], r4
+ pld [r1]
+ vrshrn.u16 d17, q9, #6
+ .ifc \type,avg
+ vld1.8 {d20}, [lr,:64], r2
+ vld1.8 {d21}, [lr,:64], r2
+ vrhadd.u8 q8, q8, q10
+ .endif
+ vext.8 d7, d6, d7, #1
+ vst1.8 {d16}, [r0,:64], r2
+ vst1.8 {d17}, [r0,:64], r2
+ bgt 1b
+
+ pop {r4-r7, pc}
+
+2: tst r6, r6
+ add r12, r12, r6
+ vdup.8 d0, r4
+ vdup.8 d1, r12
+
+ beq 4f
+
+ add r5, r1, r2
+ lsl r4, r2, #1
+ vld1.8 {d4}, [r1], r4
+ vld1.8 {d6}, [r5], r4
+
+3: pld [r5]
+ vmull.u8 q8, d4, d0
+ vmlal.u8 q8, d6, d1
+ vld1.8 {d4}, [r1], r4
+ vmull.u8 q9, d6, d0
+ vmlal.u8 q9, d4, d1
+ vld1.8 {d6}, [r5], r4
+ vrshrn.u16 d16, q8, #6
+ vrshrn.u16 d17, q9, #6
+ .ifc \type,avg
+ vld1.8 {d20}, [lr,:64], r2
+ vld1.8 {d21}, [lr,:64], r2
+ vrhadd.u8 q8, q8, q10
+ .endif
+ subs r3, r3, #2
+ pld [r1]
+ vst1.8 {d16}, [r0,:64], r2
+ vst1.8 {d17}, [r0,:64], r2
+ bgt 3b
+
+ pop {r4-r7, pc}
+
+4: vld1.8 {d4, d5}, [r1], r2
+ vld1.8 {d6, d7}, [r1], r2
+ vext.8 d5, d4, d5, #1
+ vext.8 d7, d6, d7, #1
+
+5: pld [r1]
+ subs r3, r3, #2
+ vmull.u8 q8, d4, d0
+ vmlal.u8 q8, d5, d1
+ vld1.8 {d4, d5}, [r1], r2
+ vmull.u8 q9, d6, d0
+ vmlal.u8 q9, d7, d1
+ pld [r1]
+ vext.8 d5, d4, d5, #1
+ vrshrn.u16 d16, q8, #6
+ vrshrn.u16 d17, q9, #6
+ .ifc \type,avg
+ vld1.8 {d20}, [lr,:64], r2
+ vld1.8 {d21}, [lr,:64], r2
+ vrhadd.u8 q8, q8, q10
+ .endif
+ vld1.8 {d6, d7}, [r1], r2
+ vext.8 d7, d6, d7, #1
+ vst1.8 {d16}, [r0,:64], r2
+ vst1.8 {d17}, [r0,:64], r2
+ bgt 5b
+
+ pop {r4-r7, pc}
+endfunc
+.endm
+
+/* chroma_mc4(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) */
+.macro h264_chroma_mc4 type
+function ff_\type\()_h264_chroma_mc4_neon, export=1
+ push {r4-r7, lr}
+ ldrd r4, [sp, #20]
+ .ifc \type,avg
+ mov lr, r0
+ .endif
+ pld [r1]
+ pld [r1, r2]
+
+A muls r7, r4, r5
+T mul r7, r4, r5
+T cmp r7, #0
+ rsb r6, r7, r5, lsl #3
+ rsb r12, r7, r4, lsl #3
+ sub r4, r7, r4, lsl #3
+ sub r4, r4, r5, lsl #3
+ add r4, r4, #64
+
+ beq 2f
+
+ add r5, r1, r2
+
+ vdup.8 d0, r4
+ lsl r4, r2, #1
+ vdup.8 d1, r12
+ vld1.8 {d4}, [r1], r4
+ vdup.8 d2, r6
+ vld1.8 {d6}, [r5], r4
+ vdup.8 d3, r7
+
+ vext.8 d5, d4, d5, #1
+ vext.8 d7, d6, d7, #1
+ vtrn.32 d4, d5
+ vtrn.32 d6, d7
+
+ vtrn.32 d0, d1
+ vtrn.32 d2, d3
+
+1: pld [r5]
+ vmull.u8 q8, d4, d0
+ vmlal.u8 q8, d6, d2
+ vld1.8 {d4}, [r1], r4
+ vext.8 d5, d4, d5, #1
+ vtrn.32 d4, d5
+ vmull.u8 q9, d6, d0
+ vmlal.u8 q9, d4, d2
+ vld1.8 {d6}, [r5], r4
+ vadd.i16 d16, d16, d17
+ vadd.i16 d17, d18, d19
+ vrshrn.u16 d16, q8, #6
+ subs r3, r3, #2
+ pld [r1]
+ .ifc \type,avg
+ vld1.32 {d20[0]}, [lr,:32], r2
+ vld1.32 {d20[1]}, [lr,:32], r2
+ vrhadd.u8 d16, d16, d20
+ .endif
+ vext.8 d7, d6, d7, #1
+ vtrn.32 d6, d7
+ vst1.32 {d16[0]}, [r0,:32], r2
+ vst1.32 {d16[1]}, [r0,:32], r2
+ bgt 1b
+
+ pop {r4-r7, pc}
+
+2: tst r6, r6
+ add r12, r12, r6
+ vdup.8 d0, r4
+ vdup.8 d1, r12
+ vtrn.32 d0, d1
+
+ beq 4f
+
+ vext.32 d1, d0, d1, #1
+ add r5, r1, r2
+ lsl r4, r2, #1
+ vld1.32 {d4[0]}, [r1], r4
+ vld1.32 {d4[1]}, [r5], r4
+
+3: pld [r5]
+ vmull.u8 q8, d4, d0
+ vld1.32 {d4[0]}, [r1], r4
+ vmull.u8 q9, d4, d1
+ vld1.32 {d4[1]}, [r5], r4
+ vadd.i16 d16, d16, d17
+ vadd.i16 d17, d18, d19
+ vrshrn.u16 d16, q8, #6
+ .ifc \type,avg
+ vld1.32 {d20[0]}, [lr,:32], r2
+ vld1.32 {d20[1]}, [lr,:32], r2
+ vrhadd.u8 d16, d16, d20
+ .endif
+ subs r3, r3, #2
+ pld [r1]
+ vst1.32 {d16[0]}, [r0,:32], r2
+ vst1.32 {d16[1]}, [r0,:32], r2
+ bgt 3b
+
+ pop {r4-r7, pc}
+
+4: vld1.8 {d4}, [r1], r2
+ vld1.8 {d6}, [r1], r2
+ vext.8 d5, d4, d5, #1
+ vext.8 d7, d6, d7, #1
+ vtrn.32 d4, d5
+ vtrn.32 d6, d7
+
+5: vmull.u8 q8, d4, d0
+ vmull.u8 q9, d6, d0
+ subs r3, r3, #2
+ vld1.8 {d4}, [r1], r2
+ vext.8 d5, d4, d5, #1
+ vtrn.32 d4, d5
+ vadd.i16 d16, d16, d17
+ vadd.i16 d17, d18, d19
+ pld [r1]
+ vrshrn.u16 d16, q8, #6
+ .ifc \type,avg
+ vld1.32 {d20[0]}, [lr,:32], r2
+ vld1.32 {d20[1]}, [lr,:32], r2
+ vrhadd.u8 d16, d16, d20
+ .endif
+ vld1.8 {d6}, [r1], r2
+ vext.8 d7, d6, d7, #1
+ vtrn.32 d6, d7
+ pld [r1]
+ vst1.32 {d16[0]}, [r0,:32], r2
+ vst1.32 {d16[1]}, [r0,:32], r2
+ bgt 5b
+
+ pop {r4-r7, pc}
+endfunc
+.endm
+
+.macro h264_chroma_mc2 type
+function ff_\type\()_h264_chroma_mc2_neon, export=1
+ push {r4-r6, lr}
+ ldr r4, [sp, #16]
+ ldr lr, [sp, #20]
+ pld [r1]
+ pld [r1, r2]
+ orrs r5, r4, lr
+ beq 2f
+
+ mul r5, r4, lr
+ rsb r6, r5, lr, lsl #3
+ rsb r12, r5, r4, lsl #3
+ sub r4, r5, r4, lsl #3
+ sub r4, r4, lr, lsl #3
+ add r4, r4, #64
+ vdup.8 d0, r4
+ vdup.8 d2, r12
+ vdup.8 d1, r6
+ vdup.8 d3, r5
+ vtrn.16 q0, q1
+1:
+ vld1.32 {d4[0]}, [r1], r2
+ vld1.32 {d4[1]}, [r1], r2
+ vrev64.32 d5, d4
+ vld1.32 {d5[1]}, [r1]
+ vext.8 q3, q2, q2, #1
+ vtrn.16 q2, q3
+ vmull.u8 q8, d4, d0
+ vmlal.u8 q8, d5, d1
+ .ifc \type,avg
+ vld1.16 {d18[0]}, [r0,:16], r2
+ vld1.16 {d18[1]}, [r0,:16]
+ sub r0, r0, r2
+ .endif
+ vtrn.32 d16, d17
+ vadd.i16 d16, d16, d17
+ vrshrn.u16 d16, q8, #6
+ .ifc \type,avg
+ vrhadd.u8 d16, d16, d18
+ .endif
+ vst1.16 {d16[0]}, [r0,:16], r2
+ vst1.16 {d16[1]}, [r0,:16], r2
+ subs r3, r3, #2
+ bgt 1b
+ pop {r4-r6, pc}
+2:
+ .ifc \type,put
+ ldrh_post r5, r1, r2
+ strh_post r5, r0, r2
+ ldrh_post r6, r1, r2
+ strh_post r6, r0, r2
+ .else
+ vld1.16 {d16[0]}, [r1], r2
+ vld1.16 {d16[1]}, [r1], r2
+ vld1.16 {d18[0]}, [r0,:16], r2
+ vld1.16 {d18[1]}, [r0,:16]
+ sub r0, r0, r2
+ vrhadd.u8 d16, d16, d18
+ vst1.16 {d16[0]}, [r0,:16], r2
+ vst1.16 {d16[1]}, [r0,:16], r2
+ .endif
+ subs r3, r3, #2
+ bgt 2b
+ pop {r4-r6, pc}
+endfunc
+.endm
+
+ h264_chroma_mc8 put
+ h264_chroma_mc8 avg
+ h264_chroma_mc4 put
+ h264_chroma_mc4 avg
+ h264_chroma_mc2 put
+ h264_chroma_mc2 avg
diff --git a/libavcodec/arm/h264dsp_neon.S b/libavcodec/arm/h264dsp_neon.S
index e23f7b149a..a4abf66494 100644
--- a/libavcodec/arm/h264dsp_neon.S
+++ b/libavcodec/arm/h264dsp_neon.S
@@ -21,345 +21,6 @@
#include "asm.S"
#include "neon.S"
-/* chroma_mc8(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) */
-.macro h264_chroma_mc8 type
-function ff_\type\()_h264_chroma_mc8_neon, export=1
- push {r4-r7, lr}
- ldrd r4, [sp, #20]
- .ifc \type,avg
- mov lr, r0
- .endif
- pld [r1]
- pld [r1, r2]
-
-A muls r7, r4, r5
-T mul r7, r4, r5
-T cmp r7, #0
- rsb r6, r7, r5, lsl #3
- rsb r12, r7, r4, lsl #3
- sub r4, r7, r4, lsl #3
- sub r4, r4, r5, lsl #3
- add r4, r4, #64
-
- beq 2f
-
- add r5, r1, r2
-
- vdup.8 d0, r4
- lsl r4, r2, #1
- vdup.8 d1, r12
- vld1.8 {d4, d5}, [r1], r4
- vdup.8 d2, r6
- vld1.8 {d6, d7}, [r5], r4
- vdup.8 d3, r7
-
- vext.8 d5, d4, d5, #1
- vext.8 d7, d6, d7, #1
-
-1: pld [r5]
- vmull.u8 q8, d4, d0
- vmlal.u8 q8, d5, d1
- vld1.8 {d4, d5}, [r1], r4
- vmlal.u8 q8, d6, d2
- vext.8 d5, d4, d5, #1
- vmlal.u8 q8, d7, d3
- vmull.u8 q9, d6, d0
- subs r3, r3, #2
- vmlal.u8 q9, d7, d1
- vmlal.u8 q9, d4, d2
- vmlal.u8 q9, d5, d3
- vrshrn.u16 d16, q8, #6
- vld1.8 {d6, d7}, [r5], r4
- pld [r1]
- vrshrn.u16 d17, q9, #6
- .ifc \type,avg
- vld1.8 {d20}, [lr,:64], r2
- vld1.8 {d21}, [lr,:64], r2
- vrhadd.u8 q8, q8, q10
- .endif
- vext.8 d7, d6, d7, #1
- vst1.8 {d16}, [r0,:64], r2
- vst1.8 {d17}, [r0,:64], r2
- bgt 1b
-
- pop {r4-r7, pc}
-
-2: tst r6, r6
- add r12, r12, r6
- vdup.8 d0, r4
- vdup.8 d1, r12
-
- beq 4f
-
- add r5, r1, r2
- lsl r4, r2, #1
- vld1.8 {d4}, [r1], r4
- vld1.8 {d6}, [r5], r4
-
-3: pld [r5]
- vmull.u8 q8, d4, d0
- vmlal.u8 q8, d6, d1
- vld1.8 {d4}, [r1], r4
- vmull.u8 q9, d6, d0
- vmlal.u8 q9, d4, d1
- vld1.8 {d6}, [r5], r4
- vrshrn.u16 d16, q8, #6
- vrshrn.u16 d17, q9, #6
- .ifc \type,avg
- vld1.8 {d20}, [lr,:64], r2
- vld1.8 {d21}, [lr,:64], r2
- vrhadd.u8 q8, q8, q10
- .endif
- subs r3, r3, #2
- pld [r1]
- vst1.8 {d16}, [r0,:64], r2
- vst1.8 {d17}, [r0,:64], r2
- bgt 3b
-
- pop {r4-r7, pc}
-
-4: vld1.8 {d4, d5}, [r1], r2
- vld1.8 {d6, d7}, [r1], r2
- vext.8 d5, d4, d5, #1
- vext.8 d7, d6, d7, #1
-
-5: pld [r1]
- subs r3, r3, #2
- vmull.u8 q8, d4, d0
- vmlal.u8 q8, d5, d1
- vld1.8 {d4, d5}, [r1], r2
- vmull.u8 q9, d6, d0
- vmlal.u8 q9, d7, d1
- pld [r1]
- vext.8 d5, d4, d5, #1
- vrshrn.u16 d16, q8, #6
- vrshrn.u16 d17, q9, #6
- .ifc \type,avg
- vld1.8 {d20}, [lr,:64], r2
- vld1.8 {d21}, [lr,:64], r2
- vrhadd.u8 q8, q8, q10
- .endif
- vld1.8 {d6, d7}, [r1], r2
- vext.8 d7, d6, d7, #1
- vst1.8 {d16}, [r0,:64], r2
- vst1.8 {d17}, [r0,:64], r2
- bgt 5b
-
- pop {r4-r7, pc}
-endfunc
-.endm
-
-/* chroma_mc4(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) */
-.macro h264_chroma_mc4 type
-function ff_\type\()_h264_chroma_mc4_neon, export=1
- push {r4-r7, lr}
- ldrd r4, [sp, #20]
- .ifc \type,avg
- mov lr, r0
- .endif
- pld [r1]
- pld [r1, r2]
-
-A muls r7, r4, r5
-T mul r7, r4, r5
-T cmp r7, #0
- rsb r6, r7, r5, lsl #3
- rsb r12, r7, r4, lsl #3
- sub r4, r7, r4, lsl #3
- sub r4, r4, r5, lsl #3
- add r4, r4, #64
-
- beq 2f
-
- add r5, r1, r2
-
- vdup.8 d0, r4
- lsl r4, r2, #1
- vdup.8 d1, r12
- vld1.8 {d4}, [r1], r4
- vdup.8 d2, r6
- vld1.8 {d6}, [r5], r4
- vdup.8 d3, r7
-
- vext.8 d5, d4, d5, #1
- vext.8 d7, d6, d7, #1
- vtrn.32 d4, d5
- vtrn.32 d6, d7
-
- vtrn.32 d0, d1
- vtrn.32 d2, d3
-
-1: pld [r5]
- vmull.u8 q8, d4, d0
- vmlal.u8 q8, d6, d2
- vld1.8 {d4}, [r1], r4
- vext.8 d5, d4, d5, #1
- vtrn.32 d4, d5
- vmull.u8 q9, d6, d0
- vmlal.u8 q9, d4, d2
- vld1.8 {d6}, [r5], r4
- vadd.i16 d16, d16, d17
- vadd.i16 d17, d18, d19
- vrshrn.u16 d16, q8, #6
- subs r3, r3, #2
- pld [r1]
- .ifc \type,avg
- vld1.32 {d20[0]}, [lr,:32], r2
- vld1.32 {d20[1]}, [lr,:32], r2
- vrhadd.u8 d16, d16, d20
- .endif
- vext.8 d7, d6, d7, #1
- vtrn.32 d6, d7
- vst1.32 {d16[0]}, [r0,:32], r2
- vst1.32 {d16[1]}, [r0,:32], r2
- bgt 1b
-
- pop {r4-r7, pc}
-
-2: tst r6, r6
- add r12, r12, r6
- vdup.8 d0, r4
- vdup.8 d1, r12
- vtrn.32 d0, d1
-
- beq 4f
-
- vext.32 d1, d0, d1, #1
- add r5, r1, r2
- lsl r4, r2, #1
- vld1.32 {d4[0]}, [r1], r4
- vld1.32 {d4[1]}, [r5], r4
-
-3: pld [r5]
- vmull.u8 q8, d4, d0
- vld1.32 {d4[0]}, [r1], r4
- vmull.u8 q9, d4, d1
- vld1.32 {d4[1]}, [r5], r4
- vadd.i16 d16, d16, d17
- vadd.i16 d17, d18, d19
- vrshrn.u16 d16, q8, #6
- .ifc \type,avg
- vld1.32 {d20[0]}, [lr,:32], r2
- vld1.32 {d20[1]}, [lr,:32], r2
- vrhadd.u8 d16, d16, d20
- .endif
- subs r3, r3, #2
- pld [r1]
- vst1.32 {d16[0]}, [r0,:32], r2
- vst1.32 {d16[1]}, [r0,:32], r2
- bgt 3b
-
- pop {r4-r7, pc}
-
-4: vld1.8 {d4}, [r1], r2
- vld1.8 {d6}, [r1], r2
- vext.8 d5, d4, d5, #1
- vext.8 d7, d6, d7, #1
- vtrn.32 d4, d5
- vtrn.32 d6, d7
-
-5: vmull.u8 q8, d4, d0
- vmull.u8 q9, d6, d0
- subs r3, r3, #2
- vld1.8 {d4}, [r1], r2
- vext.8 d5, d4, d5, #1
- vtrn.32 d4, d5
- vadd.i16 d16, d16, d17
- vadd.i16 d17, d18, d19
- pld [r1]
- vrshrn.u16 d16, q8, #6
- .ifc \type,avg
- vld1.32 {d20[0]}, [lr,:32], r2
- vld1.32 {d20[1]}, [lr,:32], r2
- vrhadd.u8 d16, d16, d20
- .endif
- vld1.8 {d6}, [r1], r2
- vext.8 d7, d6, d7, #1
- vtrn.32 d6, d7
- pld [r1]
- vst1.32 {d16[0]}, [r0,:32], r2
- vst1.32 {d16[1]}, [r0,:32], r2
- bgt 5b
-
- pop {r4-r7, pc}
-endfunc
-.endm
-
-.macro h264_chroma_mc2 type
-function ff_\type\()_h264_chroma_mc2_neon, export=1
- push {r4-r6, lr}
- ldr r4, [sp, #16]
- ldr lr, [sp, #20]
- pld [r1]
- pld [r1, r2]
- orrs r5, r4, lr
- beq 2f
-
- mul r5, r4, lr
- rsb r6, r5, lr, lsl #3
- rsb r12, r5, r4, lsl #3
- sub r4, r5, r4, lsl #3
- sub r4, r4, lr, lsl #3
- add r4, r4, #64
- vdup.8 d0, r4
- vdup.8 d2, r12
- vdup.8 d1, r6
- vdup.8 d3, r5
- vtrn.16 q0, q1
-1:
- vld1.32 {d4[0]}, [r1], r2
- vld1.32 {d4[1]}, [r1], r2
- vrev64.32 d5, d4
- vld1.32 {d5[1]}, [r1]
- vext.8 q3, q2, q2, #1
- vtrn.16 q2, q3
- vmull.u8 q8, d4, d0
- vmlal.u8 q8, d5, d1
- .ifc \type,avg
- vld1.16 {d18[0]}, [r0,:16], r2
- vld1.16 {d18[1]}, [r0,:16]
- sub r0, r0, r2
- .endif
- vtrn.32 d16, d17
- vadd.i16 d16, d16, d17
- vrshrn.u16 d16, q8, #6
- .ifc \type,avg
- vrhadd.u8 d16, d16, d18
- .endif
- vst1.16 {d16[0]}, [r0,:16], r2
- vst1.16 {d16[1]}, [r0,:16], r2
- subs r3, r3, #2
- bgt 1b
- pop {r4-r6, pc}
-2:
- .ifc \type,put
- ldrh_post r5, r1, r2
- strh_post r5, r0, r2
- ldrh_post r6, r1, r2
- strh_post r6, r0, r2
- .else
- vld1.16 {d16[0]}, [r1], r2
- vld1.16 {d16[1]}, [r1], r2
- vld1.16 {d18[0]}, [r0,:16], r2
- vld1.16 {d18[1]}, [r0,:16]
- sub r0, r0, r2
- vrhadd.u8 d16, d16, d18
- vst1.16 {d16[0]}, [r0,:16], r2
- vst1.16 {d16[1]}, [r0,:16], r2
- .endif
- subs r3, r3, #2
- bgt 2b
- pop {r4-r6, pc}
-endfunc
-.endm
-
- h264_chroma_mc8 put
- h264_chroma_mc8 avg
- h264_chroma_mc4 put
- h264_chroma_mc4 avg
- h264_chroma_mc2 put
- h264_chroma_mc2 avg
-
/* H.264 loop filter */
.macro h264_loop_filter_start
From f5c05b9aa5aeb6079b76f9da452f8ee4050e8955 Mon Sep 17 00:00:00 2001
From: Janne Grunau <janne-libav@jannau.net>
Date: Mon, 5 Dec 2011 21:18:05 +0000
Subject: [PATCH 3/5] rv40: NEON optimised chroma MC
Signed-off-by: Mans Rullgard <mans@mansr.com>
---
libavcodec/arm/Makefile | 2 +
libavcodec/arm/h264cmc_neon.S | 80 ++++++++++++++++++++++++++++--
libavcodec/arm/rv40dsp_init_neon.c | 38 ++++++++++++++
libavcodec/rv34dsp.h | 1 +
libavcodec/rv40dsp.c | 2 +
5 files changed, 118 insertions(+), 5 deletions(-)
create mode 100644 libavcodec/arm/rv40dsp_init_neon.c
diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index c125a59078..a948e6db3f 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -68,6 +68,8 @@ NEON-OBJS-$(CONFIG_RV30_DECODER) += arm/rv34dsp_init_neon.o \
NEON-OBJS-$(CONFIG_RV40_DECODER) += arm/rv34dsp_init_neon.o \
arm/rv34dsp_neon.o \
+ arm/rv40dsp_init_neon.o \
+ arm/h264cmc_neon.o \
NEON-OBJS-$(CONFIG_VP3_DECODER) += arm/vp3dsp_neon.o
diff --git a/libavcodec/arm/h264cmc_neon.S b/libavcodec/arm/h264cmc_neon.S
index e10adaca10..a6feadd189 100644
--- a/libavcodec/arm/h264cmc_neon.S
+++ b/libavcodec/arm/h264cmc_neon.S
@@ -21,8 +21,8 @@
#include "asm.S"
/* chroma_mc8(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) */
-.macro h264_chroma_mc8 type
-function ff_\type\()_h264_chroma_mc8_neon, export=1
+.macro h264_chroma_mc8 type, codec=h264
+function ff_\type\()_\codec\()_chroma_mc8_neon, export=1
push {r4-r7, lr}
ldrd r4, [sp, #20]
.ifc \type,avg
@@ -31,6 +31,15 @@ function ff_\type\()_h264_chroma_mc8_neon, export=1
pld [r1]
pld [r1, r2]
+ .ifc \codec,rv40
+ movrel r6, rv40bias
+ lsr r7, r5, #1
+ add r6, r6, r7, lsl #3
+ lsr r7, r4, #1
+ add r6, r6, r7, lsl #1
+ vld1.16 {d22[],d23[]}, [r6,:16]
+ .endif
+
A muls r7, r4, r5
T mul r7, r4, r5
T cmp r7, #0
@@ -67,10 +76,17 @@ T cmp r7, #0
vmlal.u8 q9, d7, d1
vmlal.u8 q9, d4, d2
vmlal.u8 q9, d5, d3
- vrshrn.u16 d16, q8, #6
vld1.8 {d6, d7}, [r5], r4
pld [r1]
+ .ifc \codec,h264
+ vrshrn.u16 d16, q8, #6
vrshrn.u16 d17, q9, #6
+ .else
+ vadd.u16 q8, q8, q11
+ vadd.u16 q9, q9, q11
+ vshrn.u16 d16, q8, #6
+ vshrn.u16 d17, q9, #6
+ .endif
.ifc \type,avg
vld1.8 {d20}, [lr,:64], r2
vld1.8 {d21}, [lr,:64], r2
@@ -102,8 +118,15 @@ T cmp r7, #0
vmull.u8 q9, d6, d0
vmlal.u8 q9, d4, d1
vld1.8 {d6}, [r5], r4
+ .ifc \codec,h264
vrshrn.u16 d16, q8, #6
vrshrn.u16 d17, q9, #6
+ .else
+ vadd.u16 q8, q8, q11
+ vadd.u16 q9, q9, q11
+ vshrn.u16 d16, q8, #6
+ vshrn.u16 d17, q9, #6
+ .endif
.ifc \type,avg
vld1.8 {d20}, [lr,:64], r2
vld1.8 {d21}, [lr,:64], r2
@@ -131,8 +154,15 @@ T cmp r7, #0
vmlal.u8 q9, d7, d1
pld [r1]
vext.8 d5, d4, d5, #1
+ .ifc \codec,h264
vrshrn.u16 d16, q8, #6
vrshrn.u16 d17, q9, #6
+ .else
+ vadd.u16 q8, q8, q11
+ vadd.u16 q9, q9, q11
+ vshrn.u16 d16, q8, #6
+ vshrn.u16 d17, q9, #6
+ .endif
.ifc \type,avg
vld1.8 {d20}, [lr,:64], r2
vld1.8 {d21}, [lr,:64], r2
@@ -149,8 +179,8 @@ endfunc
.endm
/* chroma_mc4(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y) */
-.macro h264_chroma_mc4 type
-function ff_\type\()_h264_chroma_mc4_neon, export=1
+.macro h264_chroma_mc4 type, codec=h264
+function ff_\type\()_\codec\()_chroma_mc4_neon, export=1
push {r4-r7, lr}
ldrd r4, [sp, #20]
.ifc \type,avg
@@ -159,6 +189,15 @@ function ff_\type\()_h264_chroma_mc4_neon, export=1
pld [r1]
pld [r1, r2]
+ .ifc \codec,rv40
+ movrel r6, rv40bias
+ lsr r7, r5, #1
+ add r6, r6, r7, lsl #3
+ lsr r7, r4, #1
+ add r6, r6, r7, lsl #1
+ vld1.16 {d22[],d23[]}, [r6,:16]
+ .endif
+
A muls r7, r4, r5
T mul r7, r4, r5
T cmp r7, #0
@@ -199,7 +238,12 @@ T cmp r7, #0
vld1.8 {d6}, [r5], r4
vadd.i16 d16, d16, d17
vadd.i16 d17, d18, d19
+ .ifc \codec,h264
vrshrn.u16 d16, q8, #6
+ .else
+ vadd.u16 q8, q8, q11
+ vshrn.u16 d16, q8, #6
+ .endif
subs r3, r3, #2
pld [r1]
.ifc \type,avg
@@ -236,7 +280,12 @@ T cmp r7, #0
vld1.32 {d4[1]}, [r5], r4
vadd.i16 d16, d16, d17
vadd.i16 d17, d18, d19
+ .ifc \codec,h264
vrshrn.u16 d16, q8, #6
+ .else
+ vadd.u16 q8, q8, q11
+ vshrn.u16 d16, q8, #6
+ .endif
.ifc \type,avg
vld1.32 {d20[0]}, [lr,:32], r2
vld1.32 {d20[1]}, [lr,:32], r2
@@ -266,7 +315,12 @@ T cmp r7, #0
vadd.i16 d16, d16, d17
vadd.i16 d17, d18, d19
pld [r1]
+ .ifc \codec,h264
vrshrn.u16 d16, q8, #6
+ .else
+ vadd.u16 q8, q8, q11
+ vshrn.u16 d16, q8, #6
+ .endif
.ifc \type,avg
vld1.32 {d20[0]}, [lr,:32], r2
vld1.32 {d20[1]}, [lr,:32], r2
@@ -352,9 +406,25 @@ function ff_\type\()_h264_chroma_mc2_neon, export=1
endfunc
.endm
+#if CONFIG_H264_DECODER
h264_chroma_mc8 put
h264_chroma_mc8 avg
h264_chroma_mc4 put
h264_chroma_mc4 avg
h264_chroma_mc2 put
h264_chroma_mc2 avg
+#endif
+
+#if CONFIG_RV40_DECODER
+const rv40bias
+ .short 0, 16, 32, 16
+ .short 32, 28, 32, 28
+ .short 0, 32, 16, 32
+ .short 32, 28, 32, 28
+endconst
+
+ h264_chroma_mc8 put, rv40
+ h264_chroma_mc8 avg, rv40
+ h264_chroma_mc4 put, rv40
+ h264_chroma_mc4 avg, rv40
+#endif
diff --git a/libavcodec/arm/rv40dsp_init_neon.c b/libavcodec/arm/rv40dsp_init_neon.c
new file mode 100644
index 0000000000..aa4a88da1a
--- /dev/null
+++ b/libavcodec/arm/rv40dsp_init_neon.c
@@ -0,0 +1,38 @@
+/*
+ * Copyright (c) 2011 Janne Grunau <janne-libav@jannau.net>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "libavcodec/avcodec.h"
+#include "libavcodec/rv34dsp.h"
+
+void ff_put_rv40_chroma_mc8_neon(uint8_t *, uint8_t *, int, int, int, int);
+void ff_put_rv40_chroma_mc4_neon(uint8_t *, uint8_t *, int, int, int, int);
+
+void ff_avg_rv40_chroma_mc8_neon(uint8_t *, uint8_t *, int, int, int, int);
+void ff_avg_rv40_chroma_mc4_neon(uint8_t *, uint8_t *, int, int, int, int);
+
+void ff_rv40dsp_init_neon(RV34DSPContext *c, DSPContext* dsp)
+{
+ c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_neon;
+ c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_neon;
+ c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_neon;
+ c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_neon;
+}
diff --git a/libavcodec/rv34dsp.h b/libavcodec/rv34dsp.h
index a1636e6eb5..695af06970 100644
--- a/libavcodec/rv34dsp.h
+++ b/libavcodec/rv34dsp.h
@@ -59,5 +59,6 @@ void ff_rv40dsp_init(RV34DSPContext *c, DSPContext* dsp);
void ff_rv34dsp_init_neon(RV34DSPContext *c, DSPContext *dsp);
void ff_rv40dsp_init_x86(RV34DSPContext *c, DSPContext *dsp);
+void ff_rv40dsp_init_neon(RV34DSPContext *c, DSPContext *dsp);
#endif /* AVCODEC_RV34DSP_H */
diff --git a/libavcodec/rv40dsp.c b/libavcodec/rv40dsp.c
index f193b6050d..06bdf18c42 100644
--- a/libavcodec/rv40dsp.c
+++ b/libavcodec/rv40dsp.c
@@ -534,4 +534,6 @@ av_cold void ff_rv40dsp_init(RV34DSPContext *c, DSPContext* dsp) {
if (HAVE_MMX)
ff_rv40dsp_init_x86(c, dsp);
+ if (HAVE_NEON)
+ ff_rv40dsp_init_neon(c, dsp);
}
From 6c889888662168811389e209bfbc662d70c27627 Mon Sep 17 00:00:00 2001
From: Janne Grunau <janne-libav@jannau.net>
Date: Mon, 5 Dec 2011 21:22:57 +0000
Subject: [PATCH 4/5] rv40: NEON optimised weighted prediction
Signed-off-by: Mans Rullgard <mans@mansr.com>
---
libavcodec/arm/Makefile | 1 +
libavcodec/arm/rv40dsp_init_neon.c | 6 +++
libavcodec/arm/rv40dsp_neon.S | 85 ++++++++++++++++++++++++++++++
3 files changed, 92 insertions(+)
create mode 100644 libavcodec/arm/rv40dsp_neon.S
diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index a948e6db3f..fc1711395b 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -69,6 +69,7 @@ NEON-OBJS-$(CONFIG_RV30_DECODER) += arm/rv34dsp_init_neon.o \
NEON-OBJS-$(CONFIG_RV40_DECODER) += arm/rv34dsp_init_neon.o \
arm/rv34dsp_neon.o \
arm/rv40dsp_init_neon.o \
+ arm/rv40dsp_neon.o \
arm/h264cmc_neon.o \
NEON-OBJS-$(CONFIG_VP3_DECODER) += arm/vp3dsp_neon.o
diff --git a/libavcodec/arm/rv40dsp_init_neon.c b/libavcodec/arm/rv40dsp_init_neon.c
index aa4a88da1a..3a863e1916 100644
--- a/libavcodec/arm/rv40dsp_init_neon.c
+++ b/libavcodec/arm/rv40dsp_init_neon.c
@@ -29,10 +29,16 @@ void ff_put_rv40_chroma_mc4_neon(uint8_t *, uint8_t *, int, int, int, int);
void ff_avg_rv40_chroma_mc8_neon(uint8_t *, uint8_t *, int, int, int, int);
void ff_avg_rv40_chroma_mc4_neon(uint8_t *, uint8_t *, int, int, int, int);
+void ff_rv40_weight_func_16_neon(uint8_t *, uint8_t *, uint8_t *, int, int, int);
+void ff_rv40_weight_func_8_neon(uint8_t *, uint8_t *, uint8_t *, int, int, int);
+
void ff_rv40dsp_init_neon(RV34DSPContext *c, DSPContext* dsp)
{
c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_neon;
c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_neon;
c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_neon;
c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_neon;
+
+ c->rv40_weight_pixels_tab[0] = ff_rv40_weight_func_16_neon;
+ c->rv40_weight_pixels_tab[1] = ff_rv40_weight_func_8_neon;
}
diff --git a/libavcodec/arm/rv40dsp_neon.S b/libavcodec/arm/rv40dsp_neon.S
new file mode 100644
index 0000000000..cafd98add0
--- /dev/null
+++ b/libavcodec/arm/rv40dsp_neon.S
@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2011 Janne Grunau <janne-libav@jannau.net>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "asm.S"
+
+.macro rv40_weight
+ vmovl.u8 q8, d2
+ vmovl.u8 q9, d3
+ vmovl.u8 q10, d4
+ vmovl.u8 q11, d5
+ vmull.u16 q2, d16, d0[2]
+ vmull.u16 q3, d17, d0[2]
+ vmull.u16 q8, d18, d0[2]
+ vmull.u16 q9, d19, d0[2]
+ vmull.u16 q12, d20, d0[0]
+ vmull.u16 q13, d21, d0[0]
+ vmull.u16 q14, d22, d0[0]
+ vmull.u16 q15, d23, d0[0]
+ vshrn.i32 d4, q2, #9
+ vshrn.i32 d5, q3, #9
+ vshrn.i32 d6, q8, #9
+ vshrn.i32 d7, q9, #9
+ vshrn.i32 d16, q12, #9
+ vshrn.i32 d17, q13, #9
+ vshrn.i32 d18, q14, #9
+ vshrn.i32 d19, q15, #9
+ vadd.u16 q2, q2, q8
+ vadd.u16 q3, q3, q9
+ vrshrn.i16 d2, q2, #5
+ vrshrn.i16 d3, q3, #5
+.endm
+
+/* void ff_rv40_weight_func_16_neon(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+ int w1, int w2, int stride) */
+function ff_rv40_weight_func_16_neon, export=1
+ ldr r12, [sp]
+ vmov d0, r3, r12
+ ldr r12, [sp, #4]
+ mov r3, #16
+1:
+ vld1.8 {q1}, [r1,:128], r12
+ vld1.8 {q2}, [r2,:128], r12
+ rv40_weight
+ vst1.8 {q1}, [r0,:128], r12
+ subs r3, r3, #1
+ bne 1b
+ bx lr
+endfunc
+
+/* void ff_rv40_weight_func_8_neon(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+ int w1, int w2, int stride) */
+function ff_rv40_weight_func_8_neon, export=1
+ ldr r12, [sp]
+ vmov d0, r3, r12
+ ldr r12, [sp, #4]
+ mov r3, #8
+1:
+ vld1.8 {d2}, [r1,:64], r12
+ vld1.8 {d3}, [r1,:64], r12
+ vld1.8 {d4}, [r2,:64], r12
+ vld1.8 {d5}, [r2,:64], r12
+ rv40_weight
+ vst1.8 {d2}, [r0,:64], r12
+ vst1.8 {d3}, [r0,:64], r12
+ subs r3, r3, #2
+ bne 1b
+ bx lr
+endfunc
From 4d9ec050a21c953f5a6e31483cf5ebd344afa3fd Mon Sep 17 00:00:00 2001
From: Konstantin Todorov <bioactiv3@abv.bg>
Date: Mon, 5 Dec 2011 18:40:28 +0200
Subject: [PATCH 5/5] Code cleanup - mpegvideo.c - 500-1000line
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
---
libavcodec/mpegvideo.c | 542 +++++++++++++++++++++++------------------
1 file changed, 309 insertions(+), 233 deletions(-)
diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
index 46b6232b08..78fae1026d 100644
--- a/libavcodec/mpegvideo.c
+++ b/libavcodec/mpegvideo.c
@@ -567,46 +567,56 @@ int ff_mpeg_update_thread_context(AVCodecContext *dst,
s->input_picture_number = s1->input_picture_number;
memcpy(s->picture, s1->picture, s1->picture_count * sizeof(Picture));
- memcpy(&s->last_picture, &s1->last_picture, (char*)&s1->last_picture_ptr - (char*)&s1->last_picture);
+ memcpy(&s->last_picture, &s1->last_picture,
+ (char *) &s1->last_picture_ptr - (char *) &s1->last_picture);
- s->last_picture_ptr = REBASE_PICTURE(s1->last_picture_ptr, s, s1);
- s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
- s->next_picture_ptr = REBASE_PICTURE(s1->next_picture_ptr, s, s1);
+ s->last_picture_ptr = REBASE_PICTURE(s1->last_picture_ptr, s, s1);
+ s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
+ s->next_picture_ptr = REBASE_PICTURE(s1->next_picture_ptr, s, s1);
- memcpy(s->prev_pict_types, s1->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE);
+ memcpy(s->prev_pict_types, s1->prev_pict_types,
+ PREV_PICT_TYPES_BUFFER_SIZE);
- //Error/bug resilience
+ // Error/bug resilience
s->next_p_frame_damaged = s1->next_p_frame_damaged;
s->workaround_bugs = s1->workaround_bugs;
- //MPEG4 timing info
- memcpy(&s->time_increment_bits, &s1->time_increment_bits, (char*)&s1->shape - (char*)&s1->time_increment_bits);
+ // MPEG4 timing info
+ memcpy(&s->time_increment_bits, &s1->time_increment_bits,
+ (char *) &s1->shape - (char *) &s1->time_increment_bits);
- //B-frame info
- s->max_b_frames = s1->max_b_frames;
- s->low_delay = s1->low_delay;
- s->dropable = s1->dropable;
+ // B-frame info
+ s->max_b_frames = s1->max_b_frames;
+ s->low_delay = s1->low_delay;
+ s->dropable = s1->dropable;
- //DivX handling (doesn't work)
- s->divx_packed = s1->divx_packed;
+ // DivX handling (doesn't work)
+ s->divx_packed = s1->divx_packed;
- if(s1->bitstream_buffer){
- if (s1->bitstream_buffer_size + FF_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size)
- av_fast_malloc(&s->bitstream_buffer, &s->allocated_bitstream_buffer_size, s1->allocated_bitstream_buffer_size);
- s->bitstream_buffer_size = s1->bitstream_buffer_size;
- memcpy(s->bitstream_buffer, s1->bitstream_buffer, s1->bitstream_buffer_size);
- memset(s->bitstream_buffer+s->bitstream_buffer_size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
+ if (s1->bitstream_buffer) {
+ if (s1->bitstream_buffer_size +
+ FF_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size)
+ av_fast_malloc(&s->bitstream_buffer,
+ &s->allocated_bitstream_buffer_size,
+ s1->allocated_bitstream_buffer_size);
+ s->bitstream_buffer_size = s1->bitstream_buffer_size;
+ memcpy(s->bitstream_buffer, s1->bitstream_buffer,
+ s1->bitstream_buffer_size);
+ memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
+ FF_INPUT_BUFFER_PADDING_SIZE);
}
- //MPEG2/interlacing info
- memcpy(&s->progressive_sequence, &s1->progressive_sequence, (char*)&s1->rtp_mode - (char*)&s1->progressive_sequence);
+ // MPEG2/interlacing info
+ memcpy(&s->progressive_sequence, &s1->progressive_sequence,
+ (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
- if(!s1->first_field){
- s->last_pict_type= s1->pict_type;
- if (s1->current_picture_ptr) s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f.quality;
+ if (!s1->first_field) {
+ s->last_pict_type = s1->pict_type;
+ if (s1->current_picture_ptr)
+ s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f.quality;
if (s1->pict_type != AV_PICTURE_TYPE_B) {
- s->last_non_b_pict_type= s1->pict_type;
+ s->last_non_b_pict_type = s1->pict_type;
}
}
@@ -614,35 +624,40 @@ int ff_mpeg_update_thread_context(AVCodecContext *dst,
}
/**
- * sets the given MpegEncContext to common defaults (same for encoding and decoding).
- * the changed fields will not depend upon the prior state of the MpegEncContext.
+ * sets the given MpegEncContext to common defaults
+ * (same for encoding and decoding).
+ * the changed fields will not depend upon the
+ * prior state of the MpegEncContext.
*/
-void MPV_common_defaults(MpegEncContext *s){
- s->y_dc_scale_table=
- s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
- s->chroma_qscale_table= ff_default_chroma_qscale_table;
- s->progressive_frame= 1;
- s->progressive_sequence= 1;
- s->picture_structure= PICT_FRAME;
-
- s->coded_picture_number = 0;
- s->picture_number = 0;
- s->input_picture_number = 0;
+void MPV_common_defaults(MpegEncContext *s)
+{
+ s->y_dc_scale_table =
+ s->c_dc_scale_table = ff_mpeg1_dc_scale_table;
+ s->chroma_qscale_table = ff_default_chroma_qscale_table;
+ s->progressive_frame = 1;
+ s->progressive_sequence = 1;
+ s->picture_structure = PICT_FRAME;
+
+ s->coded_picture_number = 0;
+ s->picture_number = 0;
+ s->input_picture_number = 0;
s->picture_in_gop_number = 0;
- s->f_code = 1;
- s->b_code = 1;
+ s->f_code = 1;
+ s->b_code = 1;
- s->picture_range_start = 0;
- s->picture_range_end = MAX_PICTURE_COUNT;
+ s->picture_range_start = 0;
+ s->picture_range_end = MAX_PICTURE_COUNT;
}
/**
* sets the given MpegEncContext to defaults for decoding.
- * the changed fields will not depend upon the prior state of the MpegEncContext.
+ * the changed fields will not depend upon
+ * the prior state of the MpegEncContext.
*/
-void MPV_decode_defaults(MpegEncContext *s){
+void MPV_decode_defaults(MpegEncContext *s)
+{
MPV_common_defaults(s);
}
@@ -658,157 +673,203 @@ av_cold int MPV_common_init(MpegEncContext *s)
s->avctx->active_thread_type & FF_THREAD_SLICE)) ?
s->avctx->thread_count : 1;
- if(s->codec_id == CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
+ if (s->codec_id == CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
s->mb_height = (s->height + 31) / 32 * 2;
else if (s->codec_id != CODEC_ID_H264)
s->mb_height = (s->height + 15) / 16;
- if(s->avctx->pix_fmt == PIX_FMT_NONE){
- av_log(s->avctx, AV_LOG_ERROR, "decoding to PIX_FMT_NONE is not supported.\n");
+ if (s->avctx->pix_fmt == PIX_FMT_NONE) {
+ av_log(s->avctx, AV_LOG_ERROR,
+ "decoding to PIX_FMT_NONE is not supported.\n");
return -1;
}
- if((s->encoding || (s->avctx->active_thread_type & FF_THREAD_SLICE)) &&
- (s->avctx->thread_count > MAX_THREADS || (s->avctx->thread_count > s->mb_height && s->mb_height))){
+ if ((s->encoding || (s->avctx->active_thread_type & FF_THREAD_SLICE)) &&
+ (s->avctx->thread_count > MAX_THREADS ||
+ (s->avctx->thread_count > s->mb_height && s->mb_height))) {
int max_threads = FFMIN(MAX_THREADS, s->mb_height);
- av_log(s->avctx, AV_LOG_WARNING, "too many threads (%d), reducing to %d\n",
+ av_log(s->avctx, AV_LOG_WARNING,
+ "too many threads (%d), reducing to %d\n",
s->avctx->thread_count, max_threads);
threads = max_threads;
}
- if((s->width || s->height) && av_image_check_size(s->width, s->height, 0, s->avctx))
+ if ((s->width || s->height) &&
+ av_image_check_size(s->width, s->height, 0, s->avctx))
return -1;
ff_dct_common_init(s);
- s->flags= s->avctx->flags;
- s->flags2= s->avctx->flags2;
+ s->flags = s->avctx->flags;
+ s->flags2 = s->avctx->flags2;
if (s->width && s->height) {
- s->mb_width = (s->width + 15) / 16;
- s->mb_stride = s->mb_width + 1;
- s->b8_stride = s->mb_width*2 + 1;
- s->b4_stride = s->mb_width*4 + 1;
- mb_array_size= s->mb_height * s->mb_stride;
- mv_table_size= (s->mb_height+2) * s->mb_stride + 1;
+ s->mb_width = (s->width + 15) / 16;
+ s->mb_stride = s->mb_width + 1;
+ s->b8_stride = s->mb_width * 2 + 1;
+ s->b4_stride = s->mb_width * 4 + 1;
+ mb_array_size = s->mb_height * s->mb_stride;
+ mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
/* set chroma shifts */
avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,&(s->chroma_x_shift),
&(s->chroma_y_shift) );
- /* set default edge pos, will be overriden in decode_header if needed */
- s->h_edge_pos= s->mb_width*16;
- s->v_edge_pos= s->mb_height*16;
+ /* set default edge pos, will be overriden
+ * in decode_header if needed */
+ s->h_edge_pos = s->mb_width * 16;
+ s->v_edge_pos = s->mb_height * 16;
- s->mb_num = s->mb_width * s->mb_height;
+ s->mb_num = s->mb_width * s->mb_height;
- s->block_wrap[0]=
- s->block_wrap[1]=
- s->block_wrap[2]=
- s->block_wrap[3]= s->b8_stride;
- s->block_wrap[4]=
- s->block_wrap[5]= s->mb_stride;
+ s->block_wrap[0] =
+ s->block_wrap[1] =
+ s->block_wrap[2] =
+ s->block_wrap[3] = s->b8_stride;
+ s->block_wrap[4] =
+ s->block_wrap[5] = s->mb_stride;
- y_size = s->b8_stride * (2 * s->mb_height + 1);
- c_size = s->mb_stride * (s->mb_height + 1);
- yc_size = y_size + 2 * c_size;
+ y_size = s->b8_stride * (2 * s->mb_height + 1);
+ c_size = s->mb_stride * (s->mb_height + 1);
+ yc_size = y_size + 2 * c_size;
/* convert fourcc to upper case */
- s->codec_tag = avpriv_toupper4(s->avctx->codec_tag);
+ s->codec_tag = avpriv_toupper4(s->avctx->codec_tag);
- s->stream_codec_tag = avpriv_toupper4(s->avctx->stream_codec_tag);
+ s->stream_codec_tag = avpriv_toupper4(s->avctx->stream_codec_tag);
- s->avctx->coded_frame= (AVFrame*)&s->current_picture;
+ s->avctx->coded_frame = (AVFrame *)&s->current_picture;
- FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num+1)*sizeof(int), fail) //error ressilience code looks cleaner with this
- for(y=0; y<s->mb_height; y++){
- for(x=0; x<s->mb_width; x++){
- s->mb_index2xy[ x + y*s->mb_width ] = x + y*s->mb_stride;
- }
- }
- s->mb_index2xy[ s->mb_height*s->mb_width ] = (s->mb_height-1)*s->mb_stride + s->mb_width; //FIXME really needed?
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int),
+ fail); // error ressilience code looks cleaner with this
+ for (y = 0; y < s->mb_height; y++)
+ for (x = 0; x < s->mb_width; x++)
+ s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
+
+ s->mb_index2xy[s->mb_height * s->mb_width] =
+ (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
if (s->encoding) {
/* Allocate MV tables */
- FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base , mv_table_size * 2 * sizeof(int16_t), fail)
- s->p_mv_table = s->p_mv_table_base + s->mb_stride + 1;
- s->b_forw_mv_table = s->b_forw_mv_table_base + s->mb_stride + 1;
- s->b_back_mv_table = s->b_back_mv_table_base + s->mb_stride + 1;
- s->b_bidir_forw_mv_table= s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
- s->b_bidir_back_mv_table= s->b_bidir_back_mv_table_base + s->mb_stride + 1;
- s->b_direct_mv_table = s->b_direct_mv_table_base + s->mb_stride + 1;
-
- if(s->msmpeg4_version){
- FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats, 2*2*(MAX_LEVEL+1)*(MAX_RUN+1)*2*sizeof(int), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base,
+ mv_table_size * 2 * sizeof(int16_t), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base,
+ mv_table_size * 2 * sizeof(int16_t), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base,
+ mv_table_size * 2 * sizeof(int16_t), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base,
+ mv_table_size * 2 * sizeof(int16_t), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base,
+ mv_table_size * 2 * sizeof(int16_t), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base,
+ mv_table_size * 2 * sizeof(int16_t), fail);
+ s->p_mv_table = s->p_mv_table_base +
+ s->mb_stride + 1;
+ s->b_forw_mv_table = s->b_forw_mv_table_base +
+ s->mb_stride + 1;
+ s->b_back_mv_table = s->b_back_mv_table_base +
+ s->mb_stride + 1;
+ s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base +
+ s->mb_stride + 1;
+ s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base +
+ s->mb_stride + 1;
+ s->b_direct_mv_table = s->b_direct_mv_table_base +
+ s->mb_stride + 1;
+
+ if (s->msmpeg4_version) {
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats,
+ 2 * 2 * (MAX_LEVEL + 1) *
+ (MAX_RUN + 1) * 2 * sizeof(int), fail);
}
FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
/* Allocate MB type table */
- FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type , mb_array_size * sizeof(uint16_t), fail) //needed for encoding
-
- FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
-
- FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix , 64*32 * sizeof(int), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix , 64*32 * sizeof(int), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*), fail)
-
- if(s->avctx->noise_reduction){
- FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset, 2 * 64 * sizeof(uint16_t), fail)
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size *
+ sizeof(uint16_t), fail); // needed for encoding
+
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size *
+ sizeof(int), fail);
+
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix,
+ 64 * 32 * sizeof(int), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix,
+ 64 * 32 * sizeof(int), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16,
+ 64 * 32 * 2 * sizeof(uint16_t), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16,
+ 64 * 32 * 2 * sizeof(uint16_t), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture,
+ MAX_PICTURE_COUNT * sizeof(Picture *), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture,
+ MAX_PICTURE_COUNT * sizeof(Picture *), fail);
+
+ if (s->avctx->noise_reduction) {
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset,
+ 2 * 64 * sizeof(uint16_t), fail);
}
}
}
s->picture_count = MAX_PICTURE_COUNT * FFMAX(1, s->avctx->thread_count);
- FF_ALLOCZ_OR_GOTO(s->avctx, s->picture, s->picture_count * sizeof(Picture), fail)
- for(i = 0; i < s->picture_count; i++) {
- avcodec_get_frame_defaults((AVFrame *)&s->picture[i]);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
+ s->picture_count * sizeof(Picture), fail);
+ for (i = 0; i < s->picture_count; i++) {
+ avcodec_get_frame_defaults((AVFrame *) &s->picture[i]);
}
if (s->width && s->height) {
- FF_ALLOCZ_OR_GOTO(s->avctx, s->error_status_table, mb_array_size*sizeof(uint8_t), fail)
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->error_status_table,
+ mb_array_size * sizeof(uint8_t), fail);
- if(s->codec_id==CODEC_ID_MPEG4 || (s->flags & CODEC_FLAG_INTERLACED_ME)){
+ if (s->codec_id == CODEC_ID_MPEG4 ||
+ (s->flags & CODEC_FLAG_INTERLACED_ME)) {
/* interlaced direct mode decoding tables */
- for(i=0; i<2; i++){
+ for (i = 0; i < 2; i++) {
int j, k;
- for(j=0; j<2; j++){
- for(k=0; k<2; k++){
- FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_mv_table_base[i][j][k], mv_table_size * 2 * sizeof(int16_t), fail)
- s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] + s->mb_stride + 1;
+ for (j = 0; j < 2; j++) {
+ for (k = 0; k < 2; k++) {
+ FF_ALLOCZ_OR_GOTO(s->avctx,
+ s->b_field_mv_table_base[i][j][k],
+ mv_table_size * 2 * sizeof(int16_t),
+ fail);
+ s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
+ s->mb_stride + 1;
}
- FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
- s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j]+ s->mb_stride + 1;
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j],
+ mb_array_size * 2 * sizeof(uint8_t),
+ fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j],
+ mv_table_size * 2 * sizeof(int16_t),
+ fail);
+ s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j]
+ + s->mb_stride + 1;
}
- FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i],
+ mb_array_size * 2 * sizeof(uint8_t),
+ fail);
}
}
if (s->out_format == FMT_H263) {
/* cbp values */
FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size, fail);
- s->coded_block= s->coded_block_base + s->b8_stride + 1;
+ s->coded_block = s->coded_block_base + s->b8_stride + 1;
/* cbp, ac_pred, pred_dir */
- FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table , mb_array_size * sizeof(uint8_t), fail)
- FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail)
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table,
+ mb_array_size * sizeof(uint8_t), fail);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table,
+ mb_array_size * sizeof(uint8_t), fail);
}
if (s->h263_pred || s->h263_plus || !s->encoding) {
/* dc values */
- //MN: we need these for error resilience of intra-frames
- FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
+ // MN: we need these for error resilience of intra-frames
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base,
+ yc_size * sizeof(int16_t), fail);
s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
s->dc_val[2] = s->dc_val[1] + c_size;
- for(i=0;i<yc_size;i++)
+ for (i = 0; i < yc_size; i++)
s->dc_val_base[i] = 1024;
}
@@ -817,39 +878,50 @@ av_cold int MPV_common_init(MpegEncContext *s)
memset(s->mbintra_table, 1, mb_array_size);
/* init macroblock skip table */
- FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size+2, fail);
- //Note the +1 is for a quicker mpeg4 slice_end detection
- FF_ALLOCZ_OR_GOTO(s->avctx, s->prev_pict_types, PREV_PICT_TYPES_BUFFER_SIZE, fail);
-
- s->parse_context.state= -1;
- if((s->avctx->debug&(FF_DEBUG_VIS_QP|FF_DEBUG_VIS_MB_TYPE)) || (s->avctx->debug_mv)){
- s->visualization_buffer[0] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
- s->visualization_buffer[1] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
- s->visualization_buffer[2] = av_malloc((s->mb_width*16 + 2*EDGE_WIDTH) * s->mb_height*16 + 2*EDGE_WIDTH);
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
+ // Note the + 1 is for a quicker mpeg4 slice_end detection
+ FF_ALLOCZ_OR_GOTO(s->avctx, s->prev_pict_types,
+ PREV_PICT_TYPES_BUFFER_SIZE, fail);
+
+ s->parse_context.state = -1;
+ if ((s->avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
+ (s->avctx->debug_mv)) {
+ s->visualization_buffer[0] = av_malloc((s->mb_width * 16 +
+ 2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
+ s->visualization_buffer[1] = av_malloc((s->mb_width * 16 +
+ 2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
+ s->visualization_buffer[2] = av_malloc((s->mb_width * 16 +
+ 2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
}
}
s->context_initialized = 1;
- s->thread_context[0]= s;
+ s->thread_context[0] = s;
if (s->width && s->height) {
- if (s->encoding || (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_SLICE)) {
- for(i=1; i<threads; i++){
- s->thread_context[i]= av_malloc(sizeof(MpegEncContext));
- memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
- }
+ if (s->encoding || (HAVE_THREADS &&
+ s->avctx->active_thread_type&FF_THREAD_SLICE)) {
+ for (i = 1; i < threads; i++) {
+ s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
+ memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
+ }
- for(i=0; i<threads; i++){
- if(init_duplicate_context(s->thread_context[i], s) < 0)
+ for (i = 0; i < threads; i++) {
+ if (init_duplicate_context(s->thread_context[i], s) < 0)
+ goto fail;
+ s->thread_context[i]->start_mb_y =
+ (s->mb_height * (i) + s->avctx->thread_count / 2) /
+ s->avctx->thread_count;
+ s->thread_context[i]->end_mb_y =
+ (s->mb_height * (i + 1) + s->avctx->thread_count / 2) /
+ s->avctx->thread_count;
+ }
+ } else {
+ if (init_duplicate_context(s, s) < 0)
goto fail;
- s->thread_context[i]->start_mb_y= (s->mb_height*(i ) + s->avctx->thread_count/2) / s->avctx->thread_count;
- s->thread_context[i]->end_mb_y = (s->mb_height*(i+1) + s->avctx->thread_count/2) / s->avctx->thread_count;
+ s->start_mb_y = 0;
+ s->end_mb_y = s->mb_height;
}
- } else {
- if(init_duplicate_context(s, s) < 0) goto fail;
- s->start_mb_y = 0;
- s->end_mb_y = s->mb_height;
- }
}
return 0;
@@ -863,17 +935,18 @@ void MPV_common_end(MpegEncContext *s)
{
int i, j, k;
- if (s->encoding || (HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_SLICE)) {
- for(i=0; i<s->avctx->thread_count; i++){
+ if (s->encoding || (HAVE_THREADS &&
+ s->avctx->active_thread_type & FF_THREAD_SLICE)) {
+ for (i = 0; i < s->avctx->thread_count; i++) {
free_duplicate_context(s->thread_context[i]);
}
- for(i=1; i<s->avctx->thread_count; i++){
+ for (i = 1; i < s->avctx->thread_count; i++) {
av_freep(&s->thread_context[i]);
}
} else free_duplicate_context(s);
av_freep(&s->parse_context.buffer);
- s->parse_context.buffer_size=0;
+ s->parse_context.buffer_size = 0;
av_freep(&s->mb_type);
av_freep(&s->p_mv_table_base);
@@ -882,21 +955,21 @@ void MPV_common_end(MpegEncContext *s)
av_freep(&s->b_bidir_forw_mv_table_base);
av_freep(&s->b_bidir_back_mv_table_base);
av_freep(&s->b_direct_mv_table_base);
- s->p_mv_table= NULL;
- s->b_forw_mv_table= NULL;
- s->b_back_mv_table= NULL;
- s->b_bidir_forw_mv_table= NULL;
- s->b_bidir_back_mv_table= NULL;
- s->b_direct_mv_table= NULL;
- for(i=0; i<2; i++){
- for(j=0; j<2; j++){
- for(k=0; k<2; k++){
+ s->p_mv_table = NULL;
+ s->b_forw_mv_table = NULL;
+ s->b_back_mv_table = NULL;
+ s->b_bidir_forw_mv_table = NULL;
+ s->b_bidir_back_mv_table = NULL;
+ s->b_direct_mv_table = NULL;
+ for (i = 0; i < 2; i++) {
+ for (j = 0; j < 2; j++) {
+ for (k = 0; k < 2; k++) {
av_freep(&s->b_field_mv_table_base[i][j][k]);
- s->b_field_mv_table[i][j][k]=NULL;
+ s->b_field_mv_table[i][j][k] = NULL;
}
av_freep(&s->b_field_select_table[i][j]);
av_freep(&s->p_field_mv_table_base[i][j]);
- s->p_field_mv_table[i][j]=NULL;
+ s->p_field_mv_table[i][j] = NULL;
}
av_freep(&s->p_field_select_table[i]);
}
@@ -910,7 +983,7 @@ void MPV_common_end(MpegEncContext *s)
av_freep(&s->mbskip_table);
av_freep(&s->prev_pict_types);
av_freep(&s->bitstream_buffer);
- s->allocated_bitstream_buffer_size=0;
+ s->allocated_bitstream_buffer_size = 0;
av_freep(&s->avctx->stats_out);
av_freep(&s->ac_stats);
@@ -925,37 +998,38 @@ void MPV_common_end(MpegEncContext *s)
av_freep(&s->reordered_input_picture);
av_freep(&s->dct_offset);
- if(s->picture && !s->avctx->internal->is_copy){
- for(i=0; i<s->picture_count; i++){
+ if (s->picture && !s->avctx->internal->is_copy) {
+ for (i = 0; i < s->picture_count; i++) {
free_picture(s, &s->picture[i]);
}
}
av_freep(&s->picture);
- s->context_initialized = 0;
- s->last_picture_ptr=
- s->next_picture_ptr=
- s->current_picture_ptr= NULL;
- s->linesize= s->uvlinesize= 0;
+ s->context_initialized = 0;
+ s->last_picture_ptr =
+ s->next_picture_ptr =
+ s->current_picture_ptr = NULL;
+ s->linesize = s->uvlinesize = 0;
- for(i=0; i<3; i++)
+ for (i = 0; i < 3; i++)
av_freep(&s->visualization_buffer[i]);
- if(!(s->avctx->active_thread_type&FF_THREAD_FRAME))
+ if (!(s->avctx->active_thread_type & FF_THREAD_FRAME))
avcodec_default_free_buffers(s->avctx);
}
-void init_rl(RLTable *rl, uint8_t static_store[2][2*MAX_RUN + MAX_LEVEL + 3])
+void init_rl(RLTable *rl,
+ uint8_t static_store[2][2 * MAX_RUN + MAX_LEVEL + 3])
{
- int8_t max_level[MAX_RUN+1], max_run[MAX_LEVEL+1];
- uint8_t index_run[MAX_RUN+1];
+ int8_t max_level[MAX_RUN + 1], max_run[MAX_LEVEL + 1];
+ uint8_t index_run[MAX_RUN + 1];
int last, run, level, start, end, i;
- /* If table is static, we can quit if rl->max_level[0] is not NULL */
- if(static_store && rl->max_level[0])
+ /* If table is static, we can quit if rl->max_level[0] is not NULL */
+ if (static_store && rl->max_level[0])
return;
/* compute max_level[], max_run[] and index_run[] */
- for(last=0;last<2;last++) {
+ for (last = 0; last < 2; last++) {
if (last == 0) {
start = 0;
end = rl->last;
@@ -967,8 +1041,8 @@ void init_rl(RLTable *rl, uint8_t static_store[2][2*MAX_RUN + MAX_LEVEL + 3])
memset(max_level, 0, MAX_RUN + 1);
memset(max_run, 0, MAX_LEVEL + 1);
memset(index_run, rl->n, MAX_RUN + 1);
- for(i=start;i<end;i++) {
- run = rl->table_run[i];
+ for (i = start; i < end; i++) {
+ run = rl->table_run[i];
level = rl->table_level[i];
if (index_run[run] == rl->n)
index_run[run] = i;
@@ -977,17 +1051,17 @@ void init_rl(RLTable *rl, uint8_t static_store[2][2*MAX_RUN + MAX_LEVEL + 3])
if (run > max_run[level])
max_run[level] = run;
}
- if(static_store)
+ if (static_store)
rl->max_level[last] = static_store[last];
else
rl->max_level[last] = av_malloc(MAX_RUN + 1);
memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
- if(static_store)
- rl->max_run[last] = static_store[last] + MAX_RUN + 1;
+ if (static_store)
+ rl->max_run[last] = static_store[last] + MAX_RUN + 1;
else
- rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
+ rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
- if(static_store)
+ if (static_store)
rl->index_run[last] = static_store[last] + MAX_RUN + MAX_LEVEL + 2;
else
rl->index_run[last] = av_malloc(MAX_RUN + 1);
@@ -999,77 +1073,79 @@ void init_vlc_rl(RLTable *rl)
{
int i, q;
- for(q=0; q<32; q++){
- int qmul= q*2;
- int qadd= (q-1)|1;
+ for (q = 0; q < 32; q++) {
+ int qmul = q * 2;
+ int qadd = (q - 1) | 1;
- if(q==0){
- qmul=1;
- qadd=0;
+ if (q == 0) {
+ qmul = 1;
+ qadd = 0;
}
- for(i=0; i<rl->vlc.table_size; i++){
- int code= rl->vlc.table[i][0];
- int len = rl->vlc.table[i][1];
+ for (i = 0; i < rl->vlc.table_size; i++) {
+ int code = rl->vlc.table[i][0];
+ int len = rl->vlc.table[i][1];
int level, run;
- if(len==0){ // illegal code
- run= 66;
- level= MAX_LEVEL;
- }else if(len<0){ //more bits needed
- run= 0;
- level= code;
- }else{
- if(code==rl->n){ //esc
- run= 66;
- level= 0;
- }else{
- run= rl->table_run [code] + 1;
- level= rl->table_level[code] * qmul + qadd;
- if(code >= rl->last) run+=192;
+ if (len == 0) { // illegal code
+ run = 66;
+ level = MAX_LEVEL;
+ } else if (len < 0) { // more bits needed
+ run = 0;
+ level = code;
+ } else {
+ if (code == rl->n) { // esc
+ run = 66;
+ level = 0;
+ } else {
+ run = rl->table_run[code] + 1;
+ level = rl->table_level[code] * qmul + qadd;
+ if (code >= rl->last) run += 192;
}
}
- rl->rl_vlc[q][i].len= len;
- rl->rl_vlc[q][i].level= level;
- rl->rl_vlc[q][i].run= run;
+ rl->rl_vlc[q][i].len = len;
+ rl->rl_vlc[q][i].level = level;
+ rl->rl_vlc[q][i].run = run;
}
}
}
-void ff_release_unused_pictures(MpegEncContext *s, int remove_current)
+void ff_release_unused_pictures(MpegEncContext*s, int remove_current)
{
int i;
/* release non reference frames */
- for(i=0; i<s->picture_count; i++){
- if (s->picture[i].f.data[0] && !s->picture[i].f.reference
- && (!s->picture[i].owner2 || s->picture[i].owner2 == s)
- && (remove_current || &s->picture[i] != s->current_picture_ptr)
- /*&& s->picture[i].type!=FF_BUFFER_TYPE_SHARED*/){
+ for (i = 0; i < s->picture_count; i++) {
+ if (s->picture[i].f.data[0] && !s->picture[i].f.reference &&
+ (!s->picture[i].owner2 || s->picture[i].owner2 == s) &&
+ (remove_current || &s->picture[i] != s->current_picture_ptr)
+ /* && s->picture[i].type!= FF_BUFFER_TYPE_SHARED */) {
free_frame_buffer(s, &s->picture[i]);
}
}
}
-int ff_find_unused_picture(MpegEncContext *s, int shared){
+int ff_find_unused_picture(MpegEncContext *s, int shared)
+{
int i;
- if(shared){
- for(i=s->picture_range_start; i<s->picture_range_end; i++){
+ if (shared) {
+ for (i = s->picture_range_start; i < s->picture_range_end; i++) {
if (s->picture[i].f.data[0] == NULL && s->picture[i].f.type == 0)
return i;
}
- }else{
- for(i=s->picture_range_start; i<s->picture_range_end; i++){
+ } else {
+ for (i = s->picture_range_start; i < s->picture_range_end; i++) {
if (s->picture[i].f.data[0] == NULL && s->picture[i].f.type != 0)
- return i; //FIXME
+ return i; // FIXME
}
- for(i=s->picture_range_start; i<s->picture_range_end; i++){
+ for (i = s->picture_range_start; i < s->picture_range_end; i++) {
if (s->picture[i].f.data[0] == NULL)
return i;
}
}
- av_log(s->avctx, AV_LOG_FATAL, "Internal error, picture buffer overflow\n");
+ av_log(s->avctx, AV_LOG_FATAL,
+ "Internal error, picture buffer overflow\n");
/* We could return -1, but the codec would crash trying to draw into a
* non-existing frame anyway. This is safer than waiting for a random crash.
* Also the return of this is never useful, an encoder must only allocate