rv40: x86 SIMD for biweight

Provide MMX, SSE2 and SSSE3 versions, with a fast-path when the weights are
multiples of 512 (which is often the case when the values round up nicely).

*_TIMER report for the 16x16 and 8x8 cases:
C:
9015 decicycles in 16, 524257 runs, 31 skips
2656 decicycles in 8, 524271 runs, 17 skips
MMX:
4156 decicycles in 16, 262090 runs, 54 skips
1206 decicycles in 8, 262131 runs, 13 skips
MMX on fast-path:
2760 decicycles in 16, 524222 runs, 66 skips
995 decicycles in 8, 524252 runs, 36 skips
SSE2:
2163 decicycles in 16, 262131 runs, 13 skips
832 decicycles in 8, 262137 runs, 7 skips
SSE2 with fast path:
1783 decicycles in 16, 524276 runs, 12 skips
711 decicycles in 8, 524283 runs, 5 skips
SSSE3:
2117 decicycles in 16, 262136 runs, 8 skips
814 decicycles in 8, 262143 runs, 1 skips
SSSE3 with fast path:
1315 decicycles in 16, 524285 runs, 3 skips
578 decicycles in 8, 524286 runs, 2 skips

This means around a 4% speedup for some sequences.

Signed-off-by: Diego Biurrun <diego@biurrun.de>

3 files changed, 228 insertions, 1 deletions

diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 2f29138834..930ace78c4 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -28,7 +28,8 @@ MMX-OBJS-$(CONFIG_RV30_DECODER)        += x86/rv34dsp_init.o
 YASM-OBJS-$(CONFIG_RV30_DECODER)       += x86/rv34dsp.o
 MMX-OBJS-$(CONFIG_RV40_DECODER)        += x86/rv34dsp_init.o            \
                                           x86/rv40dsp_init.o
-YASM-OBJS-$(CONFIG_RV40_DECODER)       += x86/rv34dsp.o
+YASM-OBJS-$(CONFIG_RV40_DECODER)       += x86/rv34dsp.o                 \
+                                          x86/rv40dsp.o
 
 YASM-OBJS-$(CONFIG_VC1_DECODER)        += x86/vc1dsp_yasm.o
 
diff --git a/libavcodec/x86/rv40dsp.asm b/libavcodec/x86/rv40dsp.asm
new file mode 100644
index 0000000000..bff3e7b96a
--- /dev/null
+++ b/libavcodec/x86/rv40dsp.asm
@@ -0,0 +1,207 @@
+;******************************************************************************
+;* MMX/SSE2-optimized functions for the RV40 decoder
+;* Copyright (C) 2012 Christophe Gisquet <christophe.gisquet@gmail.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 "x86inc.asm"
+%include "x86util.asm"
+
+SECTION_RODATA
+
+align 16
+shift_round:   times 8 dw 1 << (16 - 6)
+cextern pw_16
+
+SECTION .text
+
+; %1=5bits weights?, %2=dst %3=src1 %4=src3 %5=stride if sse2
+%macro RV40_WCORE  4-5
+    movh       m4, [%3 + 0]
+    movh       m5, [%4 + 0]
+%if %0 == 4
+%define OFFSET mmsize / 2
+%else
+    ; 8x8 block and sse2, stride was provided
+%define OFFSET %5
+%endif
+    movh       m6, [%3 + OFFSET]
+    movh       m7, [%4 + OFFSET]
+
+%if %1 == 0
+    ; 14bits weights
+    punpcklbw  m4, m0
+    punpcklbw  m5, m0
+    punpcklbw  m6, m0
+    punpcklbw  m7, m0
+
+    psllw      m4, 7
+    psllw      m5, 7
+    psllw      m6, 7
+    psllw      m7, 7
+    pmulhw     m4, m3
+    pmulhw     m5, m2
+    pmulhw     m6, m3
+    pmulhw     m7, m2
+
+    paddw      m4, m5
+    paddw      m6, m7
+%else
+    ; 5bits weights
+%if cpuflag(ssse3)
+    punpcklbw  m4, m5
+    punpcklbw  m6, m7
+
+    pmaddubsw  m4, m3
+    pmaddubsw  m6, m3
+%else
+    punpcklbw  m4, m0
+    punpcklbw  m5, m0
+    punpcklbw  m6, m0
+    punpcklbw  m7, m0
+
+    pmullw     m4, m3
+    pmullw     m5, m2
+    pmullw     m6, m3
+    pmullw     m7, m2
+    paddw      m4, m5
+    paddw      m6, m7
+%endif
+
+%endif
+
+    ; bias and shift down
+%if cpuflag(ssse3)
+    pmulhrsw   m4, m1
+    pmulhrsw   m6, m1
+%else
+    paddw      m4, m1
+    paddw      m6, m1
+    psrlw      m4, 5
+    psrlw      m6, 5
+%endif
+
+    packuswb   m4, m6
+%if %0 == 5
+    ; Only called for 8x8 blocks and sse2
+    movh       [%2 +  0], m4
+    movhps     [%2 + %5], m4
+%else
+    mova       [%2], m4
+%endif
+%endmacro
+
+
+%macro MAIN_LOOP   2
+%if mmsize == 8
+    RV40_WCORE %2, r0, r1, r2
+%if %1 == 16
+    RV40_WCORE %2, r0 + 8, r1 + 8, r2 + 8
+%endif
+
+    ; Prepare for next loop
+    add        r0, r5
+    add        r1, r5
+    add        r2, r5
+%else
+%ifidn %1, 8
+    RV40_WCORE %2, r0, r1, r2, r5
+    ; Prepare 2 next lines
+    lea        r0, [r0 + 2 * r5]
+    lea        r1, [r1 + 2 * r5]
+    lea        r2, [r2 + 2 * r5]
+%else
+    RV40_WCORE %2, r0, r1, r2
+    ; Prepare single next line
+    add        r0, r5
+    add        r1, r5
+    add        r2, r5
+%endif
+%endif
+
+    dec        r6
+%endmacro
+
+; rv40_weight_func_%1(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, int stride)
+; %1=size  %2=num of xmm regs
+%macro RV40_WEIGHT  2
+cglobal rv40_weight_func_%1, 6, 7, %2
+%if cpuflag(ssse3)
+    mova       m1, [shift_round]
+%else
+    mova       m1, [pw_16]
+%endif
+    pxor       m0, m0
+    mov        r6, r3
+    or         r6, r4
+    ; The weights are FP0.14 notation of fractions depending on pts.
+    ; For timebases without rounding error (i.e. PAL), the fractions
+    ; can be simplified, and several operations can be avoided.
+    ; Therefore, we check here whether they are multiples of 2^9 for
+    ; those simplifications to occur.
+    and        r6, 0x1FF
+    ; Set loop counter and increments
+%if mmsize == 8
+    mov        r6, %1
+%else
+    mov        r6, (%1 * %1) / mmsize
+%endif
+
+    ; Use result of test now
+    jz .loop_512
+    movd       m2, r3
+    movd       m3, r4
+    SPLATW     m2, m2
+    SPLATW     m3, m3
+
+.loop:
+    MAIN_LOOP  %1, 0
+    jnz        .loop
+    REP_RET
+
+    ; Weights are multiple of 512, which allows some shortcuts
+.loop_512:
+    sar        r3, 9
+    sar        r4, 9
+    movd       m2, r3
+    movd       m3, r4
+%if cpuflag(ssse3)
+    punpcklbw  m3, m2
+    SPLATW     m3, m3
+%else
+    SPLATW     m2, m2
+    SPLATW     m3, m3
+%endif
+.loop2:
+    MAIN_LOOP  %1, 1
+    jnz        .loop2
+    REP_RET
+
+%endmacro
+
+INIT_MMX mmx
+RV40_WEIGHT    8, 0
+RV40_WEIGHT   16, 0
+
+INIT_XMM sse2
+RV40_WEIGHT    8, 8
+RV40_WEIGHT   16, 8
+
+INIT_XMM ssse3
+RV40_WEIGHT    8, 8
+RV40_WEIGHT   16, 8
diff --git a/libavcodec/x86/rv40dsp_init.c b/libavcodec/x86/rv40dsp_init.c
index b429de7293..3d6c6f0fa0 100644
--- a/libavcodec/x86/rv40dsp_init.c
+++ b/libavcodec/x86/rv40dsp_init.c
@@ -40,6 +40,15 @@ void ff_avg_rv40_chroma_mc4_mmx2 (uint8_t *dst, uint8_t *src,
 void ff_avg_rv40_chroma_mc4_3dnow(uint8_t *dst, uint8_t *src,
                                   int stride, int h, int x, int y);
 
+#define DECLARE_WEIGHT(opt) \
+void ff_rv40_weight_func_16_##opt(uint8_t *dst, uint8_t *src1, uint8_t *src2, \
+                                  int w1, int w2, int stride); \
+void ff_rv40_weight_func_8_##opt (uint8_t *dst, uint8_t *src1, uint8_t *src2, \
+                                  int w1, int w2, int stride);
+DECLARE_WEIGHT(mmx)
+DECLARE_WEIGHT(sse2)
+DECLARE_WEIGHT(ssse3)
+
 void ff_rv40dsp_init_x86(RV34DSPContext *c, DSPContext *dsp)
 {
 #if HAVE_YASM
@@ -48,6 +57,8 @@ void ff_rv40dsp_init_x86(RV34DSPContext *c, DSPContext *dsp)
     if (mm_flags & AV_CPU_FLAG_MMX) {
         c->put_chroma_pixels_tab[0] = ff_put_rv40_chroma_mc8_mmx;
         c->put_chroma_pixels_tab[1] = ff_put_rv40_chroma_mc4_mmx;
+        c->rv40_weight_pixels_tab[0] = ff_rv40_weight_func_16_mmx;
+        c->rv40_weight_pixels_tab[1] = ff_rv40_weight_func_8_mmx;
     }
     if (mm_flags & AV_CPU_FLAG_MMX2) {
         c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_mmx2;
@@ -56,5 +67,13 @@ void ff_rv40dsp_init_x86(RV34DSPContext *c, DSPContext *dsp)
         c->avg_chroma_pixels_tab[0] = ff_avg_rv40_chroma_mc8_3dnow;
         c->avg_chroma_pixels_tab[1] = ff_avg_rv40_chroma_mc4_3dnow;
     }
+    if (mm_flags & AV_CPU_FLAG_SSE2) {
+        c->rv40_weight_pixels_tab[0] = ff_rv40_weight_func_16_sse2;
+        c->rv40_weight_pixels_tab[1] = ff_rv40_weight_func_8_sse2;
+    }
+    if (mm_flags & AV_CPU_FLAG_SSSE3) {
+        c->rv40_weight_pixels_tab[0] = ff_rv40_weight_func_16_ssse3;
+        c->rv40_weight_pixels_tab[1] = ff_rv40_weight_func_8_ssse3;
+    }
 #endif
 }