avcodec/x86/lossless_videodsp: Port lorens add_hfyu_left_prediction_ssse3/sse4 to 16bit

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/x86/lossless_videodsp.asm

@@ -1,5 +1,6 @@
 ;******************************************************************************
 ;* SIMD lossless video DSP utils
+;* Copyright (c) 2008 Loren Merritt
 ;* Copyright (c) 2014 Michael Niedermayer
 ;*
 ;* This file is part of FFmpeg.
@@ -21,6 +22,13 @@
 
 %include "libavutil/x86/x86util.asm"
 
+SECTION_RODATA
+
+pb_ef: times 8 db 14,15
+pb_67: times 8 db  6, 7
+pb_zzzz2323zzzzabab: db -1,-1,-1,-1, 2, 3, 2, 3,-1,-1,-1,-1,10,11,10,11
+pb_zzzzzzzz67676767: db -1,-1,-1,-1,-1,-1,-1,-1, 6, 7, 6, 7, 6, 7, 6, 7
+
 SECTION_TEXT
 
 %macro ADD_INT16_LOOP 1 ; %1 = is_aligned
@@ -84,3 +92,79 @@ cglobal add_int16, 4,4,5, dst, src, mask, w
     ADD_INT16_LOOP 1
 .unaligned:
     ADD_INT16_LOOP 0
+
+%macro ADD_HFYU_LEFT_LOOP_INT16 2 ; %1 = dst_is_aligned, %2 = src_is_aligned
+    add     wq, wq
+    add     srcq, wq
+    add     dstq, wq
+    neg     wq
+%%.loop:
+%if %2
+    mova    m1, [srcq+wq]
+%else
+    movu    m1, [srcq+wq]
+%endif
+    mova    m2, m1
+    pslld   m1, 16
+    paddw   m1, m2
+    mova    m2, m1
+
+    pshufb  m1, m3
+    paddw   m1, m2
+    pshufb  m0, m5
+%if mmsize == 16
+    mova    m2, m1
+    pshufb  m1, m4
+    paddw   m1, m2
+%endif
+    paddw   m0, m1
+    pand    m0, m7
+%if %1
+    mova    [dstq+wq], m0
+%else
+    movq    [dstq+wq], m0
+    movhps  [dstq+wq+8], m0
+%endif
+    add     wq, mmsize
+    jl %%.loop
+    mov     eax, mmsize-1
+    sub     eax, wd
+    mov     wd, eax
+    shl     wd, 8
+    lea     eax, [wd+eax-1]
+    movd    m1, eax
+    pshufb  m0, m1
+    movd    eax, m0
+    RET
+%endmacro
+
+; int add_hfyu_left_prediction_int16(uint16_t *dst, const uint16_t *src, unsigned mask, int w, int left)
+INIT_MMX ssse3
+cglobal add_hfyu_left_prediction_int16, 4,4,8, dst, src, mask, w, left
+.skip_prologue:
+    mova    m5, [pb_67]
+    mova    m3, [pb_zzzz2323zzzzabab]
+    movd    m0, leftm
+    psllq   m0, 48
+    movd    m7, maskm
+    SPLATW  m7 ,m7
+    ADD_HFYU_LEFT_LOOP_INT16 1, 1
+
+INIT_XMM sse4
+cglobal add_hfyu_left_prediction_int16, 4,4,8, dst, src, mask, w, left
+    mova    m5, [pb_ef]
+    mova    m4, [pb_zzzzzzzz67676767]
+    mova    m3, [pb_zzzz2323zzzzabab]
+    movd    m0, leftm
+    pslldq  m0, 14
+    movd    m7, maskm
+    SPLATW  m7 ,m7
+    test    srcq, 15
+    jnz .src_unaligned
+    test    dstq, 15
+    jnz .dst_unaligned
+    ADD_HFYU_LEFT_LOOP_INT16 1, 1
+.dst_unaligned:
+    ADD_HFYU_LEFT_LOOP_INT16 0, 1
+.src_unaligned:
+    ADD_HFYU_LEFT_LOOP_INT16 0, 0

libavcodec/x86/lossless_videodsp_init.c

@@ -23,6 +23,8 @@
 
 void ff_add_int16_mmx(uint16_t *dst, const uint16_t *src, unsigned mask, int w);
 void ff_add_int16_sse2(uint16_t *dst, const uint16_t *src, unsigned mask, int w);
+int ff_add_hfyu_left_prediction_int16_ssse3(uint16_t *dst, const uint16_t *src, unsigned mask, int w, int acc);
+int ff_add_hfyu_left_prediction_int16_sse4(uint16_t *dst, const uint16_t *src, unsigned mask, int w, int acc);
 
 void ff_llviddsp_init_x86(LLVidDSPContext *c)
 {
@@ -35,4 +37,12 @@ void ff_llviddsp_init_x86(LLVidDSPContext *c)
     if (EXTERNAL_SSE2(cpu_flags)) {
         c->add_int16 = ff_add_int16_sse2;
     }
+
+    if (EXTERNAL_SSSE3(cpu_flags)) {
+        c->add_hfyu_left_prediction_int16 = ff_add_hfyu_left_prediction_int16_ssse3;
+    }
+
+    if (EXTERNAL_SSE4(cpu_flags)) {
+        c->add_hfyu_left_prediction_int16 = ff_add_hfyu_left_prediction_int16_sse4;
+    }
 }