huffyuvencdsp: move functions only used by huffyuv from lossless_videodsp

Signed-off-by: James Almer <jamrial@gmail.com>

4 files changed, 128 insertions, 66 deletions

diff --git a/libavcodec/x86/huffyuvencdsp.asm b/libavcodec/x86/huffyuvencdsp.asm
index a55a1de65d..78ad202249 100644
--- a/libavcodec/x86/huffyuvencdsp.asm
+++ b/libavcodec/x86/huffyuvencdsp.asm
@@ -148,3 +148,116 @@ DIFF_BYTES_PROLOGUE
     DIFF_BYTES_BODY    u, u
 %undef i
 %endif
+
+%macro INT16_LOOP 2 ; %1 = a/u (aligned/unaligned), %2 = add/sub
+    movd    m4, maskd
+    SPLATW  m4, m4
+    add     wd, wd
+    test    wq, 2*mmsize - 1
+    jz %%.tomainloop
+    push  tmpq
+%%.wordloop:
+    sub     wq, 2
+%ifidn %2, add
+    mov   tmpw, [srcq+wq]
+    add   tmpw, [dstq+wq]
+%else
+    mov   tmpw, [src1q+wq]
+    sub   tmpw, [src2q+wq]
+%endif
+    and   tmpw, maskw
+    mov     [dstq+wq], tmpw
+    test    wq, 2*mmsize - 1
+    jnz %%.wordloop
+    pop   tmpq
+%%.tomainloop:
+%ifidn %2, add
+    add     srcq, wq
+%else
+    add     src1q, wq
+    add     src2q, wq
+%endif
+    add     dstq, wq
+    neg     wq
+    jz      %%.end
+%%.loop:
+%ifidn %2, add
+    mov%1   m0, [srcq+wq]
+    mov%1   m1, [dstq+wq]
+    mov%1   m2, [srcq+wq+mmsize]
+    mov%1   m3, [dstq+wq+mmsize]
+%else
+    mov%1   m0, [src1q+wq]
+    mov%1   m1, [src2q+wq]
+    mov%1   m2, [src1q+wq+mmsize]
+    mov%1   m3, [src2q+wq+mmsize]
+%endif
+    p%2w    m0, m1
+    p%2w    m2, m3
+    pand    m0, m4
+    pand    m2, m4
+    mov%1   [dstq+wq]       , m0
+    mov%1   [dstq+wq+mmsize], m2
+    add     wq, 2*mmsize
+    jl %%.loop
+%%.end:
+    RET
+%endmacro
+
+%if ARCH_X86_32
+INIT_MMX mmx
+cglobal diff_int16, 5,5,5, dst, src1, src2, mask, w, tmp
+    INT16_LOOP a, sub
+%endif
+
+INIT_XMM sse2
+cglobal diff_int16, 5,5,5, dst, src1, src2, mask, w, tmp
+    test src1q, mmsize-1
+    jnz .unaligned
+    test src2q, mmsize-1
+    jnz .unaligned
+    test dstq, mmsize-1
+    jnz .unaligned
+    INT16_LOOP a, sub
+.unaligned:
+    INT16_LOOP u, sub
+
+INIT_MMX mmxext
+cglobal sub_hfyu_median_pred_int16, 7,7,0, dst, src1, src2, mask, w, left, left_top
+    add      wd, wd
+    movd    mm7, maskd
+    SPLATW  mm7, mm7
+    movq    mm0, [src1q]
+    movq    mm2, [src2q]
+    psllq   mm0, 16
+    psllq   mm2, 16
+    movd    mm6, [left_topq]
+    por     mm0, mm6
+    movd    mm6, [leftq]
+    por     mm2, mm6
+    xor     maskq, maskq
+.loop:
+    movq    mm1, [src1q + maskq]
+    movq    mm3, [src2q + maskq]
+    movq    mm4, mm2
+    psubw   mm2, mm0
+    paddw   mm2, mm1
+    pand    mm2, mm7
+    movq    mm5, mm4
+    pmaxsw  mm4, mm1
+    pminsw  mm1, mm5
+    pminsw  mm4, mm2
+    pmaxsw  mm4, mm1
+    psubw   mm3, mm4
+    pand    mm3, mm7
+    movq    [dstq + maskq], mm3
+    add     maskq, 8
+    movq    mm0, [src1q + maskq - 2]
+    movq    mm2, [src2q + maskq - 2]
+    cmp     maskq, wq
+        jb .loop
+    movzx maskd, word [src1q + wq - 2]
+    mov [left_topq], maskd
+    movzx maskd, word [src2q + wq - 2]
+    mov [leftq], maskd
+    RET
diff --git a/libavcodec/x86/huffyuvencdsp_mmx.c b/libavcodec/x86/huffyuvencdsp_mmx.c
index 9767b212da..2402021823 100644
--- a/libavcodec/x86/huffyuvencdsp_mmx.c
+++ b/libavcodec/x86/huffyuvencdsp_mmx.c
@@ -24,6 +24,7 @@
 
 #include "libavutil/attributes.h"
 #include "libavutil/cpu.h"
+#include "libavutil/pixdesc.h"
 #include "libavutil/x86/asm.h"
 #include "libavutil/x86/cpu.h"
 #include "libavcodec/huffyuvencdsp.h"
@@ -35,6 +36,12 @@ void ff_diff_bytes_sse2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
                         intptr_t w);
 void ff_diff_bytes_avx2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
                         intptr_t w);
+void ff_diff_int16_mmx (uint16_t *dst, const uint16_t *src1, const uint16_t *src2,
+                        unsigned mask, int w);
+void ff_diff_int16_sse2(uint16_t *dst, const uint16_t *src1, const uint16_t *src2,
+                        unsigned mask, int w);
+void ff_sub_hfyu_median_pred_int16_mmxext(uint16_t *dst, const uint16_t *src1, const uint16_t *src2,
+                                          unsigned mask, int w, int *left, int *left_top);
 
 #if HAVE_INLINE_ASM
 
@@ -80,12 +87,14 @@ static void sub_hfyu_median_pred_mmxext(uint8_t *dst, const uint8_t *src1,
 
 #endif /* HAVE_INLINE_ASM */
 
-av_cold void ff_huffyuvencdsp_init_x86(HuffYUVEncDSPContext *c)
+av_cold void ff_huffyuvencdsp_init_x86(HuffYUVEncDSPContext *c, AVCodecContext *avctx)
 {
     av_unused int cpu_flags = av_get_cpu_flags();
+    const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(avctx->pix_fmt);
 
     if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags)) {
         c->diff_bytes = ff_diff_bytes_mmx;
+        c->diff_int16 = ff_diff_int16_mmx;
     }
 
 #if HAVE_INLINE_ASM
@@ -94,8 +103,13 @@ av_cold void ff_huffyuvencdsp_init_x86(HuffYUVEncDSPContext *c)
     }
 #endif /* HAVE_INLINE_ASM */
 
+    if (EXTERNAL_MMXEXT(cpu_flags) && pix_desc && pix_desc->comp[0].depth<16) {
+        c->sub_hfyu_median_pred_int16 = ff_sub_hfyu_median_pred_int16_mmxext;
+    }
+
     if (EXTERNAL_SSE2(cpu_flags)) {
         c->diff_bytes = ff_diff_bytes_sse2;
+        c->diff_int16 = ff_diff_int16_sse2;
     }
 
     if (EXTERNAL_AVX2_FAST(cpu_flags)) {
diff --git a/libavcodec/x86/lossless_videodsp.asm b/libavcodec/x86/lossless_videodsp.asm
index a6ce5fe62b..bcc40ec061 100644
--- a/libavcodec/x86/lossless_videodsp.asm
+++ b/libavcodec/x86/lossless_videodsp.asm
@@ -288,25 +288,6 @@ cglobal add_int16, 4,4,5, dst, src, mask, w, tmp
 .unaligned:
     INT16_LOOP u, add
 
-%if ARCH_X86_32
-INIT_MMX mmx
-cglobal diff_int16, 5,5,5, dst, src1, src2, mask, w, tmp
-    INT16_LOOP a, sub
-%endif
-
-INIT_XMM sse2
-cglobal diff_int16, 5,5,5, dst, src1, src2, mask, w, tmp
-    test src1q, mmsize-1
-    jnz .unaligned
-    test src2q, mmsize-1
-    jnz .unaligned
-    test dstq, mmsize-1
-    jnz .unaligned
-    INT16_LOOP a, sub
-.unaligned:
-    INT16_LOOP u, sub
-
-
 %macro ADD_HFYU_LEFT_LOOP_INT16 2 ; %1 = dst alignment (a/u), %2 = src alignment (a/u)
     add     wd, wd
     add     srcq, wq
@@ -443,42 +424,3 @@ cglobal add_hfyu_median_pred_int16, 7,7,0, dst, top, diff, mask, w, left, left_t
     movzx   r2d, word [topq-2]
     mov [left_topq], r2d
     RET
-
-cglobal sub_hfyu_median_pred_int16, 7,7,0, dst, src1, src2, mask, w, left, left_top
-    add      wd, wd
-    movd    mm7, maskd
-    SPLATW  mm7, mm7
-    movq    mm0, [src1q]
-    movq    mm2, [src2q]
-    psllq   mm0, 16
-    psllq   mm2, 16
-    movd    mm6, [left_topq]
-    por     mm0, mm6
-    movd    mm6, [leftq]
-    por     mm2, mm6
-    xor     maskq, maskq
-.loop:
-    movq    mm1, [src1q + maskq]
-    movq    mm3, [src2q + maskq]
-    movq    mm4, mm2
-    psubw   mm2, mm0
-    paddw   mm2, mm1
-    pand    mm2, mm7
-    movq    mm5, mm4
-    pmaxsw  mm4, mm1
-    pminsw  mm1, mm5
-    pminsw  mm4, mm2
-    pmaxsw  mm4, mm1
-    psubw   mm3, mm4
-    pand    mm3, mm7
-    movq    [dstq + maskq], mm3
-    add     maskq, 8
-    movq    mm0, [src1q + maskq - 2]
-    movq    mm2, [src2q + maskq - 2]
-    cmp     maskq, wq
-        jb .loop
-    movzx maskd, word [src1q + wq - 2]
-    mov [left_topq], maskd
-    movzx maskd, word [src2q + wq - 2]
-    mov [leftq], maskd
-    RET
diff --git a/libavcodec/x86/lossless_videodsp_init.c b/libavcodec/x86/lossless_videodsp_init.c
index 465feef81b..2dc662d8b1 100644
--- a/libavcodec/x86/lossless_videodsp_init.c
+++ b/libavcodec/x86/lossless_videodsp_init.c
@@ -41,12 +41,9 @@ int  ff_add_left_pred_sse4(uint8_t *dst, const uint8_t *src,
 
 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);
-void ff_diff_int16_mmx (uint16_t *dst, const uint16_t *src1, const uint16_t *src2, unsigned mask, int w);
-void ff_diff_int16_sse2(uint16_t *dst, const uint16_t *src1, const uint16_t *src2, unsigned mask, int w);
 int ff_add_hfyu_left_pred_int16_ssse3(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc);
 int ff_add_hfyu_left_pred_int16_sse4(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc);
 void ff_add_hfyu_median_pred_int16_mmxext(uint16_t *dst, const uint16_t *top, const uint16_t *diff, unsigned mask, int w, int *left, int *left_top);
-void ff_sub_hfyu_median_pred_int16_mmxext(uint16_t *dst, const uint16_t *src1, const uint16_t *src2, unsigned mask, int w, int *left, int *left_top);
 
 #if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32
 static void add_median_pred_cmov(uint8_t *dst, const uint8_t *top,
@@ -98,9 +95,7 @@ void ff_llviddsp_init_x86(LLVidDSPContext *c, AVCodecContext *avctx)
 
     if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags)) {
         c->add_bytes = ff_add_bytes_mmx;
-
         c->add_int16 = ff_add_int16_mmx;
-        c->diff_int16 = ff_diff_int16_mmx;
     }
 
     if (ARCH_X86_32 && EXTERNAL_MMXEXT(cpu_flags)) {
@@ -111,7 +106,6 @@ void ff_llviddsp_init_x86(LLVidDSPContext *c, AVCodecContext *avctx)
 
     if (EXTERNAL_MMXEXT(cpu_flags) && pix_desc && pix_desc->comp[0].depth<16) {
         c->add_hfyu_median_pred_int16 = ff_add_hfyu_median_pred_int16_mmxext;
-        c->sub_hfyu_median_pred_int16 = ff_sub_hfyu_median_pred_int16_mmxext;
     }
 
     if (EXTERNAL_SSE2(cpu_flags)) {
@@ -119,7 +113,6 @@ void ff_llviddsp_init_x86(LLVidDSPContext *c, AVCodecContext *avctx)
         c->add_median_pred = ff_add_median_pred_sse2;
 
         c->add_int16 = ff_add_int16_sse2;
-        c->diff_int16 = ff_diff_int16_sse2;
     }
 
     if (EXTERNAL_SSSE3(cpu_flags)) {