x86/hevc: add 12bits support for MC

cherry picked from commit 3fcb7a4595a6f40100a22110a5805e3b7510c0fd

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

3 files changed, 155 insertions, 6 deletions

diff --git a/libavcodec/x86/hevc_mc.asm b/libavcodec/x86/hevc_mc.asm
index 5cf37d0194..81b7d9e4e3 100644
--- a/libavcodec/x86/hevc_mc.asm
+++ b/libavcodec/x86/hevc_mc.asm
@@ -21,11 +21,14 @@
 %include "libavutil/x86/x86util.asm"
 
 SECTION_RODATA
-pw_8:                   times 8 dw 512
-pw_10:                  times 8 dw 2048
-pw_bi_8:                times 8 dw 256
-pw_bi_10:               times 8 dw 1024
-max_pixels_10:          times 8  dw 1023
+pw_8:                   times 8 dw (1 << 9)
+pw_10:                  times 8 dw (1 << 11)
+pw_12:                  times 8 dw (1 << 13)
+pw_bi_8:                times 8 dw (1 << 8)
+pw_bi_10:               times 8 dw (1 << 10)
+pw_bi_12:               times 8 dw (1 << 12)
+max_pixels_10:          times 8  dw ((1 << 10)-1)
+max_pixels_12:          times 8  dw ((1 << 12)-1)
 zero:                   times 4  dd 0
 one_per_32:             times 4  dd 1
 
@@ -51,6 +54,7 @@ hevc_epel_filters_%4_%1 times %2 d%3 -2, 58
 
 EPEL_TABLE  8, 8, b, sse4
 EPEL_TABLE 10, 4, w, sse4
+EPEL_TABLE 12, 4, w, sse4
 
 %macro QPEL_TABLE 4
 hevc_qpel_filters_%4_%1 times %2 d%3  -1,  4
@@ -69,6 +73,7 @@ hevc_qpel_filters_%4_%1 times %2 d%3  -1,  4
 
 QPEL_TABLE  8, 8, b, sse4
 QPEL_TABLE 10, 4, w, sse4
+QPEL_TABLE 12, 4, w, sse4
 
 %define hevc_qpel_filters_sse4_14 hevc_qpel_filters_sse4_10
 
@@ -295,6 +300,29 @@ QPEL_TABLE 10, 4, w, sse4
 %endif
 %endmacro
 
+%macro PEL_12STORE2 3
+    movd           [%1], %2
+%endmacro
+%macro PEL_12STORE4 3
+    movq           [%1], %2
+%endmacro
+%macro PEL_12STORE6 3
+    movq           [%1], %2
+    psrldq            %2, 8
+    movd         [%1+8], %2
+%endmacro
+%macro PEL_12STORE8 3
+    movdqa         [%1], %2
+%endmacro
+%macro PEL_12STORE12 3
+    movdqa         [%1], %2
+    movq        [%1+16], %3
+%endmacro
+%macro PEL_12STORE16 3
+    PEL_12STORE8      %1, %2, %3
+    movdqa       [%1+16], %3
+%endmacro
+
 %macro PEL_10STORE2 3
     movd           [%1], %2
 %endmacro
@@ -1229,6 +1257,11 @@ WEIGHTING_FUNCS 4, 10
 WEIGHTING_FUNCS 6, 10
 WEIGHTING_FUNCS 8, 10
 
+WEIGHTING_FUNCS 2, 12
+WEIGHTING_FUNCS 4, 12
+WEIGHTING_FUNCS 6, 12
+WEIGHTING_FUNCS 8, 12
+
 HEVC_PUT_HEVC_PEL_PIXELS  2, 8
 HEVC_PUT_HEVC_PEL_PIXELS  4, 8
 HEVC_PUT_HEVC_PEL_PIXELS  6, 8
@@ -1241,6 +1274,10 @@ HEVC_PUT_HEVC_PEL_PIXELS 4, 10
 HEVC_PUT_HEVC_PEL_PIXELS 6, 10
 HEVC_PUT_HEVC_PEL_PIXELS 8, 10
 
+HEVC_PUT_HEVC_PEL_PIXELS 2, 12
+HEVC_PUT_HEVC_PEL_PIXELS 4, 12
+HEVC_PUT_HEVC_PEL_PIXELS 6, 12
+HEVC_PUT_HEVC_PEL_PIXELS 8, 12
 
 HEVC_PUT_HEVC_EPEL 2,  8
 HEVC_PUT_HEVC_EPEL 4,  8
@@ -1255,6 +1292,10 @@ HEVC_PUT_HEVC_EPEL 4, 10
 HEVC_PUT_HEVC_EPEL 6, 10
 HEVC_PUT_HEVC_EPEL 8, 10
 
+HEVC_PUT_HEVC_EPEL 2, 12
+HEVC_PUT_HEVC_EPEL 4, 12
+HEVC_PUT_HEVC_EPEL 6, 12
+HEVC_PUT_HEVC_EPEL 8, 12
 
 HEVC_PUT_HEVC_EPEL_HV 2,  8
 HEVC_PUT_HEVC_EPEL_HV 4,  8
@@ -1266,6 +1307,10 @@ HEVC_PUT_HEVC_EPEL_HV 4, 10
 HEVC_PUT_HEVC_EPEL_HV 6, 10
 HEVC_PUT_HEVC_EPEL_HV 8, 10
 
+HEVC_PUT_HEVC_EPEL_HV 2, 12
+HEVC_PUT_HEVC_EPEL_HV 4, 12
+HEVC_PUT_HEVC_EPEL_HV 6, 12
+HEVC_PUT_HEVC_EPEL_HV 8, 12
 
 HEVC_PUT_HEVC_QPEL 4,  8
 HEVC_PUT_HEVC_QPEL 8,  8
@@ -1275,6 +1320,9 @@ HEVC_PUT_HEVC_QPEL 16, 8
 HEVC_PUT_HEVC_QPEL 4, 10
 HEVC_PUT_HEVC_QPEL 8, 10
 
+HEVC_PUT_HEVC_QPEL 4, 12
+HEVC_PUT_HEVC_QPEL 8, 12
+
 HEVC_PUT_HEVC_QPEL_HV 2, 8
 HEVC_PUT_HEVC_QPEL_HV 4, 8
 HEVC_PUT_HEVC_QPEL_HV 6, 8
@@ -1285,4 +1333,9 @@ HEVC_PUT_HEVC_QPEL_HV 4, 10
 HEVC_PUT_HEVC_QPEL_HV 6, 10
 HEVC_PUT_HEVC_QPEL_HV 8, 10
 
+HEVC_PUT_HEVC_QPEL_HV 2, 12
+HEVC_PUT_HEVC_QPEL_HV 4, 12
+HEVC_PUT_HEVC_QPEL_HV 6, 12
+HEVC_PUT_HEVC_QPEL_HV 8, 12
+
 %endif ; ARCH_X86_64
diff --git a/libavcodec/x86/hevcdsp.h b/libavcodec/x86/hevcdsp.h
index 661a860bd8..6062d8e055 100644
--- a/libavcodec/x86/hevcdsp.h
+++ b/libavcodec/x86/hevcdsp.h
@@ -95,33 +95,41 @@ void ff_hevc_put_hevc_bi_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t dstst
 ///////////////////////////////////////////////////////////////////////////////
 EPEL_PROTOTYPES(pel_pixels ,  8, sse4);
 EPEL_PROTOTYPES(pel_pixels , 10, sse4);
+EPEL_PROTOTYPES(pel_pixels , 12, sse4);
 ///////////////////////////////////////////////////////////////////////////////
 // EPEL
 ///////////////////////////////////////////////////////////////////////////////
 EPEL_PROTOTYPES(epel_h ,  8, sse4);
 EPEL_PROTOTYPES(epel_h , 10, sse4);
+EPEL_PROTOTYPES(epel_h , 12, sse4);
 
 EPEL_PROTOTYPES(epel_v ,  8, sse4);
 EPEL_PROTOTYPES(epel_v , 10, sse4);
+EPEL_PROTOTYPES(epel_v , 12, sse4);
 
 EPEL_PROTOTYPES(epel_hv ,  8, sse4);
 EPEL_PROTOTYPES(epel_hv , 10, sse4);
+EPEL_PROTOTYPES(epel_hv , 12, sse4);
 
 ///////////////////////////////////////////////////////////////////////////////
 // QPEL
 ///////////////////////////////////////////////////////////////////////////////
 QPEL_PROTOTYPES(qpel_h ,  8, sse4);
 QPEL_PROTOTYPES(qpel_h , 10, sse4);
+QPEL_PROTOTYPES(qpel_h , 12, sse4);
 
 QPEL_PROTOTYPES(qpel_v,  8, sse4);
 QPEL_PROTOTYPES(qpel_v, 10, sse4);
+QPEL_PROTOTYPES(qpel_v, 12, sse4);
 
 QPEL_PROTOTYPES(qpel_hv,  8, sse4);
 QPEL_PROTOTYPES(qpel_hv, 10, sse4);
+QPEL_PROTOTYPES(qpel_hv, 12, sse4);
 
 
 WEIGHTING_PROTOTYPES(8, sse4);
 WEIGHTING_PROTOTYPES(10, sse4);
+WEIGHTING_PROTOTYPES(12, sse4);
 
 ///////////////////////////////////////////////////////////////////////////////
 // IDCT
diff --git a/libavcodec/x86/hevcdsp_init.c b/libavcodec/x86/hevcdsp_init.c
index ebe9847048..6fb94aaf0b 100644
--- a/libavcodec/x86/hevcdsp_init.c
+++ b/libavcodec/x86/hevcdsp_init.c
@@ -167,13 +167,18 @@ mc_rep_funcs(pel_pixels, 8, 16, 64, sse4);
 mc_rep_funcs(pel_pixels, 8, 16, 48, sse4);
 mc_rep_funcs(pel_pixels, 8, 16, 32, sse4);
 mc_rep_funcs(pel_pixels, 8,  8, 24, sse4);
-
 mc_rep_funcs(pel_pixels,10,  8, 64, sse4);
 mc_rep_funcs(pel_pixels,10,  8, 48, sse4);
 mc_rep_funcs(pel_pixels,10,  8, 32, sse4);
 mc_rep_funcs(pel_pixels,10,  8, 24, sse4);
 mc_rep_funcs(pel_pixels,10,  8, 16, sse4);
 mc_rep_funcs(pel_pixels,10,  4, 12, sse4);
+mc_rep_funcs(pel_pixels,12,  8, 64, sse4);
+mc_rep_funcs(pel_pixels,12,  8, 48, sse4);
+mc_rep_funcs(pel_pixels,12,  8, 32, sse4);
+mc_rep_funcs(pel_pixels,12,  8, 24, sse4);
+mc_rep_funcs(pel_pixels,12,  8, 16, sse4);
+mc_rep_funcs(pel_pixels,12,  4, 12, sse4);
 
 mc_rep_funcs(epel_h, 8, 16, 64, sse4);
 mc_rep_funcs(epel_h, 8, 16, 48, sse4);
@@ -185,6 +190,12 @@ mc_rep_funcs(epel_h,10,  8, 32, sse4);
 mc_rep_funcs(epel_h,10,  8, 24, sse4);
 mc_rep_funcs(epel_h,10,  8, 16, sse4);
 mc_rep_funcs(epel_h,10,  4, 12, sse4);
+mc_rep_funcs(epel_h,12,  8, 64, sse4);
+mc_rep_funcs(epel_h,12,  8, 48, sse4);
+mc_rep_funcs(epel_h,12,  8, 32, sse4);
+mc_rep_funcs(epel_h,12,  8, 24, sse4);
+mc_rep_funcs(epel_h,12,  8, 16, sse4);
+mc_rep_funcs(epel_h,12,  4, 12, sse4);
 mc_rep_funcs(epel_v, 8, 16, 64, sse4);
 mc_rep_funcs(epel_v, 8, 16, 48, sse4);
 mc_rep_funcs(epel_v, 8, 16, 32, sse4);
@@ -195,6 +206,12 @@ mc_rep_funcs(epel_v,10,  8, 32, sse4);
 mc_rep_funcs(epel_v,10,  8, 24, sse4);
 mc_rep_funcs(epel_v,10,  8, 16, sse4);
 mc_rep_funcs(epel_v,10,  4, 12, sse4);
+mc_rep_funcs(epel_v,12,  8, 64, sse4);
+mc_rep_funcs(epel_v,12,  8, 48, sse4);
+mc_rep_funcs(epel_v,12,  8, 32, sse4);
+mc_rep_funcs(epel_v,12,  8, 24, sse4);
+mc_rep_funcs(epel_v,12,  8, 16, sse4);
+mc_rep_funcs(epel_v,12,  4, 12, sse4);
 mc_rep_funcs(epel_hv, 8,  8, 64, sse4);
 mc_rep_funcs(epel_hv, 8,  8, 48, sse4);
 mc_rep_funcs(epel_hv, 8,  8, 32, sse4);
@@ -207,6 +224,12 @@ mc_rep_funcs(epel_hv,10,  8, 32, sse4);
 mc_rep_funcs(epel_hv,10,  8, 24, sse4);
 mc_rep_funcs(epel_hv,10,  8, 16, sse4);
 mc_rep_funcs(epel_hv,10,  4, 12, sse4);
+mc_rep_funcs(epel_hv,12,  8, 64, sse4);
+mc_rep_funcs(epel_hv,12,  8, 48, sse4);
+mc_rep_funcs(epel_hv,12,  8, 32, sse4);
+mc_rep_funcs(epel_hv,12,  8, 24, sse4);
+mc_rep_funcs(epel_hv,12,  8, 16, sse4);
+mc_rep_funcs(epel_hv,12,  4, 12, sse4);
 
 mc_rep_funcs(qpel_h, 8, 16, 64, sse4);
 mc_rep_funcs(qpel_h, 8, 16, 48, sse4);
@@ -218,6 +241,12 @@ mc_rep_funcs(qpel_h,10,  8, 32, sse4);
 mc_rep_funcs(qpel_h,10,  8, 24, sse4);
 mc_rep_funcs(qpel_h,10,  8, 16, sse4);
 mc_rep_funcs(qpel_h,10,  4, 12, sse4);
+mc_rep_funcs(qpel_h,12,  8, 64, sse4);
+mc_rep_funcs(qpel_h,12,  8, 48, sse4);
+mc_rep_funcs(qpel_h,12,  8, 32, sse4);
+mc_rep_funcs(qpel_h,12,  8, 24, sse4);
+mc_rep_funcs(qpel_h,12,  8, 16, sse4);
+mc_rep_funcs(qpel_h,12,  4, 12, sse4);
 mc_rep_funcs(qpel_v, 8, 16, 64, sse4);
 mc_rep_funcs(qpel_v, 8, 16, 48, sse4);
 mc_rep_funcs(qpel_v, 8, 16, 32, sse4);
@@ -228,6 +257,12 @@ mc_rep_funcs(qpel_v,10,  8, 32, sse4);
 mc_rep_funcs(qpel_v,10,  8, 24, sse4);
 mc_rep_funcs(qpel_v,10,  8, 16, sse4);
 mc_rep_funcs(qpel_v,10,  4, 12, sse4);
+mc_rep_funcs(qpel_v,12,  8, 64, sse4);
+mc_rep_funcs(qpel_v,12,  8, 48, sse4);
+mc_rep_funcs(qpel_v,12,  8, 32, sse4);
+mc_rep_funcs(qpel_v,12,  8, 24, sse4);
+mc_rep_funcs(qpel_v,12,  8, 16, sse4);
+mc_rep_funcs(qpel_v,12,  4, 12, sse4);
 mc_rep_funcs(qpel_hv, 8,  8, 64, sse4);
 mc_rep_funcs(qpel_hv, 8,  8, 48, sse4);
 mc_rep_funcs(qpel_hv, 8,  8, 32, sse4);
@@ -240,6 +275,12 @@ mc_rep_funcs(qpel_hv,10,  8, 32, sse4);
 mc_rep_funcs(qpel_hv,10,  8, 24, sse4);
 mc_rep_funcs(qpel_hv,10,  8, 16, sse4);
 mc_rep_funcs(qpel_hv,10,  4, 12, sse4);
+mc_rep_funcs(qpel_hv,12,  8, 64, sse4);
+mc_rep_funcs(qpel_hv,12,  8, 48, sse4);
+mc_rep_funcs(qpel_hv,12,  8, 32, sse4);
+mc_rep_funcs(qpel_hv,12,  8, 24, sse4);
+mc_rep_funcs(qpel_hv,12,  8, 16, sse4);
+mc_rep_funcs(qpel_hv,12,  4, 12, sse4);
 
 #define mc_rep_uni_w(bitd, step, W, opt) \
 void ff_hevc_put_hevc_uni_w##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t dststride, int16_t *_src, ptrdiff_t _srcstride,\
@@ -270,6 +311,13 @@ mc_rep_uni_w(10, 8, 32, sse4);
 mc_rep_uni_w(10, 8, 48, sse4);
 mc_rep_uni_w(10, 8, 64, sse4);
 
+mc_rep_uni_w(12, 6, 12, sse4);
+mc_rep_uni_w(12, 8, 16, sse4);
+mc_rep_uni_w(12, 8, 24, sse4);
+mc_rep_uni_w(12, 8, 32, sse4);
+mc_rep_uni_w(12, 8, 48, sse4);
+mc_rep_uni_w(12, 8, 64, sse4);
+
 #define mc_rep_bi_w(bitd, step, W, opt) \
 void ff_hevc_put_hevc_bi_w##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t dststride, int16_t *_src, ptrdiff_t _srcstride, \
                                               int16_t *_src2, ptrdiff_t _src2stride, int height,                        \
@@ -302,6 +350,13 @@ mc_rep_bi_w(10, 8, 32, sse4);
 mc_rep_bi_w(10, 8, 48, sse4);
 mc_rep_bi_w(10, 8, 64, sse4);
 
+mc_rep_bi_w(12, 6, 12, sse4);
+mc_rep_bi_w(12, 8, 16, sse4);
+mc_rep_bi_w(12, 8, 24, sse4);
+mc_rep_bi_w(12, 8, 32, sse4);
+mc_rep_bi_w(12, 8, 48, sse4);
+mc_rep_bi_w(12, 8, 64, sse4);
+
 #define mc_uni_w_func(name, bitd, W, opt) \
 void ff_hevc_put_hevc_uni_w_##name##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t _dststride,         \
                                                       uint8_t *_src, ptrdiff_t _srcstride,          \
@@ -348,6 +403,17 @@ mc_uni_w_funcs(qpel_h, 10, sse4);
 mc_uni_w_funcs(qpel_v, 10, sse4);
 mc_uni_w_funcs(qpel_hv, 10, sse4);
 
+mc_uni_w_funcs(pel_pixels, 12, sse4);
+mc_uni_w_func(pel_pixels, 12, 6, sse4);
+mc_uni_w_funcs(epel_h, 12, sse4);
+mc_uni_w_func(epel_h, 12, 6, sse4);
+mc_uni_w_funcs(epel_v, 12, sse4);
+mc_uni_w_func(epel_v, 12, 6, sse4);
+mc_uni_w_funcs(epel_hv, 12, sse4);
+mc_uni_w_func(epel_hv, 12, 6, sse4);
+mc_uni_w_funcs(qpel_h, 12, sse4);
+mc_uni_w_funcs(qpel_v, 12, sse4);
+mc_uni_w_funcs(qpel_hv, 12, sse4);
 
 #define mc_bi_w_func(name, bitd, W, opt) \
 void ff_hevc_put_hevc_bi_w_##name##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t _dststride,           \
@@ -397,6 +463,17 @@ mc_bi_w_funcs(qpel_h, 10, sse4);
 mc_bi_w_funcs(qpel_v, 10, sse4);
 mc_bi_w_funcs(qpel_hv, 10, sse4);
 
+mc_bi_w_funcs(pel_pixels, 12, sse4);
+mc_bi_w_func(pel_pixels, 12, 6, sse4);
+mc_bi_w_funcs(epel_h, 12, sse4);
+mc_bi_w_func(epel_h, 12, 6, sse4);
+mc_bi_w_funcs(epel_v, 12, sse4);
+mc_bi_w_func(epel_v, 12, 6, sse4);
+mc_bi_w_funcs(epel_hv, 12, sse4);
+mc_bi_w_func(epel_hv, 12, 6, sse4);
+mc_bi_w_funcs(qpel_h, 12, sse4);
+mc_bi_w_funcs(qpel_v, 12, sse4);
+mc_bi_w_funcs(qpel_hv, 12, sse4);
 #endif //ARCH_X86_64 && HAVE_SSE4_EXTERNAL
 
 
@@ -515,5 +592,16 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
             c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_12_ssse3;
             c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_12_ssse3;
         }
+        if (EXTERNAL_SSE4(mm_flags) && ARCH_X86_64) {
+            EPEL_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 12, sse4);
+            EPEL_LINKS(c->put_hevc_epel, 0, 1, epel_h,     12, sse4);
+            EPEL_LINKS(c->put_hevc_epel, 1, 0, epel_v,     12, sse4);
+            EPEL_LINKS(c->put_hevc_epel, 1, 1, epel_hv,    12, sse4);
+
+            QPEL_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 12, sse4);
+            QPEL_LINKS(c->put_hevc_qpel, 0, 1, qpel_h,     12, sse4);
+            QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v,     12, sse4);
+            QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv,    12, sse4);
+        }
     }
 }