libswscale: Adds ff_hscale8to15_4_avx2 and ff_hscale8to15_X4_avx2 for all filter sizes.

Fixes so that fate under 64 bit Windows passes.

These functions replace all ff_hscale8to15_*_ssse3 when avx2 is available.

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

5 files changed, 171 insertions, 0 deletions

diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index 708facba67..64aa0b9804 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -1105,4 +1105,6 @@ void ff_sws_slice_worker(void *priv, int jobnr, int threadnr,
 //number of extra lines to process
 #define MAX_LINES_AHEAD 4
 
+//shuffle filter and filterPos for hyScale and hcScale filters in avx2
+void ff_shuffle_filter_coefficients(SwsContext *c, int* filterPos, int filterSize, int16_t *filter, int dstW);
 #endif /* SWSCALE_SWSCALE_INTERNAL_H */
diff --git a/libswscale/utils.c b/libswscale/utils.c
index ae92ac9fbc..d4a72d3ce1 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -278,6 +278,41 @@ static const FormatEntry format_entries[] = {
     [AV_PIX_FMT_P416LE]      = { 1, 0 },
 };
 
+void ff_shuffle_filter_coefficients(SwsContext *c, int *filterPos, int filterSize, int16_t *filter, int dstW){
+#if ARCH_X86_64
+    int i, j, k, l;
+    int cpu_flags = av_get_cpu_flags();
+    if (EXTERNAL_AVX2_FAST(cpu_flags)){
+        if ((c->srcBpc == 8) && (c->dstBpc <= 14)){
+            if (dstW % 16 == 0){
+                if (filter != NULL){
+                    for (i = 0; i < dstW; i += 8){
+                        FFSWAP(int, filterPos[i + 2], filterPos[i+4]);
+                        FFSWAP(int, filterPos[i + 3], filterPos[i+5]);
+                    }
+                    if (filterSize > 4){
+                        int16_t *tmp2 = av_malloc(dstW * filterSize * 2);
+                        memcpy(tmp2, filter, dstW * filterSize * 2);
+                        for (i = 0; i < dstW; i += 16){//pixel
+                            for (k = 0; k < filterSize / 4; ++k){//fcoeff
+                                for (j = 0; j < 16; ++j){//inner pixel
+                                    for (l = 0; l < 4; ++l){//coeff
+                                        int from = i * filterSize + j * filterSize + k * 4 + l;
+                                        int to = (i) * filterSize + j * 4 + l + k * 64;
+                                        filter[to] = tmp2[from];
+                                    }
+                                }
+                            }
+                        }
+                        av_free(tmp2);
+                    }
+                }
+            }
+        }
+    }
+#endif
+}
+
 int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
 {
     return (unsigned)pix_fmt < FF_ARRAY_ELEMS(format_entries) ?
@@ -1801,6 +1836,7 @@ av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
                            get_local_pos(c, 0, 0, 0),
                            get_local_pos(c, 0, 0, 0))) < 0)
                 goto fail;
+            ff_shuffle_filter_coefficients(c, c->hLumFilterPos, c->hLumFilterSize, c->hLumFilter, dstW);
             if ((ret = initFilter(&c->hChrFilter, &c->hChrFilterPos,
                            &c->hChrFilterSize, c->chrXInc,
                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
@@ -1810,6 +1846,7 @@ av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
                            get_local_pos(c, c->chrSrcHSubSample, c->src_h_chr_pos, 0),
                            get_local_pos(c, c->chrDstHSubSample, c->dst_h_chr_pos, 0))) < 0)
                 goto fail;
+            ff_shuffle_filter_coefficients(c, c->hChrFilterPos, c->hChrFilterSize, c->hChrFilter, c->chrDstW);
         }
     } // initialize horizontal stuff
 
diff --git a/libswscale/x86/Makefile b/libswscale/x86/Makefile
index bfe383364e..68391494be 100644
--- a/libswscale/x86/Makefile
+++ b/libswscale/x86/Makefile
@@ -11,6 +11,7 @@ OBJS-$(CONFIG_XMM_CLOBBER_TEST) += x86/w64xmmtest.o
 X86ASM-OBJS                     += x86/input.o                          \
                                    x86/output.o                         \
                                    x86/scale.o                          \
+                                   x86/scale_avx2.o                          \
                                    x86/rgb_2_rgb.o                      \
                                    x86/yuv_2_rgb.o                      \
                                    x86/yuv2yuvX.o                       \
diff --git a/libswscale/x86/scale_avx2.asm b/libswscale/x86/scale_avx2.asm
new file mode 100644
index 0000000000..4e4fe5d794
--- /dev/null
+++ b/libswscale/x86/scale_avx2.asm
@@ -0,0 +1,112 @@
+;******************************************************************************
+;* x86-optimized horizontal line scaling functions
+;* Copyright 2020 Google LLC
+;* Copyright (c) 2011 Ronald S. Bultje <rsbultje@gmail.com>
+;*
+;* This file is part of FFmpeg.
+;*
+;* FFmpeg 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.
+;*
+;* FFmpeg 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 FFmpeg; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+SECTION_RODATA 32
+
+swizzle: dd 0, 4, 1, 5, 2, 6, 3, 7
+four: times 8 dd 4
+
+SECTION .text
+
+;-----------------------------------------------------------------------------
+; horizontal line scaling
+;
+; void hscale8to15_<filterSize>_<opt>
+;                   (SwsContext *c, int16_t *dst,
+;                    int dstW, const uint8_t *src,
+;                    const int16_t *filter,
+;                    const int32_t *filterPos, int filterSize);
+;
+; Scale one horizontal line. Input is 8-bit width Filter is 14 bits. Output is
+; 15 bits (in int16_t). Each output pixel is generated from $filterSize input
+; pixels, the position of the first pixel is given in filterPos[nOutputPixel].
+;-----------------------------------------------------------------------------
+
+%macro SCALE_FUNC 1
+cglobal hscale8to15_%1, 7, 9, 16, pos0, dst, w, srcmem, filter, fltpos, fltsize, count, inner
+  pxor m0, m0
+  mova m15, [swizzle]
+  mov countq, $0
+  movsxd wq, wd
+%ifidn %1, X4
+  mova m14, [four]
+  shr fltsized, 2
+%endif
+.loop:
+  movu m1, [fltposq]
+  movu m2, [fltposq+32]
+%ifidn %1, X4
+  pxor m9, m9
+  pxor m10, m10
+  pxor m11, m11
+  pxor m12, m12
+  mov innerq, $0
+.innerloop:
+%endif
+  vpcmpeqd  m13, m13
+  vpgatherdd m3,[srcmemq + m1], m13
+  vpcmpeqd  m13, m13
+  vpgatherdd m4,[srcmemq + m2], m13
+  vpunpcklbw m5, m3, m0
+  vpunpckhbw m6, m3, m0
+  vpunpcklbw m7, m4, m0
+  vpunpckhbw m8, m4, m0
+  vpmaddwd m5, m5, [filterq]
+  vpmaddwd m6, m6, [filterq + 32]
+  vpmaddwd m7, m7, [filterq + 64]
+  vpmaddwd m8, m8, [filterq + 96]
+  add filterq, $80
+%ifidn %1, X4
+  paddd m9, m5
+  paddd m10, m6
+  paddd m11, m7
+  paddd m12, m8
+  paddd m1, m14
+  paddd m2, m14
+  add innerq, $1
+  cmp innerq, fltsizeq
+  jl .innerloop
+  vphaddd m5, m9, m10
+  vphaddd m6, m11, m12
+%else
+  vphaddd m5, m5, m6
+  vphaddd m6, m7, m8
+%endif
+  vpsrad  m5, 7
+  vpsrad  m6, 7
+  vpackssdw m5, m5, m6
+  vpermd m5, m15, m5
+  vmovdqu [dstq + countq * 2], m5
+  add fltposq, $40
+  add countq, $10
+  cmp countq, wq
+  jl .loop
+REP_RET
+%endmacro
+
+%if ARCH_X86_64
+INIT_YMM avx2
+SCALE_FUNC 4
+SCALE_FUNC X4
+%endif
diff --git a/libswscale/x86/swscale.c b/libswscale/x86/swscale.c
index 0848a31461..164b06d6ba 100644
--- a/libswscale/x86/swscale.c
+++ b/libswscale/x86/swscale.c
@@ -276,6 +276,9 @@ SCALE_FUNCS_SSE(sse2);
 SCALE_FUNCS_SSE(ssse3);
 SCALE_FUNCS_SSE(sse4);
 
+SCALE_FUNC(4, 8, 15, avx2);
+SCALE_FUNC(X4, 8, 15, avx2);
+
 #define VSCALEX_FUNC(size, opt) \
 void ff_yuv2planeX_ ## size ## _ ## opt(const int16_t *filter, int filterSize, \
                                         const int16_t **src, uint8_t *dest, int dstW, \
@@ -568,6 +571,22 @@ switch(c->dstBpc){ \
     }
 
 #if ARCH_X86_64
+#define ASSIGN_AVX2_SCALE_FUNC(hscalefn, filtersize) \
+    switch (filtersize) { \
+    case 4:  hscalefn = ff_hscale8to15_4_avx2; break; \
+    default:  hscalefn = ff_hscale8to15_X4_avx2; break; \
+             break; \
+    }
+
+    if (EXTERNAL_AVX2_FAST(cpu_flags)){
+      if ((c->srcBpc == 8) && (c->dstBpc <= 14)){
+        if(c->chrDstW % 16 == 0)
+          ASSIGN_AVX2_SCALE_FUNC(c->hcScale, c->hChrFilterSize);
+        if(c->dstW % 16 == 0)
+          ASSIGN_AVX2_SCALE_FUNC(c->hyScale, c->hLumFilterSize);
+      }
+    }
+
     if (EXTERNAL_AVX2_FAST(cpu_flags)) {
         switch (c->dstFormat) {
         case AV_PIX_FMT_NV12: