x86: simple_idct(_put): 10bits versions

Modeled from the prores version. Clips to [0;1023] and is bitexact.
Bitexactness requires to add offsets in different places compared to
prores or C, and makes the function approximately 2% slower.

For 16 frames of a DNxHD 4:2:2 10bits test sequence:

C:    60861 decicycles in idct, 1048205 runs,    371 skips
sse2: 27567 decicycles in idct, 1048216 runs,    360 skips
avx:  26272 decicycles in idct, 1048171 runs,    405 skips

The add version is not implemented, so the corresponding dsp
function is set to NULL to make it clear in a code executing it.

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

libavcodec/x86/Makefile

@@ -126,6 +126,7 @@ YASM-OBJS-$(CONFIG_QPELDSP)            += x86/qpeldsp.o                 \
                                           x86/fpel.o                    \
                                           x86/qpel.o
 YASM-OBJS-$(CONFIG_RV34DSP)            += x86/rv34dsp.o
+YASM-OBJS-$(CONFIG_IDCTDSP)            += x86/simple_idct10.o
 YASM-OBJS-$(CONFIG_VIDEODSP)           += x86/videodsp.o
 YASM-OBJS-$(CONFIG_VP3DSP)             += x86/vp3dsp.o
 YASM-OBJS-$(CONFIG_VP8DSP)             += x86/vp8dsp.o                  \

libavcodec/x86/idctdsp_init.c

@@ -85,4 +85,24 @@ av_cold void ff_idctdsp_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
         c->put_pixels_clamped        = ff_put_pixels_clamped_sse2;
         c->add_pixels_clamped        = ff_add_pixels_clamped_sse2;
     }
+
+    if (ARCH_X86_64 &&
+        avctx->bits_per_raw_sample == 10 && avctx->lowres == 0 &&
+        (avctx->idct_algo == FF_IDCT_AUTO ||
+         avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
+         avctx->idct_algo == FF_IDCT_SIMPLE)) {
+        if (EXTERNAL_SSE2(cpu_flags)) {
+            c->idct_put  = ff_simple_idct10_put_sse2;
+            c->idct_add  = NULL;
+            c->idct      = ff_simple_idct10_sse2;
+            c->perm_type = FF_IDCT_PERM_TRANSPOSE;
+
+        }
+        if (EXTERNAL_AVX(cpu_flags)) {
+            c->idct_put  = ff_simple_idct10_put_avx;
+            c->idct_add  = NULL;
+            c->idct      = ff_simple_idct10_avx;
+            c->perm_type = FF_IDCT_PERM_TRANSPOSE;
+        }
+    }
 }

libavcodec/x86/proresdsp.asm

@@ -37,17 +37,17 @@ cextern pw_1019
 
 section .text align=16
 
-%macro idct_put_fn 0
+%macro idct_fn 0
 cglobal prores_idct_put_10, 4, 4, 15
-    IDCT_PUT_FN    pw_1, 15, pw_88, 18, pw_4, pw_1019, r3
+    IDCT_FN    pw_1, 15, pw_88, 18, pw_4, pw_1019, r3
     RET
 %endmacro
 
 INIT_XMM sse2
-idct_put_fn
+idct_fn
 %if HAVE_AVX_EXTERNAL
 INIT_XMM avx
-idct_put_fn
+idct_fn
 %endif
 
 %endif

libavcodec/x86/simple_idct.h

@@ -25,4 +25,10 @@ void ff_simple_idct_mmx(int16_t *block);
 void ff_simple_idct_add_mmx(uint8_t *dest, int line_size, int16_t *block);
 void ff_simple_idct_put_mmx(uint8_t *dest, int line_size, int16_t *block);
 
+void ff_simple_idct10_sse2(int16_t *block);
+void ff_simple_idct10_avx(int16_t *block);
+
+void ff_simple_idct10_put_sse2(uint8_t *dest, int line_size, int16_t *block);
+void ff_simple_idct10_put_avx(uint8_t *dest, int line_size, int16_t *block);
+
 #endif /* AVCODEC_X86_SIMPLE_IDCT_H */

libavcodec/x86/simple_idct10.asm

@@ -0,0 +1,58 @@
+;******************************************************************************
+;* x86-SIMD-optimized IDCT for prores
+;* this is identical to "simple" IDCT written by Michael Niedermayer
+;* except for the clip range
+;*
+;* Copyright (c) 2011 Ronald S. Bultje <rsbultje@gmail.com>
+;* Copyright (c) 2015 Christophe Gisquet
+;*
+;* 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
+;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+%if ARCH_X86_64
+
+SECTION_RODATA
+
+cextern pw_16
+cextern pw_1023
+pd_round_12: times 4 dd 1<<(12-1)
+pd_round_19: times 4 dd 1<<(19-1)
+
+%include "libavcodec/x86/simple_idct10_template.asm"
+
+section .text align=16
+
+%macro idct_fn 0
+cglobal simple_idct10, 1, 1, 16
+    IDCT_FN    "", 12, "", 19
+    RET
+
+cglobal simple_idct10_put, 3, 3, 16
+    IDCT_FN    "", 12, "", 19, 0, pw_1023
+    RET
+%endmacro
+
+INIT_XMM sse2
+idct_fn
+%if HAVE_AVX_EXTERNAL
+INIT_XMM avx
+idct_fn
+%endif
+
+%endif

libavcodec/x86/simple_idct10_template.asm

@@ -128,18 +128,18 @@ cextern w7_min_w5
     psubd       m3,  m9            ; a1[4-7] intermediate
 
     ; load/store
-    mova   [r2+  0], m0
+    mova   [COEFFS+  0], m0
-    mova   [r2+ 32], m2
+    mova   [COEFFS+ 32], m2
-    mova   [r2+ 64], m4
+    mova   [COEFFS+ 64], m4
-    mova   [r2+ 96], m6
+    mova   [COEFFS+ 96], m6
-    mova        m10,[r2+ 16]       ; { row[1] }[0-7]
+    mova        m10,[COEFFS+ 16]       ; { row[1] }[0-7]
-    mova        m8, [r2+ 48]       ; { row[3] }[0-7]
+    mova        m8, [COEFFS+ 48]       ; { row[3] }[0-7]
-    mova        m13,[r2+ 80]       ; { row[5] }[0-7]
+    mova        m13,[COEFFS+ 80]       ; { row[5] }[0-7]
-    mova        m14,[r2+112]       ; { row[7] }[0-7]
+    mova        m14,[COEFFS+112]       ; { row[7] }[0-7]
-    mova   [r2+ 16], m1
+    mova   [COEFFS+ 16], m1
-    mova   [r2+ 48], m3
+    mova   [COEFFS+ 48], m3
-    mova   [r2+ 80], m5
+    mova   [COEFFS+ 80], m5
-    mova   [r2+112], m7
+    mova   [COEFFS+112], m7
 %if %0 == 3
     pmullw      m10,[%3+ 16]
     pmullw      m8, [%3+ 48]
@@ -210,17 +210,17 @@ cextern w7_min_w5
     ; row[5] = (a2 - b2) >> 15;
     ; row[3] = (a3 + b3) >> 15;
     ; row[4] = (a3 - b3) >> 15;
-    mova        m8, [r2+ 0]        ; a0[0-3]
+    mova        m8, [COEFFS+ 0]        ; a0[0-3]
-    mova        m9, [r2+16]        ; a0[4-7]
+    mova        m9, [COEFFS+16]        ; a0[4-7]
     SUMSUB_SHPK m8,  m9,  m10, m11, m0,  m1,  %2
-    mova        m0, [r2+32]        ; a1[0-3]
+    mova        m0, [COEFFS+32]        ; a1[0-3]
-    mova        m1, [r2+48]        ; a1[4-7]
+    mova        m1, [COEFFS+48]        ; a1[4-7]
     SUMSUB_SHPK m0,  m1,  m9,  m11, m2,  m3,  %2
-    mova        m1, [r2+64]        ; a2[0-3]
+    mova        m1, [COEFFS+64]        ; a2[0-3]
-    mova        m2, [r2+80]        ; a2[4-7]
+    mova        m2, [COEFFS+80]        ; a2[4-7]
     SUMSUB_SHPK m1,  m2,  m11, m3,  m4,  m5,  %2
-    mova        m2, [r2+96]        ; a3[0-3]
+    mova        m2, [COEFFS+96]        ; a3[0-3]
-    mova        m3, [r2+112]       ; a3[4-7]
+    mova        m3, [COEFFS+112]       ; a3[4-7]
     SUMSUB_SHPK m2,  m3,  m4,  m5,  m6,  m7,  %2
 %endmacro
 
@@ -235,15 +235,21 @@ cextern w7_min_w5
 ; %6 = max pixel value
 ; %7 = qmat (for prores)
 
-%macro IDCT_PUT_FN 6-7
+%macro IDCT_FN 4-7
+%if %0 == 4
+    ; No clamping, means pure idct
+%xdefine COEFFS r0
+%else
     movsxd      r1,  r1d
+%xdefine COEFFS r2
+%endif
 
     ; for (i = 0; i < 8; i++)
     ;     idctRowCondDC(block + i*8);
-    mova        m10,[r2+ 0]        ; { row[0] }[0-7]
+    mova        m10,[COEFFS+ 0]        ; { row[0] }[0-7]
-    mova        m8, [r2+32]        ; { row[2] }[0-7]
+    mova        m8, [COEFFS+32]        ; { row[2] }[0-7]
-    mova        m13,[r2+64]        ; { row[4] }[0-7]
+    mova        m13,[COEFFS+64]        ; { row[4] }[0-7]
-    mova        m12,[r2+96]        ; { row[6] }[0-7]
+    mova        m12,[COEFFS+96]        ; { row[6] }[0-7]
 
 %if %0 == 7
     pmullw      m10,[%7+ 0]
@@ -258,10 +264,10 @@ cextern w7_min_w5
 
     ; transpose for second part of IDCT
     TRANSPOSE8x8W 8, 0, 1, 2, 4, 11, 9, 10, 3
-    mova   [r2+ 16], m0
+    mova   [COEFFS+ 16], m0
-    mova   [r2+ 48], m2
+    mova   [COEFFS+ 48], m2
-    mova   [r2+ 80], m11
+    mova   [COEFFS+ 80], m11
-    mova   [r2+112], m10
+    mova   [COEFFS+112], m10
     SWAP         8,  10
     SWAP         1,   8
     SWAP         4,  13
@@ -272,6 +278,17 @@ cextern w7_min_w5
     IDCT_1D     %3, %4
 
     ; clip/store
+%if %0 == 4
+    ; No clamping, means pure idct
+    mova  [r0+  0], m8
+    mova  [r0+ 16], m0
+    mova  [r0+ 32], m1
+    mova  [r0+ 48], m2
+    mova  [r0+ 64], m4
+    mova  [r0+ 80], m11
+    mova  [r0+ 96], m9
+    mova  [r0+112], m10
+%else
 %ifidn %5, 0
     pxor        m3, m3
 %else
@@ -305,6 +322,7 @@ cextern w7_min_w5
     mova  [r0+r1  ], m11
     mova  [r0+r1*2], m9
     mova  [r0+r2  ], m10
+%endif
 %endmacro
 
 %endif