dilim/FFmpeg
1
0 Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2026-11-06 05:17:42 +00:00
Code Issues Packages Projects Releases Wiki Activity

swscale/uops: check every product of an input before fusing it

Browse Source 
exact_prod() only checked the products with the two ends of the input
range, so a coefficient times inputs in [0, 1] or [0, 2] passed as
exact although the values in between are not, and with SWS_BITEXACT the
x86 backend fused those products with FMA where the C backend rounds
them. A product is exact for every input when the coefficient is a power
of two, or when the inputs are integers whose bits fit in the
significand next to those of the coefficient.

Signed-off-by: Kacper Michajłow <kasper93@gmail.com>
This commit is contained in:
Kacper Michajłow committed 2026-11-02 18:19:15 +00:00
1 parent 7626d21bf9
commit b8f362cea4
3 files changed
+62 -17
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

libswscale/uops.c
+20 -11
View File
@@ -18,9 +18,12 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <float.h>
#include <math.h>
#include <stdbool.h>
#include "libavutil/avassert.h"
#include "libavutil/intmath.h"
#include "libavutil/mem.h"
#include "libavutil/refstruct.h"
@@ -236,26 +239,32 @@ static SwsPixelType pixel_type_to_int(SwsUOpFlags flags, const SwsPixelType type
return SWS_PIXEL_NONE;
}
static bool exact_product_f32(float a, float b)
static int significant_bits_f32(float f)
{
volatile float prod = a * b;
volatile float result = b ? prod / b : 0.0f;
return !b || result == a;
int exp;
const uint32_t mant = ldexpf(frexpf(fabsf(f), &exp), FLT_MANT_DIG);
return mant ? FLT_MANT_DIG - ff_ctz(mant) : 0;
}
static bool exact_prod(SwsPixel coef, const SwsOpPad *input)
{
if (ff_sws_pixel_type_is_int(input->type))
return true;
else if (!input->min.den || !input->max.den)
return false; /* unknown bounds */
const SwsPixel min = ff_sws_pixel_from_q64(input->type, input->min);
const SwsPixel max = ff_sws_pixel_from_q64(input->type, input->max);
switch (input->type) {
case SWS_PIXEL_F32:
return exact_product_f32(coef.f32, min.f32) &&
exact_product_f32(coef.f32, max.f32);
case SWS_PIXEL_F32: {
const int bits = significant_bits_f32(coef.f32);
if (bits <= 1)
return true; /* power of two */
if (!(input->flags & SWS_COMP_EXACT) || input->min.den != 1 || input->max.den != 1)
return false; /* not only integers, or unknown bounds */
const uint64_t lo = FFABS64U(input->min.num), hi = FFABS64U(input->max.num);
const uint64_t max = FFMAX(lo, hi);
if (max >= 1ULL << FLT_MANT_DIG)
return false;
return bits + (max ? av_log2(max) + 1 : 0) <= FLT_MANT_DIG;
}
}
av_unreachable("Invalid pixel type!");
libswscale/x86/uops_macros.asm
+14 -2
View File
@@ -825,6 +825,7 @@
%macro DECL_F32_LINEAR_FMA 1
DECL_OP %1, f32_linear_fma_x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefa8, 0xfffe8
DECL_OP %1, f32_linear_fma_x_xXx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefa8, 0xfffea
DECL_OP %1, f32_linear_fma_x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefa8, 0xfffef
DECL_OP %1, f32_linear_fma_x_x000x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefae, 0xfffee
DECL_OP %1, f32_linear_fma_x_X000x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefae, 0xfffef
@@ -834,14 +835,24 @@
DECL_OP %1, f32_linear_fma_x_xXX00 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefb8, 0xffffe
DECL_OP %1, f32_linear_fma_y_0x000 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x2, 0x41001, 0xbefbe, 0xfffbf
DECL_OP %1, f32_linear_fma_xyz_xxx0x_xxx0x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfa108
DECL_OP %1, f32_linear_fma_xyz_XXX0x_XxX0x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfbdaf
DECL_OP %1, f32_linear_fma_xyz_xxx0x_xxx0x_Xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfa508
DECL_OP %1, f32_linear_fma_xyz_xxx0x_xxx0x_XXx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfad08
DECL_OP %1, f32_linear_fma_xyz_XxX0x_xxX0x_Xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfa58d
DECL_OP %1, f32_linear_fma_xyz_XXX0x_XXX0x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfbdef
DECL_OP %1, f32_linear_fma_xyz_x0x0x_xxx0x_xx00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbb10a, 0xfb10a
DECL_OP %1, f32_linear_fma_xyz_X0x0x_Xxx0x_Xx00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbb10a, 0xfb52b
DECL_OP %1, f32_linear_fma_xyz_X0X0x_XXX0x_XX00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbb10a, 0xfbdef
DECL_OP %1, f32_linear_fma_xyz_xxx00_xxx0x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfa118
DECL_OP %1, f32_linear_fma_xyz_xXX00_XxX0x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfbdbe
DECL_OP %1, f32_linear_fma_xyz_xxx00_xXx0x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfa158
DECL_OP %1, f32_linear_fma_xyz_xxx00_xxX0x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfa198
DECL_OP %1, f32_linear_fma_xyz_xxx00_xxX0x_Xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfa598
DECL_OP %1, f32_linear_fma_xyz_xxX00_xxX0x_XxX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfb59c
DECL_OP %1, f32_linear_fma_xyz_xXX00_XXX0x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfbdfe
DECL_OP %1, f32_linear_fma_xyz_x000x_0x00x_00x0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbadae, 0xfadae
DECL_OP %1, f32_linear_fma_xyz_X000x_0x00x_00X0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbadae, 0xfbdaf
DECL_OP %1, f32_linear_fma_xyz_X000x_0X00x_00X0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbadae, 0xfbdef
DECL_OP %1, f32_linear_fma_xyz_x0000_0x000_00x00 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbefbe, 0xfefbe
DECL_OP %1, f32_linear_fma_xyz_10x0x_1xx0x_1x00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40421, 0xbb10a, 0xfb52b
DECL_OP %1, f32_linear_fma_xyz_10X0x_1XX0x_1X00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40421, 0xbb10a, 0xfbdef
DECL_OP %1, f32_linear_fma_w_000x0 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x8, 0x01041, 0xbefbe, 0xbffff
DECL_OP %1, f32_linear_fma_xw_x000x_000x0 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x9, 0x01040, 0xbefae, 0xbffee
@@ -850,6 +861,7 @@
DECL_OP %1, f32_linear_fma_xw_xXX00_000x0 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x9, 0x01040, 0xbefb8, 0xbfffe
DECL_OP %1, f32_linear_fma_xyzw_xxx0x_xxx0x_xxx0x_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xba108, 0xba108
DECL_OP %1, f32_linear_fma_xyzw_XXX0x_XXX0x_XXX0x_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xba108, 0xbbdef
DECL_OP %1, f32_linear_fma_xyzw_x0x0x_xxx0x_xx00x_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xbb10a, 0xbb10a
DECL_OP %1, f32_linear_fma_xyzw_X0X0x_XXX0x_XX00x_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xbb10a, 0xbbdef
DECL_OP %1, f32_linear_fma_xyzw_x0000_0x000_00x00_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xbefbe, 0xbefbe
%endmacro
libswscale/x86/uops_macros.h
+28 -4
View File
@@ -1087,6 +1087,7 @@
MACRO(__VA_ARGS__, f32_linear_xyzw_x0000_0x000_00x00_000x0 , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR , .mask = 0xf, .par.lin.one = 0x00000, .par.lin.zero = 0xbefbe)
#define SWS_FOR_F32_LINEAR_FMA(MACRO, ...) \
MACRO(__VA_ARGS__, f32_linear_fma_x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefa8, 0xfffe8) \
MACRO(__VA_ARGS__, f32_linear_fma_x_xXx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefa8, 0xfffea) \
MACRO(__VA_ARGS__, f32_linear_fma_x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefa8, 0xfffef) \
MACRO(__VA_ARGS__, f32_linear_fma_x_x000x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefae, 0xfffee) \
MACRO(__VA_ARGS__, f32_linear_fma_x_X000x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefae, 0xfffef) \
@@ -1096,14 +1097,24 @@
MACRO(__VA_ARGS__, f32_linear_fma_x_xXX00 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x1, 0x41040, 0xbefb8, 0xffffe) \
MACRO(__VA_ARGS__, f32_linear_fma_y_0x000 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x2, 0x41001, 0xbefbe, 0xfffbf) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx0x_xxx0x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfa108) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_XXX0x_XxX0x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfbdaf) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx0x_xxx0x_Xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfa508) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx0x_xxx0x_XXx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfad08) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_XxX0x_xxX0x_Xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfa58d) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_XXX0x_XXX0x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba108, 0xfbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_x0x0x_xxx0x_xx00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbb10a, 0xfb10a) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_X0x0x_Xxx0x_Xx00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbb10a, 0xfb52b) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_X0X0x_XXX0x_XX00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbb10a, 0xfbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx00_xxx0x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfa118) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xXX00_XxX0x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfbdbe) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx00_xXx0x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfa158) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx00_xxX0x_xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfa198) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx00_xxX0x_Xxx0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfa598) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxX00_xxX0x_XxX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfb59c) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xXX00_XXX0x_XXX0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xba118, 0xfbdfe) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_x000x_0x00x_00x0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbadae, 0xfadae) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_X000x_0x00x_00X0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbadae, 0xfbdaf) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_X000x_0X00x_00X0x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbadae, 0xfbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_x0000_0x000_00x00 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40000, 0xbefbe, 0xfefbe) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_10x0x_1xx0x_1x00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40421, 0xbb10a, 0xfb52b) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_10X0x_1XX0x_1X00x , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x7, 0x40421, 0xbb10a, 0xfbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_w_000x0 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x8, 0x01041, 0xbefbe, 0xbffff) \
MACRO(__VA_ARGS__, f32_linear_fma_xw_x000x_000x0 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x9, 0x01040, 0xbefae, 0xbffee) \
@@ -1112,10 +1123,12 @@
MACRO(__VA_ARGS__, f32_linear_fma_xw_xXX00_000x0 , SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0x9, 0x01040, 0xbefb8, 0xbfffe) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_xxx0x_xxx0x_xxx0x_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xba108, 0xba108) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_XXX0x_XXX0x_XXX0x_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xba108, 0xbbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_x0x0x_xxx0x_xx00x_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xbb10a, 0xbb10a) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_X0X0x_XXX0x_XX00x_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xbb10a, 0xbbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_x0000_0x000_00x00_000x0, SWS_PIXEL_F32, SWS_UOP_LINEAR_FMA , 0xf, 0x00000, 0xbefbe, 0xbefbe)
#define SWS_FOR_STRUCT_F32_LINEAR_FMA(MACRO, ...) \
MACRO(__VA_ARGS__, f32_linear_fma_x_xxx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x1, .par.lin.one = 0x41040, .par.lin.zero = 0xbefa8, .par.lin.exact = 0xfffe8) \
MACRO(__VA_ARGS__, f32_linear_fma_x_xXx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x1, .par.lin.one = 0x41040, .par.lin.zero = 0xbefa8, .par.lin.exact = 0xfffea) \
MACRO(__VA_ARGS__, f32_linear_fma_x_XXX0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x1, .par.lin.one = 0x41040, .par.lin.zero = 0xbefa8, .par.lin.exact = 0xfffef) \
MACRO(__VA_ARGS__, f32_linear_fma_x_x000x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x1, .par.lin.one = 0x41040, .par.lin.zero = 0xbefae, .par.lin.exact = 0xfffee) \
MACRO(__VA_ARGS__, f32_linear_fma_x_X000x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x1, .par.lin.one = 0x41040, .par.lin.zero = 0xbefae, .par.lin.exact = 0xfffef) \
@@ -1125,14 +1138,24 @@
MACRO(__VA_ARGS__, f32_linear_fma_x_xXX00 , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x1, .par.lin.one = 0x41040, .par.lin.zero = 0xbefb8, .par.lin.exact = 0xffffe) \
MACRO(__VA_ARGS__, f32_linear_fma_y_0x000 , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x2, .par.lin.one = 0x41001, .par.lin.zero = 0xbefbe, .par.lin.exact = 0xfffbf) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx0x_xxx0x_xxx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba108, .par.lin.exact = 0xfa108) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_XXX0x_XxX0x_XXX0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba108, .par.lin.exact = 0xfbdaf) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx0x_xxx0x_Xxx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba108, .par.lin.exact = 0xfa508) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx0x_xxx0x_XXx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba108, .par.lin.exact = 0xfad08) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_XxX0x_xxX0x_Xxx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba108, .par.lin.exact = 0xfa58d) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_XXX0x_XXX0x_XXX0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba108, .par.lin.exact = 0xfbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_x0x0x_xxx0x_xx00x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xbb10a, .par.lin.exact = 0xfb10a) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_X0x0x_Xxx0x_Xx00x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xbb10a, .par.lin.exact = 0xfb52b) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_X0X0x_XXX0x_XX00x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xbb10a, .par.lin.exact = 0xfbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx00_xxx0x_xxx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba118, .par.lin.exact = 0xfa118) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xXX00_XxX0x_XXX0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba118, .par.lin.exact = 0xfbdbe) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx00_xXx0x_xxx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba118, .par.lin.exact = 0xfa158) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx00_xxX0x_xxx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba118, .par.lin.exact = 0xfa198) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxx00_xxX0x_Xxx0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba118, .par.lin.exact = 0xfa598) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xxX00_xxX0x_XxX0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba118, .par.lin.exact = 0xfb59c) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_xXX00_XXX0x_XXX0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xba118, .par.lin.exact = 0xfbdfe) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_x000x_0x00x_00x0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xbadae, .par.lin.exact = 0xfadae) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_X000x_0x00x_00X0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xbadae, .par.lin.exact = 0xfbdaf) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_X000x_0X00x_00X0x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xbadae, .par.lin.exact = 0xfbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_x0000_0x000_00x00 , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40000, .par.lin.zero = 0xbefbe, .par.lin.exact = 0xfefbe) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_10x0x_1xx0x_1x00x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40421, .par.lin.zero = 0xbb10a, .par.lin.exact = 0xfb52b) \
MACRO(__VA_ARGS__, f32_linear_fma_xyz_10X0x_1XX0x_1X00x , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x7, .par.lin.one = 0x40421, .par.lin.zero = 0xbb10a, .par.lin.exact = 0xfbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_w_000x0 , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x8, .par.lin.one = 0x01041, .par.lin.zero = 0xbefbe, .par.lin.exact = 0xbffff) \
MACRO(__VA_ARGS__, f32_linear_fma_xw_x000x_000x0 , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x9, .par.lin.one = 0x01040, .par.lin.zero = 0xbefae, .par.lin.exact = 0xbffee) \
@@ -1141,6 +1164,7 @@
MACRO(__VA_ARGS__, f32_linear_fma_xw_xXX00_000x0 , .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0x9, .par.lin.one = 0x01040, .par.lin.zero = 0xbefb8, .par.lin.exact = 0xbfffe) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_xxx0x_xxx0x_xxx0x_000x0, .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0xf, .par.lin.one = 0x00000, .par.lin.zero = 0xba108, .par.lin.exact = 0xba108) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_XXX0x_XXX0x_XXX0x_000x0, .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0xf, .par.lin.one = 0x00000, .par.lin.zero = 0xba108, .par.lin.exact = 0xbbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_x0x0x_xxx0x_xx00x_000x0, .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0xf, .par.lin.one = 0x00000, .par.lin.zero = 0xbb10a, .par.lin.exact = 0xbb10a) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_X0X0x_XXX0x_XX00x_000x0, .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0xf, .par.lin.one = 0x00000, .par.lin.zero = 0xbb10a, .par.lin.exact = 0xbbdef) \
MACRO(__VA_ARGS__, f32_linear_fma_xyzw_x0000_0x000_00x00_000x0, .type = SWS_PIXEL_F32, .uop = SWS_UOP_LINEAR_FMA , .mask = 0xf, .par.lin.one = 0x00000, .par.lin.zero = 0xbefbe, .par.lin.exact = 0xbefbe)
#define SWS_FOR_F32_DITHER(MACRO, ...) \