mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-05-05 09:38:07 +02:00
mesa: Replace the float[4] unpack code with util/format's.
The gallium drivers lose 10-15k of binary by not having this duplicated unpack code, but classic i965 gains 215k of binary by pulling in u_format's pack/unpack code. The cost to classic will go down as we drop more of the duplicated code. Reviewed-by: Marek Olšák <marek.olsak@amd.com> Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6297>
This commit is contained in:
Eric Anholt
Marge Bot
parent
83f97cd0be
commit
13ae72aaae
4 changed files with 9 additions and 166 deletions
|
|
@ -25,15 +25,19 @@
|
|||
#ifndef FORMAT_UNPACK_H
|
||||
#define FORMAT_UNPACK_H
|
||||
|
||||
#include "util/format/u_format.h"
|
||||
#include "formats.h"
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
extern void
|
||||
static inline void
|
||||
_mesa_unpack_rgba_row(mesa_format format, uint32_t n,
|
||||
const void *src, float dst[][4]);
|
||||
const void *src, float dst[][4])
|
||||
{
|
||||
util_format_unpack_rgba(format, dst, src, n);
|
||||
}
|
||||
|
||||
extern void
|
||||
_mesa_unpack_ubyte_rgba_row(mesa_format format, uint32_t n,
|
||||
|
|
|
|||
|
|
@ -68,136 +68,6 @@ for f in formats:
|
|||
rgb_formats.append(f)
|
||||
%>
|
||||
|
||||
/* float unpacking functions */
|
||||
|
||||
%for f in rgb_formats:
|
||||
%if f.name in ('MESA_FORMAT_R9G9B9E5_FLOAT', 'MESA_FORMAT_R11G11B10_FLOAT'):
|
||||
<% continue %>
|
||||
%elif f.is_int() and not f.is_normalized():
|
||||
<% continue %>
|
||||
%elif f.is_compressed():
|
||||
<% continue %>
|
||||
%endif
|
||||
|
||||
static inline void
|
||||
unpack_float_${f.short_name()}(const void *void_src, float dst[4])
|
||||
{
|
||||
${f.datatype()} *src = (${f.datatype()} *)void_src;
|
||||
%if f.layout == parser.PACKED:
|
||||
%for c in f.channels:
|
||||
%if c.type != 'x':
|
||||
${c.datatype()} ${c.name} = UNPACK(*src, ${c.shift}, ${c.size});
|
||||
%endif
|
||||
%endfor
|
||||
%elif f.layout == parser.ARRAY:
|
||||
%for (i, c) in enumerate(f.channels):
|
||||
%if c.type != 'x':
|
||||
${c.datatype()} ${c.name} = src[${i}];
|
||||
%endif
|
||||
%endfor
|
||||
%else:
|
||||
<% assert False %>
|
||||
%endif
|
||||
|
||||
%for i in range(4):
|
||||
<% s = f.swizzle[i] %>
|
||||
%if 0 <= s and s <= parser.Swizzle.SWIZZLE_W:
|
||||
<% c = f.channels[s] %>
|
||||
%if c.type == parser.UNSIGNED:
|
||||
%if f.colorspace == 'srgb' and c.name in 'rgb':
|
||||
<% assert c.size == 8 %>
|
||||
dst[${i}] = util_format_srgb_8unorm_to_linear_float(${c.name});
|
||||
%else:
|
||||
dst[${i}] = _mesa_unorm_to_float(${c.name}, ${c.size});
|
||||
%endif
|
||||
%elif c.type == parser.SIGNED:
|
||||
dst[${i}] = _mesa_snorm_to_float(${c.name}, ${c.size});
|
||||
%elif c.type == parser.FLOAT:
|
||||
%if c.size == 32:
|
||||
dst[${i}] = ${c.name};
|
||||
%elif c.size == 16:
|
||||
dst[${i}] = _mesa_half_to_float(${c.name});
|
||||
%else:
|
||||
<% assert False %>
|
||||
%endif
|
||||
%else:
|
||||
<% assert False %>
|
||||
%endif
|
||||
%elif s == parser.Swizzle.SWIZZLE_ZERO:
|
||||
dst[${i}] = 0.0f;
|
||||
%elif s == parser.Swizzle.SWIZZLE_ONE:
|
||||
dst[${i}] = 1.0f;
|
||||
%else:
|
||||
<% assert False %>
|
||||
%endif
|
||||
%endfor
|
||||
}
|
||||
%endfor
|
||||
|
||||
static void
|
||||
unpack_float_r9g9b9e5_float(const void *src, float dst[4])
|
||||
{
|
||||
rgb9e5_to_float3(*(const uint32_t *)src, dst);
|
||||
dst[3] = 1.0f;
|
||||
}
|
||||
|
||||
static void
|
||||
unpack_float_r11g11b10_float(const void *src, float dst[4])
|
||||
{
|
||||
r11g11b10f_to_float3(*(const uint32_t *)src, dst);
|
||||
dst[3] = 1.0f;
|
||||
}
|
||||
|
||||
static void
|
||||
unpack_float_ycbcr(const void *src, float dst[][4], uint32_t n)
|
||||
{
|
||||
uint32_t i;
|
||||
for (i = 0; i < n; i++) {
|
||||
const uint16_t *src0 = ((const uint16_t *) src) + i * 2; /* even */
|
||||
const uint16_t *src1 = src0 + 1; /* odd */
|
||||
const uint8_t y0 = (*src0 >> 8) & 0xff; /* luminance */
|
||||
const uint8_t cb = *src0 & 0xff; /* chroma U */
|
||||
const uint8_t y1 = (*src1 >> 8) & 0xff; /* luminance */
|
||||
const uint8_t cr = *src1 & 0xff; /* chroma V */
|
||||
const uint8_t y = (i & 1) ? y1 : y0; /* choose even/odd luminance */
|
||||
float r = 1.164F * (y - 16) + 1.596F * (cr - 128);
|
||||
float g = 1.164F * (y - 16) - 0.813F * (cr - 128) - 0.391F * (cb - 128);
|
||||
float b = 1.164F * (y - 16) + 2.018F * (cb - 128);
|
||||
r *= (1.0F / 255.0F);
|
||||
g *= (1.0F / 255.0F);
|
||||
b *= (1.0F / 255.0F);
|
||||
dst[i][0] = CLAMP(r, 0.0F, 1.0F);
|
||||
dst[i][1] = CLAMP(g, 0.0F, 1.0F);
|
||||
dst[i][2] = CLAMP(b, 0.0F, 1.0F);
|
||||
dst[i][3] = 1.0F;
|
||||
}
|
||||
}
|
||||
|
||||
static void
|
||||
unpack_float_ycbcr_rev(const void *src, float dst[][4], uint32_t n)
|
||||
{
|
||||
uint32_t i;
|
||||
for (i = 0; i < n; i++) {
|
||||
const uint16_t *src0 = ((const uint16_t *) src) + i * 2; /* even */
|
||||
const uint16_t *src1 = src0 + 1; /* odd */
|
||||
const uint8_t y0 = *src0 & 0xff; /* luminance */
|
||||
const uint8_t cr = (*src0 >> 8) & 0xff; /* chroma V */
|
||||
const uint8_t y1 = *src1 & 0xff; /* luminance */
|
||||
const uint8_t cb = (*src1 >> 8) & 0xff; /* chroma U */
|
||||
const uint8_t y = (i & 1) ? y1 : y0; /* choose even/odd luminance */
|
||||
float r = 1.164F * (y - 16) + 1.596F * (cr - 128);
|
||||
float g = 1.164F * (y - 16) - 0.813F * (cr - 128) - 0.391F * (cb - 128);
|
||||
float b = 1.164F * (y - 16) + 2.018F * (cb - 128);
|
||||
r *= (1.0F / 255.0F);
|
||||
g *= (1.0F / 255.0F);
|
||||
b *= (1.0F / 255.0F);
|
||||
dst[i][0] = CLAMP(r, 0.0F, 1.0F);
|
||||
dst[i][1] = CLAMP(g, 0.0F, 1.0F);
|
||||
dst[i][2] = CLAMP(b, 0.0F, 1.0F);
|
||||
dst[i][3] = 1.0F;
|
||||
}
|
||||
}
|
||||
|
||||
/* ubyte packing functions */
|
||||
|
||||
%for f in rgb_formats:
|
||||
|
|
@ -305,38 +175,6 @@ unpack_int_${f.short_name()}(const void *void_src, uint32_t dst[4])
|
|||
%endfor
|
||||
|
||||
|
||||
void
|
||||
_mesa_unpack_rgba_row(mesa_format format, uint32_t n,
|
||||
const void *src, float dst[][4])
|
||||
{
|
||||
uint8_t *s = (uint8_t *)src;
|
||||
uint32_t i;
|
||||
|
||||
switch (format) {
|
||||
%for f in rgb_formats:
|
||||
%if f.is_compressed():
|
||||
<% continue %>
|
||||
%elif f.is_int() and not f.is_normalized():
|
||||
<% continue %>
|
||||
%endif
|
||||
case ${f.name}:
|
||||
for (i = 0; i < n; ++i) {
|
||||
unpack_float_${f.short_name()}(s, dst[i]);
|
||||
s += ${f.block_size() // 8};
|
||||
}
|
||||
break;
|
||||
%endfor
|
||||
case MESA_FORMAT_YCBCR:
|
||||
unpack_float_ycbcr(src, dst, n);
|
||||
break;
|
||||
case MESA_FORMAT_YCBCR_REV:
|
||||
unpack_float_ycbcr_rev(src, dst, n);
|
||||
break;
|
||||
default:
|
||||
unreachable("bad format");
|
||||
}
|
||||
}
|
||||
|
||||
void
|
||||
_mesa_unpack_ubyte_rgba_row(mesa_format format, uint32_t n,
|
||||
const void *src, uint8_t dst[][4])
|
||||
|
|
|
|||
|
|
@ -55,7 +55,8 @@
|
|||
#include "glformats.h"
|
||||
#include "format_utils.h"
|
||||
#include "format_pack.h"
|
||||
|
||||
#include "format_unpack.h"
|
||||
#include "util/format/u_format.h"
|
||||
|
||||
/**
|
||||
* Flip the 8 bits in each byte of the given array.
|
||||
|
|
|
|||
|
|
@ -38,7 +38,7 @@ test(
|
|||
executable(
|
||||
'main_test',
|
||||
[files_main_test, main_dispatch_h],
|
||||
include_directories : [inc_include, inc_src, inc_mapi, inc_mesa],
|
||||
include_directories : [inc_include, inc_src, inc_mapi, inc_mesa, inc_gallium],
|
||||
dependencies : [idep_gtest, dep_clock, dep_dl, dep_thread, idep_mesautil],
|
||||
link_with : [libmesa_classic, link_main_test],
|
||||
),
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue