softpipe: add image support to softpipe (v3)

This adds support for ARB_shader_image_load_store to softpipe.

v2: add RESQ support (Ilia)
v3: constify, cleanup internals, add some comments (Brian).

Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This commit is contained in:
Dave Airlie 2016-03-22 07:59:35 +10:00
parent 0d1f679ded
commit eb9ad9faa3
14 changed files with 928 additions and 12 deletions

View File

@ -497,8 +497,10 @@ tgsi_exec_get_shader_param(enum pipe_shader_cap param)
case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
return 0;
case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
return PIPE_MAX_SHADER_IMAGES;
case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
return 32;
}

View File

@ -10,6 +10,7 @@ C_SOURCES := \
sp_flush.h \
sp_fs_exec.c \
sp_fs.h \
sp_image.c \
sp_limits.h \
sp_prim_vbuf.c \
sp_prim_vbuf.h \
@ -31,6 +32,7 @@ C_SOURCES := \
sp_state_blend.c \
sp_state_clip.c \
sp_state_derived.c \
sp_state_image.c \
sp_state.h \
sp_state_rasterizer.c \
sp_state_sampler.c \

View File

@ -50,7 +50,7 @@
#include "sp_query.h"
#include "sp_screen.h"
#include "sp_tex_sample.h"
#include "sp_image.h"
static void
softpipe_destroy( struct pipe_context *pipe )
@ -199,6 +199,10 @@ softpipe_create_context(struct pipe_screen *screen,
softpipe->tgsi.sampler[i] = sp_create_tgsi_sampler();
}
for (i = 0; i < PIPE_SHADER_TYPES; i++) {
softpipe->tgsi.image[i] = sp_create_tgsi_image();
}
softpipe->dump_fs = debug_get_bool_option( "SOFTPIPE_DUMP_FS", FALSE );
softpipe->dump_gs = debug_get_bool_option( "SOFTPIPE_DUMP_GS", FALSE );
@ -216,6 +220,7 @@ softpipe_create_context(struct pipe_screen *screen,
softpipe_init_streamout_funcs(&softpipe->pipe);
softpipe_init_texture_funcs( &softpipe->pipe );
softpipe_init_vertex_funcs(&softpipe->pipe);
softpipe_init_image_funcs(&softpipe->pipe);
softpipe->pipe.set_framebuffer_state = softpipe_set_framebuffer_state;
@ -223,7 +228,8 @@ softpipe_create_context(struct pipe_screen *screen,
softpipe->pipe.clear = softpipe_clear;
softpipe->pipe.flush = softpipe_flush_wrapped;
softpipe->pipe.texture_barrier = softpipe_texture_barrier;
softpipe->pipe.memory_barrier = softpipe_memory_barrier;
softpipe->pipe.render_condition = softpipe_render_condition;
/*
@ -272,6 +278,16 @@ softpipe_create_context(struct pipe_screen *screen,
(struct tgsi_sampler *)
softpipe->tgsi.sampler[PIPE_SHADER_GEOMETRY]);
draw_image(softpipe->draw,
PIPE_SHADER_VERTEX,
(struct tgsi_image *)
softpipe->tgsi.image[PIPE_SHADER_VERTEX]);
draw_image(softpipe->draw,
PIPE_SHADER_GEOMETRY,
(struct tgsi_image *)
softpipe->tgsi.image[PIPE_SHADER_GEOMETRY]);
if (debug_get_bool_option( "SOFTPIPE_NO_RAST", FALSE ))
softpipe->no_rast = TRUE;

View File

@ -83,6 +83,7 @@ struct softpipe_context {
struct pipe_scissor_state scissors[PIPE_MAX_VIEWPORTS];
struct pipe_sampler_view *sampler_views[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_SAMPLER_VIEWS];
struct pipe_image_view images[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_IMAGES];
struct pipe_viewport_state viewports[PIPE_MAX_VIEWPORTS];
struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
struct pipe_index_buffer index_buffer;
@ -172,6 +173,7 @@ struct softpipe_context {
/** TGSI exec things */
struct {
struct sp_tgsi_sampler *sampler[PIPE_SHADER_TYPES];
struct sp_tgsi_image *image[PIPE_SHADER_TYPES];
} tgsi;
struct tgsi_exec_machine *fs_machine;

View File

@ -168,3 +168,29 @@ softpipe_flush_resource(struct pipe_context *pipe,
return TRUE;
}
void softpipe_texture_barrier(struct pipe_context *pipe)
{
struct softpipe_context *softpipe = softpipe_context(pipe);
uint i, sh;
for (sh = 0; sh < Elements(softpipe->tex_cache); sh++) {
for (i = 0; i < softpipe->num_sampler_views[sh]; i++) {
sp_flush_tex_tile_cache(softpipe->tex_cache[sh][i]);
}
}
for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++)
if (softpipe->cbuf_cache[i])
sp_flush_tile_cache(softpipe->cbuf_cache[i]);
if (softpipe->zsbuf_cache)
sp_flush_tile_cache(softpipe->zsbuf_cache);
softpipe->dirty_render_cache = FALSE;
}
void softpipe_memory_barrier(struct pipe_context *pipe, unsigned flags)
{
softpipe_texture_barrier(pipe);
}

View File

@ -55,4 +55,6 @@ softpipe_flush_resource(struct pipe_context *pipe,
boolean cpu_access,
boolean do_not_block);
void softpipe_texture_barrier(struct pipe_context *pipe);
void softpipe_memory_barrier(struct pipe_context *pipe, unsigned flags);
#endif

View File

@ -62,14 +62,15 @@ sp_exec_fragment_shader(const struct sp_fragment_shader_variant *var)
static void
exec_prepare( const struct sp_fragment_shader_variant *var,
struct tgsi_exec_machine *machine,
struct tgsi_sampler *sampler )
struct tgsi_sampler *sampler,
struct tgsi_image *image )
{
/*
* Bind tokens/shader to the interpreter's machine state.
*/
tgsi_exec_machine_bind_shader(machine,
var->tokens,
sampler, NULL);
sampler, image);
}
@ -127,6 +128,7 @@ exec_run( const struct sp_fragment_shader_variant *var,
/* convert 0 to 1.0 and 1 to -1.0 */
machine->Face = (float) (quad->input.facing * -2 + 1);
machine->NonHelperMask = quad->inout.mask;
quad->inout.mask &= tgsi_exec_machine_run( machine );
if (quad->inout.mask == 0)
return FALSE;

View File

@ -0,0 +1,762 @@
/*
* Copyright 2016 Red Hat.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* on the rights to use, copy, modify, merge, publish, distribute, sub
* license, and/or sell copies of the Software, and to permit persons to whom
* the Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
* THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#include "sp_context.h"
#include "sp_image.h"
#include "sp_texture.h"
#include "util/u_format.h"
/*
* Get the offset into the base image
* first element for a buffer or layer/level for texture.
*/
static uint32_t
get_image_offset(const struct softpipe_resource *spr,
const struct pipe_image_view *iview,
enum pipe_format format, unsigned r_coord)
{
int base_layer = 0;
if (spr->base.target == PIPE_BUFFER)
return iview->u.buf.first_element * util_format_get_blocksize(format);
if (spr->base.target == PIPE_TEXTURE_1D_ARRAY ||
spr->base.target == PIPE_TEXTURE_2D_ARRAY ||
spr->base.target == PIPE_TEXTURE_CUBE_ARRAY ||
spr->base.target == PIPE_TEXTURE_CUBE ||
spr->base.target == PIPE_TEXTURE_3D)
base_layer = r_coord + iview->u.tex.first_layer;
return softpipe_get_tex_image_offset(spr, iview->u.tex.level, base_layer);
}
/*
* Does this texture instruction have a layer or depth parameter.
*/
static inline bool
has_layer_or_depth(unsigned tgsi_tex_instr)
{
return (tgsi_tex_instr == TGSI_TEXTURE_3D ||
tgsi_tex_instr == TGSI_TEXTURE_CUBE ||
tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY ||
tgsi_tex_instr == TGSI_TEXTURE_2D_ARRAY ||
tgsi_tex_instr == TGSI_TEXTURE_CUBE_ARRAY ||
tgsi_tex_instr == TGSI_TEXTURE_2D_ARRAY_MSAA);
}
/*
* Is this texture instruction a single non-array coordinate.
*/
static inline bool
has_1coord(unsigned tgsi_tex_instr)
{
return (tgsi_tex_instr == TGSI_TEXTURE_BUFFER ||
tgsi_tex_instr == TGSI_TEXTURE_1D ||
tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY);
}
/*
* check the bounds vs w/h/d
*/
static inline bool
bounds_check(int width, int height, int depth,
int s, int t, int r)
{
if (s < 0 || s >= width)
return false;
if (t < 0 || t >= height)
return false;
if (r < 0 || r >= depth)
return false;
return true;
}
/*
* Checks if the texture target compatible with the image resource
* pipe target.
*/
static inline bool
has_compat_target(unsigned pipe_target, unsigned tgsi_target)
{
switch (pipe_target) {
case PIPE_TEXTURE_1D:
if (tgsi_target == TGSI_TEXTURE_1D)
return true;
break;
case PIPE_TEXTURE_2D:
if (tgsi_target == TGSI_TEXTURE_2D)
return true;
break;
case PIPE_TEXTURE_RECT:
if (tgsi_target == TGSI_TEXTURE_RECT)
return true;
break;
case PIPE_TEXTURE_3D:
if (tgsi_target == TGSI_TEXTURE_3D ||
tgsi_target == TGSI_TEXTURE_2D)
return true;
break;
case PIPE_TEXTURE_CUBE:
if (tgsi_target == TGSI_TEXTURE_CUBE ||
tgsi_target == TGSI_TEXTURE_2D)
return true;
break;
case PIPE_TEXTURE_1D_ARRAY:
if (tgsi_target == TGSI_TEXTURE_1D ||
tgsi_target == TGSI_TEXTURE_1D_ARRAY)
return true;
break;
case PIPE_TEXTURE_2D_ARRAY:
if (tgsi_target == TGSI_TEXTURE_2D ||
tgsi_target == TGSI_TEXTURE_2D_ARRAY)
return true;
break;
case PIPE_TEXTURE_CUBE_ARRAY:
if (tgsi_target == TGSI_TEXTURE_CUBE ||
tgsi_target == TGSI_TEXTURE_CUBE_ARRAY ||
tgsi_target == TGSI_TEXTURE_2D)
return true;
break;
case PIPE_BUFFER:
return (tgsi_target == TGSI_TEXTURE_BUFFER);
}
return false;
}
static bool
get_dimensions(const struct pipe_image_view *iview,
const struct softpipe_resource *spr,
unsigned tgsi_tex_instr,
enum pipe_format pformat,
unsigned *width,
unsigned *height,
unsigned *depth)
{
if (tgsi_tex_instr == TGSI_TEXTURE_BUFFER) {
*width = iview->u.buf.last_element - iview->u.buf.first_element + 1;
*height = 1;
*depth = 1;
/*
* Bounds check the buffer size from the view
* and the buffer size from the underlying buffer.
*/
if (util_format_get_stride(pformat, *width) >
util_format_get_stride(spr->base.format, spr->base.width0))
return false;
} else {
unsigned level;
level = spr->base.target == PIPE_BUFFER ? 0 : iview->u.tex.level;
*width = u_minify(spr->base.width0, level);
*height = u_minify(spr->base.height0, level);
if (spr->base.target == TGSI_TEXTURE_3D)
*depth = u_minify(spr->base.depth0, level);
else
*depth = spr->base.array_size;
/* Make sure the resource and view have compatiable formats */
if (util_format_get_blocksize(pformat) >
util_format_get_blocksize(spr->base.format))
return false;
}
return true;
}
static void
fill_coords(const struct tgsi_image_params *params,
unsigned index,
const int s[TGSI_QUAD_SIZE],
const int t[TGSI_QUAD_SIZE],
const int r[TGSI_QUAD_SIZE],
int *s_coord, int *t_coord, int *r_coord)
{
*s_coord = s[index];
*t_coord = has_1coord(params->tgsi_tex_instr) ? 0 : t[index];
*r_coord = has_layer_or_depth(params->tgsi_tex_instr) ?
(params->tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY ? t[index] : r[index]) : 0;
}
/*
* Implement the image LOAD operation.
*/
static void
sp_tgsi_load(const struct tgsi_image *image,
const struct tgsi_image_params *params,
const int s[TGSI_QUAD_SIZE],
const int t[TGSI_QUAD_SIZE],
const int r[TGSI_QUAD_SIZE],
const int sample[TGSI_QUAD_SIZE],
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image;
struct pipe_image_view *iview;
struct softpipe_resource *spr;
unsigned width, height, depth;
unsigned stride;
int c, j;
char *data_ptr;
unsigned offset = 0;
if (params->unit > PIPE_MAX_SHADER_IMAGES)
goto fail_write_all_zero;
iview = &sp_img->sp_iview[params->unit];
spr = (struct softpipe_resource *)iview->resource;
if (!spr)
goto fail_write_all_zero;
if (!has_compat_target(spr->base.target, params->tgsi_tex_instr))
goto fail_write_all_zero;
if (!get_dimensions(iview, spr, params->tgsi_tex_instr,
params->format, &width, &height, &depth))
return;
stride = util_format_get_stride(params->format, width);
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
int s_coord, t_coord, r_coord;
bool fill_zero = false;
if (!(params->execmask & (1 << j)))
fill_zero = true;
fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord);
if (!bounds_check(width, height, depth,
s_coord, t_coord, r_coord))
fill_zero = true;
if (fill_zero) {
int nc = util_format_get_nr_components(params->format);
int ival = util_format_is_pure_integer(params->format);
for (c = 0; c < 4; c++) {
rgba[c][j] = 0;
if (c == 3 && nc < 4) {
if (ival)
((int32_t *)rgba[c])[j] = 1;
else
rgba[c][j] = 1.0;
}
}
continue;
}
offset = get_image_offset(spr, iview, params->format, r_coord);
data_ptr = (char *)spr->data + offset;
if (util_format_is_pure_sint(params->format)) {
int32_t sdata[4];
util_format_read_4i(params->format,
sdata, 0,
data_ptr, stride,
s_coord, t_coord, 1, 1);
for (c = 0; c < 4; c++)
((int32_t *)rgba[c])[j] = sdata[c];
} else if (util_format_is_pure_uint(params->format)) {
uint32_t sdata[4];
util_format_read_4ui(params->format,
sdata, 0,
data_ptr, stride,
s_coord, t_coord, 1, 1);
for (c = 0; c < 4; c++)
((uint32_t *)rgba[c])[j] = sdata[c];
} else {
float sdata[4];
util_format_read_4f(params->format,
sdata, 0,
data_ptr, stride,
s_coord, t_coord, 1, 1);
for (c = 0; c < 4; c++)
rgba[c][j] = sdata[c];
}
}
return;
fail_write_all_zero:
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
for (c = 0; c < 4; c++)
rgba[c][j] = 0;
}
return;
}
/*
* Implement the image STORE operation.
*/
static void
sp_tgsi_store(const struct tgsi_image *image,
const struct tgsi_image_params *params,
const int s[TGSI_QUAD_SIZE],
const int t[TGSI_QUAD_SIZE],
const int r[TGSI_QUAD_SIZE],
const int sample[TGSI_QUAD_SIZE],
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image;
struct pipe_image_view *iview;
struct softpipe_resource *spr;
unsigned width, height, depth;
unsigned stride;
char *data_ptr;
int j, c;
unsigned offset = 0;
unsigned pformat = params->format;
if (params->unit > PIPE_MAX_SHADER_IMAGES)
return;
iview = &sp_img->sp_iview[params->unit];
spr = (struct softpipe_resource *)iview->resource;
if (!spr)
return;
if (!has_compat_target(spr->base.target, params->tgsi_tex_instr))
return;
if (params->format == PIPE_FORMAT_NONE)
pformat = spr->base.format;
if (!get_dimensions(iview, spr, params->tgsi_tex_instr,
pformat, &width, &height, &depth))
return;
stride = util_format_get_stride(pformat, width);
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
int s_coord, t_coord, r_coord;
if (!(params->execmask & (1 << j)))
continue;
fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord);
if (!bounds_check(width, height, depth,
s_coord, t_coord, r_coord))
continue;
offset = get_image_offset(spr, iview, pformat, r_coord);
data_ptr = (char *)spr->data + offset;
if (util_format_is_pure_sint(pformat)) {
int32_t sdata[4];
for (c = 0; c < 4; c++)
sdata[c] = ((int32_t *)rgba[c])[j];
util_format_write_4i(pformat, sdata, 0, data_ptr, stride,
s_coord, t_coord, 1, 1);
} else if (util_format_is_pure_uint(pformat)) {
uint32_t sdata[4];
for (c = 0; c < 4; c++)
sdata[c] = ((uint32_t *)rgba[c])[j];
util_format_write_4ui(pformat, sdata, 0, data_ptr, stride,
s_coord, t_coord, 1, 1);
} else {
float sdata[4];
for (c = 0; c < 4; c++)
sdata[c] = rgba[c][j];
util_format_write_4f(pformat, sdata, 0, data_ptr, stride,
s_coord, t_coord, 1, 1);
}
}
}
/*
* Implement atomic operations on unsigned integers.
*/
static void
handle_op_uint(const struct pipe_image_view *iview,
const struct tgsi_image_params *params,
bool just_read,
char *data_ptr,
uint qi,
unsigned stride,
unsigned opcode,
int s,
int t,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
uint c;
int nc = util_format_get_nr_components(params->format);
unsigned sdata[4];
util_format_read_4ui(params->format,
sdata, 0,
data_ptr, stride,
s, t, 1, 1);
if (just_read) {
for (c = 0; c < nc; c++) {
((uint32_t *)rgba[c])[qi] = sdata[c];
}
return;
}
switch (opcode) {
case TGSI_OPCODE_ATOMUADD:
for (c = 0; c < nc; c++) {
unsigned temp = sdata[c];
sdata[c] += ((uint32_t *)rgba[c])[qi];
((uint32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMXCHG:
for (c = 0; c < nc; c++) {
unsigned temp = sdata[c];
sdata[c] = ((uint32_t *)rgba[c])[qi];
((uint32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMCAS:
for (c = 0; c < nc; c++) {
unsigned dst_x = sdata[c];
unsigned cmp_x = ((uint32_t *)rgba[c])[qi];
unsigned src_x = ((uint32_t *)rgba2[c])[qi];
unsigned temp = sdata[c];
sdata[c] = (dst_x == cmp_x) ? src_x : dst_x;
((uint32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMAND:
for (c = 0; c < nc; c++) {
unsigned temp = sdata[c];
sdata[c] &= ((uint32_t *)rgba[c])[qi];
((uint32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMOR:
for (c = 0; c < nc; c++) {
unsigned temp = sdata[c];
sdata[c] |= ((uint32_t *)rgba[c])[qi];
((uint32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMXOR:
for (c = 0; c < nc; c++) {
unsigned temp = sdata[c];
sdata[c] ^= ((uint32_t *)rgba[c])[qi];
((uint32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMUMIN:
for (c = 0; c < nc; c++) {
unsigned dst_x = sdata[c];
unsigned src_x = ((uint32_t *)rgba[c])[qi];
sdata[c] = MIN2(dst_x, src_x);
((uint32_t *)rgba[c])[qi] = dst_x;
}
break;
case TGSI_OPCODE_ATOMUMAX:
for (c = 0; c < nc; c++) {
unsigned dst_x = sdata[c];
unsigned src_x = ((uint32_t *)rgba[c])[qi];
sdata[c] = MAX2(dst_x, src_x);
((uint32_t *)rgba[c])[qi] = dst_x;
}
break;
case TGSI_OPCODE_ATOMIMIN:
for (c = 0; c < nc; c++) {
int dst_x = sdata[c];
int src_x = ((uint32_t *)rgba[c])[qi];
sdata[c] = MIN2(dst_x, src_x);
((uint32_t *)rgba[c])[qi] = dst_x;
}
break;
case TGSI_OPCODE_ATOMIMAX:
for (c = 0; c < nc; c++) {
int dst_x = sdata[c];
int src_x = ((uint32_t *)rgba[c])[qi];
sdata[c] = MAX2(dst_x, src_x);
((uint32_t *)rgba[c])[qi] = dst_x;
}
break;
default:
assert(!"Unexpected TGSI opcode in sp_tgsi_op");
break;
}
util_format_write_4ui(params->format, sdata, 0, data_ptr, stride,
s, t, 1, 1);
}
/*
* Implement atomic operations on signed integers.
*/
static void
handle_op_int(const struct pipe_image_view *iview,
const struct tgsi_image_params *params,
bool just_read,
char *data_ptr,
uint qi,
unsigned stride,
unsigned opcode,
int s,
int t,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
uint c;
int nc = util_format_get_nr_components(params->format);
int sdata[4];
util_format_read_4i(params->format,
sdata, 0,
data_ptr, stride,
s, t, 1, 1);
if (just_read) {
for (c = 0; c < nc; c++) {
((int32_t *)rgba[c])[qi] = sdata[c];
}
return;
}
switch (opcode) {
case TGSI_OPCODE_ATOMUADD:
for (c = 0; c < nc; c++) {
int temp = sdata[c];
sdata[c] += ((int32_t *)rgba[c])[qi];
((int32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMXCHG:
for (c = 0; c < nc; c++) {
int temp = sdata[c];
sdata[c] = ((int32_t *)rgba[c])[qi];
((int32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMCAS:
for (c = 0; c < nc; c++) {
int dst_x = sdata[c];
int cmp_x = ((int32_t *)rgba[c])[qi];
int src_x = ((int32_t *)rgba2[c])[qi];
int temp = sdata[c];
sdata[c] = (dst_x == cmp_x) ? src_x : dst_x;
((int32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMAND:
for (c = 0; c < nc; c++) {
int temp = sdata[c];
sdata[c] &= ((int32_t *)rgba[c])[qi];
((int32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMOR:
for (c = 0; c < nc; c++) {
int temp = sdata[c];
sdata[c] |= ((int32_t *)rgba[c])[qi];
((int32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMXOR:
for (c = 0; c < nc; c++) {
int temp = sdata[c];
sdata[c] ^= ((int32_t *)rgba[c])[qi];
((int32_t *)rgba[c])[qi] = temp;
}
break;
case TGSI_OPCODE_ATOMUMIN:
for (c = 0; c < nc; c++) {
int dst_x = sdata[c];
int src_x = ((int32_t *)rgba[c])[qi];
sdata[c] = MIN2(dst_x, src_x);
((int32_t *)rgba[c])[qi] = dst_x;
}
break;
case TGSI_OPCODE_ATOMUMAX:
for (c = 0; c < nc; c++) {
int dst_x = sdata[c];
int src_x = ((int32_t *)rgba[c])[qi];
sdata[c] = MAX2(dst_x, src_x);
((int32_t *)rgba[c])[qi] = dst_x;
}
break;
case TGSI_OPCODE_ATOMIMIN:
for (c = 0; c < nc; c++) {
int dst_x = sdata[c];
int src_x = ((int32_t *)rgba[c])[qi];
sdata[c] = MIN2(dst_x, src_x);
((int32_t *)rgba[c])[qi] = dst_x;
}
break;
case TGSI_OPCODE_ATOMIMAX:
for (c = 0; c < nc; c++) {
int dst_x = sdata[c];
int src_x = ((int32_t *)rgba[c])[qi];
sdata[c] = MAX2(dst_x, src_x);
((int32_t *)rgba[c])[qi] = dst_x;
}
break;
default:
assert(!"Unexpected TGSI opcode in sp_tgsi_op");
break;
}
util_format_write_4i(params->format, sdata, 0, data_ptr, stride,
s, t, 1, 1);
}
/*
* Implement atomic image operations.
*/
static void
sp_tgsi_op(const struct tgsi_image *image,
const struct tgsi_image_params *params,
unsigned opcode,
const int s[TGSI_QUAD_SIZE],
const int t[TGSI_QUAD_SIZE],
const int r[TGSI_QUAD_SIZE],
const int sample[TGSI_QUAD_SIZE],
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image;
struct pipe_image_view *iview;
struct softpipe_resource *spr;
unsigned width, height, depth;
unsigned stride;
int j, c;
unsigned offset;
char *data_ptr;
if (params->unit > PIPE_MAX_SHADER_IMAGES)
return;
iview = &sp_img->sp_iview[params->unit];
spr = (struct softpipe_resource *)iview->resource;
if (!spr)
goto fail_write_all_zero;
if (!has_compat_target(spr->base.target, params->tgsi_tex_instr))
goto fail_write_all_zero;
if (!get_dimensions(iview, spr, params->tgsi_tex_instr,
params->format, &width, &height, &depth))
goto fail_write_all_zero;
stride = util_format_get_stride(spr->base.format, width);
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
int s_coord, t_coord, r_coord;
bool just_read = false;
fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord);
if (!bounds_check(width, height, depth,
s_coord, t_coord, r_coord)) {
int nc = util_format_get_nr_components(params->format);
int ival = util_format_is_pure_integer(params->format);
int c;
for (c = 0; c < 4; c++) {
rgba[c][j] = 0;
if (c == 3 && nc < 4) {
if (ival)
((int32_t *)rgba[c])[j] = 1;
else
rgba[c][j] = 1.0;
}
}
continue;
}
/* just readback the value for atomic if execmask isn't set */
if (!(params->execmask & (1 << j))) {
just_read = true;
}
offset = get_image_offset(spr, iview, params->format, r_coord);
data_ptr = (char *)spr->data + offset;
/* we should see atomic operations on r32 formats */
if (util_format_is_pure_uint(params->format))
handle_op_uint(iview, params, just_read, data_ptr, j, stride,
opcode, s_coord, t_coord, rgba, rgba2);
else if (util_format_is_pure_sint(params->format))
handle_op_int(iview, params, just_read, data_ptr, j, stride,
opcode, s_coord, t_coord, rgba, rgba2);
else
assert(0);
}
return;
fail_write_all_zero:
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
for (c = 0; c < 4; c++)
rgba[c][j] = 0;
}
return;
}
static void
sp_tgsi_get_dims(const struct tgsi_image *image,
const struct tgsi_image_params *params,
int dims[4])
{
struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image;
struct pipe_image_view *iview;
struct softpipe_resource *spr;
int level;
if (params->unit > PIPE_MAX_SHADER_IMAGES)
return;
iview = &sp_img->sp_iview[params->unit];
spr = (struct softpipe_resource *)iview->resource;
if (!spr)
return;
if (params->tgsi_tex_instr == TGSI_TEXTURE_BUFFER) {
dims[0] = iview->u.buf.last_element - iview->u.buf.first_element + 1;
dims[1] = dims[2] = dims[3] = 0;
return;
}
level = iview->u.tex.level;
dims[0] = u_minify(spr->base.width0, level);
switch (params->tgsi_tex_instr) {
case TGSI_TEXTURE_1D_ARRAY:
dims[1] = iview->u.tex.last_layer - iview->u.tex.first_layer + 1;
/* fallthrough */
case TGSI_TEXTURE_1D:
return;
case TGSI_TEXTURE_2D_ARRAY:
dims[2] = iview->u.tex.last_layer - iview->u.tex.first_layer + 1;
/* fallthrough */
case TGSI_TEXTURE_2D:
case TGSI_TEXTURE_CUBE:
case TGSI_TEXTURE_RECT:
dims[1] = u_minify(spr->base.height0, level);
return;
case TGSI_TEXTURE_3D:
dims[1] = u_minify(spr->base.height0, level);
dims[2] = u_minify(spr->base.depth0, level);
return;
case TGSI_TEXTURE_CUBE_ARRAY:
dims[1] = u_minify(spr->base.height0, level);
dims[2] = (iview->u.tex.last_layer - iview->u.tex.first_layer + 1) / 6;
break;
default:
assert(!"unexpected texture target in sp_get_dims()");
return;
}
}
struct sp_tgsi_image *
sp_create_tgsi_image(void)
{
struct sp_tgsi_image *img = CALLOC_STRUCT(sp_tgsi_image);
if (!img)
return NULL;
img->base.load = sp_tgsi_load;
img->base.store = sp_tgsi_store;
img->base.op = sp_tgsi_op;
img->base.get_dims = sp_tgsi_get_dims;
return img;
};

View File

@ -0,0 +1,37 @@
/*
* Copyright 2016 Red Hat.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* on the rights to use, copy, modify, merge, publish, distribute, sub
* license, and/or sell copies of the Software, and to permit persons to whom
* the Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
* THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#ifndef SP_IMAGE_H
#define SP_IMAGE_H
#include "tgsi/tgsi_exec.h"
struct sp_tgsi_image
{
struct tgsi_image base;
struct pipe_image_view sp_iview[PIPE_MAX_SHADER_IMAGES];
};
struct sp_tgsi_image *
sp_create_tgsi_image(void);
#endif

View File

@ -56,6 +56,7 @@
struct tgsi_sampler;
struct tgsi_image;
struct tgsi_exec_machine;
struct vertex_info;
@ -81,7 +82,8 @@ struct sp_fragment_shader_variant
void (*prepare)(const struct sp_fragment_shader_variant *shader,
struct tgsi_exec_machine *machine,
struct tgsi_sampler *sampler);
struct tgsi_sampler *sampler,
struct tgsi_image *image);
unsigned (*run)(const struct sp_fragment_shader_variant *shader,
struct tgsi_exec_machine *machine,
@ -149,6 +151,9 @@ softpipe_init_streamout_funcs(struct pipe_context *pipe);
void
softpipe_init_vertex_funcs(struct pipe_context *pipe);
void
softpipe_init_image_funcs(struct pipe_context *pipe);
void
softpipe_set_framebuffer_state(struct pipe_context *,
const struct pipe_framebuffer_state *);

View File

@ -343,7 +343,8 @@ update_fragment_shader(struct softpipe_context *softpipe, unsigned prim)
softpipe->fs_variant->prepare(softpipe->fs_variant,
softpipe->fs_machine,
(struct tgsi_sampler *) softpipe->
tgsi.sampler[PIPE_SHADER_FRAGMENT]);
tgsi.sampler[PIPE_SHADER_FRAGMENT],
(struct tgsi_image *)softpipe->tgsi.image[PIPE_SHADER_FRAGMENT]);
}
else {
softpipe->fs_variant = NULL;

View File

@ -0,0 +1,57 @@
/*
* Copyright 2016 Red Hat.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* on the rights to use, copy, modify, merge, publish, distribute, sub
* license, and/or sell copies of the Software, and to permit persons to whom
* the Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
* THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#include "sp_context.h"
#include "sp_state.h"
#include "sp_image.h"
static void softpipe_set_shader_images(struct pipe_context *pipe,
unsigned shader,
unsigned start,
unsigned num,
struct pipe_image_view *images)
{
struct softpipe_context *softpipe = softpipe_context(pipe);
unsigned i;
assert(shader < PIPE_SHADER_TYPES);
assert(start + num <= Elements(softpipe->sampler_views[shader]));
/* set the new images */
for (i = 0; i < num; i++) {
int idx = start + i;
if (images) {
pipe_resource_reference(&softpipe->tgsi.image[shader]->sp_iview[idx].resource, images[i].resource);
softpipe->tgsi.image[shader]->sp_iview[idx] = images[i];
}
else {
pipe_resource_reference(&softpipe->tgsi.image[shader]->sp_iview[idx].resource, NULL);
memset(&softpipe->tgsi.image[shader]->sp_iview[idx], 0, sizeof(struct pipe_image_view));
}
}
}
void softpipe_init_image_funcs(struct pipe_context *pipe)
{
pipe->set_shader_images = softpipe_set_shader_images;
}

View File

@ -270,9 +270,9 @@ softpipe_resource_get_handle(struct pipe_screen *screen,
* Helper function to compute offset (in bytes) for a particular
* texture level/face/slice from the start of the buffer.
*/
static unsigned
sp_get_tex_image_offset(const struct softpipe_resource *spr,
unsigned level, unsigned layer)
unsigned
softpipe_get_tex_image_offset(const struct softpipe_resource *spr,
unsigned level, unsigned layer)
{
unsigned offset = spr->level_offset[level];
@ -422,7 +422,7 @@ softpipe_transfer_map(struct pipe_context *pipe,
pt->stride = spr->stride[level];
pt->layer_stride = spr->img_stride[level];
spt->offset = sp_get_tex_image_offset(spr, level, box->z);
spt->offset = softpipe_get_tex_image_offset(spr, level, box->z);
spt->offset +=
box->y / util_format_get_blockheight(format) * spt->base.stride +

View File

@ -116,5 +116,7 @@ softpipe_init_screen_texture_funcs(struct pipe_screen *screen);
extern void
softpipe_init_texture_funcs(struct pipe_context *pipe);
unsigned
softpipe_get_tex_image_offset(const struct softpipe_resource *spr,
unsigned level, unsigned layer);
#endif /* SP_TEXTURE */