anv: factor out host/gpu internal shaders interfaces

This will prevent host/gpu structure definitions to go out of sync.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Tested-by: Felix DeGrood <felix.j.degrood@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25361>
This commit is contained in:
Lionel Landwerlin 2023-09-17 13:39:46 +03:00 committed by Marge Bot
parent da1e39faa4
commit db335d9b73
7 changed files with 143 additions and 132 deletions

View file

@ -24,40 +24,7 @@
#ifndef ANV_GENERATED_INDIRECT_DRAWS_H
#define ANV_GENERATED_INDIRECT_DRAWS_H
#include <stdint.h>
#define ANV_GENERATED_FLAG_INDEXED BITFIELD_BIT(0)
#define ANV_GENERATED_FLAG_PREDICATED BITFIELD_BIT(1)
#define ANV_GENERATED_FLAG_DRAWID BITFIELD_BIT(2)
#define ANV_GENERATED_FLAG_BASE BITFIELD_BIT(3)
/* This needs to match common_generated_draws.glsl :
*
* layout(set = 0, binding = 4) uniform block
*/
struct anv_generated_indirect_draw_params {
/* Draw ID buffer address (only used on Gfx9) */
uint64_t draw_id_addr;
/* Indirect data buffer address (only used on Gfx9) */
uint64_t indirect_data_addr;
/* Stride between each elements of the indirect data buffer */
uint32_t indirect_data_stride;
uint32_t flags; /* 0-7: bits, 8-15: mocs, 16-23: cmd_dws */
/* Base number of the draw ID, it is added to the index computed from the
* gl_FragCoord
*/
uint32_t draw_base;
/* Maximum number of draws (equals to draw_count for indirect draws without
* an indirect count)
*/
uint32_t max_draw_count;
/* Instance multiplier for multi view */
uint32_t instance_multiplier;
/* Address where to jump at after the generated draw (only used with
* indirect draw count variants)
*/
uint64_t end_addr;
};
#include "shaders/interface.h"
struct anv_generated_indirect_params {
struct anv_generated_indirect_draw_params draw;
@ -84,43 +51,6 @@ struct anv_generated_indirect_params {
struct anv_generated_indirect_params *prev;
};
#define ANV_COPY_QUERY_FLAG_RESULT64 BITFIELD_BIT(0)
#define ANV_COPY_QUERY_FLAG_AVAILABLE BITFIELD_BIT(1)
#define ANV_COPY_QUERY_FLAG_DELTA BITFIELD_BIT(2)
#define ANV_COPY_QUERY_FLAG_PARTIAL BITFIELD_BIT(3)
/* This needs to match common_query_copy.glsl :
*
* layout(set = 0, binding = 2) uniform block
*/
struct anv_query_copy_shader_params {
/* ANV_COPY_QUERY_FLAG_* flags */
uint32_t flags;
/* Number of queries to copy */
uint32_t num_queries;
/* Number of items to write back in the results per query */
uint32_t num_items;
/* First query to copy result from */
uint32_t query_base;
/* Query stride in bytes */
uint32_t query_stride;
/* Offset at which the data should be read from */
uint32_t query_data_offset;
/* Stride of destination writes */
uint32_t destination_stride;
/* We need to be 64 bit aligned, or 32 bit builds get
* very unhappy.
*/
uint32_t padding;
};
struct anv_query_copy_params {
struct anv_query_copy_shader_params copy;

View file

@ -21,12 +21,7 @@
* IN THE SOFTWARE.
*/
#define BITFIELD_BIT(i) (1u << i)
#define ANV_GENERATED_FLAG_INDEXED BITFIELD_BIT(0)
#define ANV_GENERATED_FLAG_PREDICATED BITFIELD_BIT(1)
#define ANV_GENERATED_FLAG_DRAWID BITFIELD_BIT(2)
#define ANV_GENERATED_FLAG_BASE BITFIELD_BIT(3)
#include "interface.h"
/* These 3 bindings will be accessed through A64 messages */
layout(set = 0, binding = 0, std430) buffer Storage0 {
@ -56,14 +51,7 @@ layout(set = 0, binding = 3) buffer Storage3 {
/* This data will be provided through push constants. */
layout(set = 0, binding = 4) uniform block {
uint64_t draw_id_addr;
uint64_t indirect_data_addr;
uint indirect_data_stride;
uint flags;
uint draw_base;
uint max_draw_count;
uint instance_multiplier;
uint64_t end_addr;
anv_generated_indirect_draw_params params;
};
void write_VERTEX_BUFFER_STATE(uint write_offset,
@ -146,15 +134,15 @@ void write_MI_BATCH_BUFFER_START(uint write_offset,
void end_generated_draws(uint cmd_idx, uint draw_id, uint draw_count)
{
uint _3dprim_dw_size = (flags >> 16) & 0xff;
uint _3dprim_dw_size = (params.flags >> 16) & 0xff;
/* We can have an indirect draw count = 0. */
uint last_draw_id = draw_count == 0 ? 0 : (min(draw_count, max_draw_count) - 1);
uint last_draw_id = draw_count == 0 ? 0 : (min(draw_count, params.max_draw_count) - 1);
uint jump_offset = draw_count == 0 ? 0 : _3dprim_dw_size;
if (draw_id == last_draw_id && draw_count < max_draw_count) {
if (draw_id == last_draw_id && draw_count < params.max_draw_count) {
/* Only write a jump forward in the batch if we have fewer elements than
* the max draw count.
*/
write_MI_BATCH_BUFFER_START(cmd_idx + jump_offset, end_addr);
write_MI_BATCH_BUFFER_START(cmd_idx + jump_offset, params.end_addr);
}
}

View file

@ -21,12 +21,7 @@
* IN THE SOFTWARE.
*/
#define BITFIELD_BIT(i) (1u << i)
#define ANV_COPY_QUERY_FLAG_RESULT64 BITFIELD_BIT(0)
#define ANV_COPY_QUERY_FLAG_AVAILABLE BITFIELD_BIT(1)
#define ANV_COPY_QUERY_FLAG_DELTA BITFIELD_BIT(2)
#define ANV_COPY_QUERY_FLAG_PARTIAL BITFIELD_BIT(3)
#include "interface.h"
/* These 3 bindings will be accessed through A64 messages */
layout(set = 0, binding = 0, std430) buffer Storage0 {
@ -39,34 +34,28 @@ layout(set = 0, binding = 1, std430) buffer Storage1 {
/* This data will be provided through push constants. */
layout(set = 0, binding = 2) uniform block {
uint flags;
uint num_queries;
uint num_items;
uint query_base;
uint query_stride;
uint query_data_offset;
uint destination_stride;
anv_query_copy_shader_params params;
};
void query_copy(uint item_idx)
{
if (item_idx >= num_queries)
if (item_idx >= params.num_queries)
return;
bool is_result64 = (flags & ANV_COPY_QUERY_FLAG_RESULT64) != 0;
bool write_available = (flags & ANV_COPY_QUERY_FLAG_AVAILABLE) != 0;
bool compute_delta = (flags & ANV_COPY_QUERY_FLAG_DELTA) != 0;
bool partial_result = (flags & ANV_COPY_QUERY_FLAG_PARTIAL) != 0;
bool is_result64 = (params.flags & ANV_COPY_QUERY_FLAG_RESULT64) != 0;
bool write_available = (params.flags & ANV_COPY_QUERY_FLAG_AVAILABLE) != 0;
bool compute_delta = (params.flags & ANV_COPY_QUERY_FLAG_DELTA) != 0;
bool partial_result = (params.flags & ANV_COPY_QUERY_FLAG_PARTIAL) != 0;
uint query_byte = (query_base + item_idx) * query_stride;
uint query_data_byte = query_byte + query_data_offset;
uint destination_byte = item_idx * destination_stride;
uint query_byte = (params.query_base + item_idx) * params.query_stride;
uint query_data_byte = query_byte + params.query_data_offset;
uint destination_byte = item_idx * params.destination_stride;
uint64_t availability = query_data[query_byte / 4];
uint query_data_dword = query_data_byte / 4;
uint dest_dword = destination_byte / 4;
for (uint i = 0; i < num_items; i++) {
for (uint i = 0; i < params.num_items; i++) {
uint item_data_dword = query_data_dword + i * 2 * (compute_delta ? 2 : 1);
uint64_t v;

View file

@ -29,14 +29,14 @@
void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
{
bool is_indexed = (flags & ANV_GENERATED_FLAG_INDEXED) != 0;
bool is_predicated = (flags & ANV_GENERATED_FLAG_PREDICATED) != 0;
uint indirect_data_offset = item_idx * indirect_data_stride / 4;
bool is_indexed = (params.flags & ANV_GENERATED_FLAG_INDEXED) != 0;
bool is_predicated = (params.flags & ANV_GENERATED_FLAG_PREDICATED) != 0;
uint indirect_data_offset = item_idx * params.indirect_data_stride / 4;
if (is_indexed) {
/* Loading a VkDrawIndexedIndirectCommand */
uint index_count = indirect_data[indirect_data_offset + 0];
uint instance_count = indirect_data[indirect_data_offset + 1] * instance_multiplier;
uint instance_count = indirect_data[indirect_data_offset + 1] * params.instance_multiplier;
uint first_index = indirect_data[indirect_data_offset + 2];
uint vertex_offset = indirect_data[indirect_data_offset + 3];
uint first_instance = indirect_data[indirect_data_offset + 4];
@ -55,7 +55,7 @@ void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
} else {
/* Loading a VkDrawIndirectCommand structure */
uint vertex_count = indirect_data[indirect_data_offset + 0];
uint instance_count = indirect_data[indirect_data_offset + 1] * instance_multiplier;
uint instance_count = indirect_data[indirect_data_offset + 1] * params.instance_multiplier;
uint first_vertex = indirect_data[indirect_data_offset + 2];
uint first_instance = indirect_data[indirect_data_offset + 3];
@ -76,10 +76,10 @@ void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
void main()
{
uint _3dprim_dw_size = (flags >> 16) & 0xff;
uint _3dprim_dw_size = (params.flags >> 16) & 0xff;
uint item_idx = uint(gl_FragCoord.y) * 8192 + uint(gl_FragCoord.x);
uint cmd_idx = item_idx * _3dprim_dw_size;
uint draw_id = draw_base + item_idx;
uint draw_id = params.draw_base + item_idx;
uint draw_count = _draw_count;
if (draw_id < draw_count)

View file

@ -29,17 +29,17 @@
void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
{
bool is_indexed = (flags & ANV_GENERATED_FLAG_INDEXED) != 0;
bool is_predicated = (flags & ANV_GENERATED_FLAG_PREDICATED) != 0;
bool uses_base = (flags & ANV_GENERATED_FLAG_BASE) != 0;
bool uses_drawid = (flags & ANV_GENERATED_FLAG_DRAWID) != 0;
uint mocs = (flags >> 8) & 0xff;
uint indirect_data_offset = item_idx * indirect_data_stride / 4;
bool is_indexed = (params.flags & ANV_GENERATED_FLAG_INDEXED) != 0;
bool is_predicated = (params.flags & ANV_GENERATED_FLAG_PREDICATED) != 0;
bool uses_base = (params.flags & ANV_GENERATED_FLAG_BASE) != 0;
bool uses_drawid = (params.flags & ANV_GENERATED_FLAG_DRAWID) != 0;
uint mocs = (params.flags >> 8) & 0xff;
uint indirect_data_offset = item_idx * params.indirect_data_stride / 4;
if (is_indexed) {
/* Loading a VkDrawIndexedIndirectCommand */
uint index_count = indirect_data[indirect_data_offset + 0];
uint instance_count = indirect_data[indirect_data_offset + 1] * instance_multiplier;
uint instance_count = indirect_data[indirect_data_offset + 1] * params.instance_multiplier;
uint first_index = indirect_data[indirect_data_offset + 2];
uint vertex_offset = indirect_data[indirect_data_offset + 3];
uint first_instance = indirect_data[indirect_data_offset + 4];
@ -56,7 +56,7 @@ void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
cmd_idx += 1;
if (uses_base) {
uint64_t indirect_draw_data_addr =
indirect_data_addr + item_idx * indirect_data_stride + 12;
params.indirect_data_addr + item_idx * params.indirect_data_stride + 12;
write_VERTEX_BUFFER_STATE(cmd_idx,
mocs,
31,
@ -65,7 +65,7 @@ void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
cmd_idx += 4;
}
if (uses_drawid) {
uint64_t draw_idx_addr = draw_id_addr + 4 * item_idx;
uint64_t draw_idx_addr = params.draw_id_addr + 4 * item_idx;
draw_ids[item_idx] = draw_id;
write_VERTEX_BUFFER_STATE(cmd_idx,
mocs,
@ -86,7 +86,7 @@ void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
} else {
/* Loading a VkDrawIndirectCommand structure */
uint vertex_count = indirect_data[indirect_data_offset + 0];
uint instance_count = indirect_data[indirect_data_offset + 1] * instance_multiplier;
uint instance_count = indirect_data[indirect_data_offset + 1] * params.instance_multiplier;
uint first_vertex = indirect_data[indirect_data_offset + 2];
uint first_instance = indirect_data[indirect_data_offset + 3];
@ -102,7 +102,7 @@ void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
cmd_idx += 1;
if (uses_base) {
uint64_t indirect_draw_data_addr =
indirect_data_addr + item_idx * indirect_data_stride + 8;
params.indirect_data_addr + item_idx * params.indirect_data_stride + 8;
write_VERTEX_BUFFER_STATE(cmd_idx,
mocs,
31,
@ -111,7 +111,7 @@ void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
cmd_idx += 4;
}
if (uses_drawid) {
uint64_t draw_idx_addr = draw_id_addr + 4 * item_idx;
uint64_t draw_idx_addr = params.draw_id_addr + 4 * item_idx;
draw_ids[item_idx] = draw_id;
write_VERTEX_BUFFER_STATE(cmd_idx,
mocs,
@ -134,10 +134,10 @@ void write_draw(uint item_idx, uint cmd_idx, uint draw_id)
void main()
{
uint _3dprim_dw_size = (flags >> 16) & 0xff;
uint _3dprim_dw_size = (params.flags >> 16) & 0xff;
uint item_idx = uint(gl_FragCoord.y) * 8192 + uint(gl_FragCoord.x);
uint cmd_idx = item_idx * _3dprim_dw_size;
uint draw_id = draw_base + item_idx;
uint draw_id = params.draw_base + item_idx;
uint draw_count = _draw_count;
if (draw_id < draw_count)

View file

@ -0,0 +1,103 @@
/*
* Copyright © 2023 Intel Corporation
*
* 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
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* 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 NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS 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 _SHADERS_INTERFACE_H_
#define _SHADERS_INTERFACE_H_
#ifdef _IN_SHADER_
#define BITFIELD_BIT(i) (1u << i)
#define uint32_t uint
#else
#include "util/macros.h"
#include <stdint.h>
#endif
#define ANV_GENERATED_FLAG_INDEXED BITFIELD_BIT(0)
#define ANV_GENERATED_FLAG_PREDICATED BITFIELD_BIT(1)
#define ANV_GENERATED_FLAG_DRAWID BITFIELD_BIT(2)
#define ANV_GENERATED_FLAG_BASE BITFIELD_BIT(3)
struct anv_generated_indirect_draw_params {
/* Draw ID buffer address (only used on Gfx9) */
uint64_t draw_id_addr;
/* Indirect data buffer address (only used on Gfx9) */
uint64_t indirect_data_addr;
/* Stride between each elements of the indirect data buffer */
uint32_t indirect_data_stride;
uint32_t flags; /* 0-7: bits, 8-15: mocs, 16-23: cmd_dws */
/* Base number of the draw ID, it is added to the index computed from the
* gl_FragCoord
*/
uint32_t draw_base;
/* Maximum number of draws (equals to draw_count for indirect draws without
* an indirect count)
*/
uint32_t max_draw_count;
/* Instance multiplier for multi view */
uint32_t instance_multiplier;
/* Address where to jump at after the generated draw (only used with
* indirect draw count variants)
*/
uint64_t end_addr;
};
#define ANV_COPY_QUERY_FLAG_RESULT64 BITFIELD_BIT(0)
#define ANV_COPY_QUERY_FLAG_AVAILABLE BITFIELD_BIT(1)
#define ANV_COPY_QUERY_FLAG_DELTA BITFIELD_BIT(2)
#define ANV_COPY_QUERY_FLAG_PARTIAL BITFIELD_BIT(3)
/* This needs to match common_query_copy.glsl :
*
* layout(set = 0, binding = 2) uniform block
*/
struct anv_query_copy_shader_params {
/* ANV_COPY_QUERY_FLAG_* flags */
uint32_t flags;
/* Number of queries to copy */
uint32_t num_queries;
/* Number of items to write back in the results per query */
uint32_t num_items;
/* First query to copy result from */
uint32_t query_base;
/* Query stride in bytes */
uint32_t query_stride;
/* Offset at which the data should be read from */
uint32_t query_data_offset;
/* Stride of destination writes */
uint32_t destination_stride;
/* We need to be 64 bit aligned, or 32 bit builds get
* very unhappy.
*/
uint32_t padding;
};
#endif /* _SHADERS_INTERFACE_H_ */

View file

@ -58,5 +58,6 @@ foreach item : anv_internal_shaders
'--glsl-version', '450',
'--stage', stage,
'-I' + meson.current_source_dir(),
'-D_IN_SHADER_=1',
])
endforeach