Commit bf7302a6 authored by Christian König's avatar Christian König
Browse files

radeonsi: rework state handling v2



Add a complete new state handling for SI.

v2: fix spelling error
Signed-off-by: default avatarChristian König <deathsimple@vodafone.de>
parent 27382c0f
......@@ -10,4 +10,6 @@ C_SOURCES := \
evergreen_hw_context.c \
evergreen_state.c \
r600_translate.c \
r600_state_common.c
r600_state_common.c \
radeonsi_pm4.c \
si_state.c
......@@ -24,6 +24,7 @@
* Jerome Glisse
*/
#include "r600_hw_context_priv.h"
#include "radeonsi_pm4.h"
#include "radeonsi_pipe.h"
#include "sid.h"
#include "util/u_memory.h"
......@@ -563,6 +564,7 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags)
ctx->pm4_dirty_cdwords += enable_block->pm4_ndwords;
enable_block->nreg_dirty = enable_block->nreg;
}
si_pm4_reset_emitted(ctx);
}
void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence_bo, unsigned offset, unsigned value)
......
......@@ -801,6 +801,8 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
rdraw.db_render_control = dsa->db_render_control;
/* Emit states. */
rctx->pm4_dirty_cdwords += si_pm4_dirty_dw(rctx);
r600_need_cs_space(rctx, 0, TRUE);
LIST_FOR_EACH_ENTRY_SAFE(state, next_state, &rctx->dirty_states, head) {
......@@ -809,6 +811,7 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &rctx->dirty,list) {
r600_context_block_emit_dirty(rctx, dirty_block);
}
si_pm4_emit_dirty(rctx);
rctx->pm4_dirty_cdwords = 0;
/* Enable stream out if needed. */
......
......@@ -36,6 +36,8 @@
#include "util/u_slab.h"
#include "r600.h"
#include "radeonsi_public.h"
#include "radeonsi_pm4.h"
#include "si_state.h"
#include "r600_resource.h"
#include "sid.h"
......@@ -320,6 +322,10 @@ struct r600_context {
struct pipe_index_buffer index_buffer;
struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
unsigned nr_vertex_buffers;
/* SI state handling */
union si_state queued;
union si_state emitted;
};
static INLINE void r600_emit_atom(struct r600_context *rctx, struct r600_atom *atom)
......
/*
* Copyright 2012 Advanced Micro Devices, Inc.
*
* 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.
*
* Authors:
* Christian König <christian.koenig@amd.com>
*/
#include "util/u_memory.h"
#include "radeonsi_pipe.h"
#include "radeonsi_pm4.h"
#include "sid.h"
#include "r600_hw_context_priv.h"
#define NUMBER_OF_STATES (sizeof(union si_state) / sizeof(struct si_pm4_state *))
void si_pm4_set_reg(struct si_pm4_state *state, unsigned reg, uint32_t val)
{
unsigned opcode, count;
if (reg >= SI_CONFIG_REG_OFFSET && reg <= SI_CONFIG_REG_END) {
opcode = PKT3_SET_CONFIG_REG;
reg -= SI_CONFIG_REG_OFFSET;
} else if (reg >= SI_SH_REG_OFFSET && reg <= SI_SH_REG_END) {
opcode = PKT3_SET_SH_REG;
reg -= SI_SH_REG_OFFSET;
} else if (reg >= SI_CONTEXT_REG_OFFSET && reg <= SI_CONTEXT_REG_END) {
opcode = PKT3_SET_CONTEXT_REG;
reg -= SI_CONTEXT_REG_OFFSET;
} else {
R600_ERR("Invalid register offset %08x!\n", reg);
return;
}
reg >>= 2;
if (opcode != state->last_opcode || reg != (state->last_reg + 1)) {
state->last_opcode = opcode;
state->last_pm4 = state->ndw++;
state->pm4[state->ndw++] = reg;
}
state->last_reg = reg;
count = state->ndw - state->last_pm4 - 1;
state->pm4[state->last_pm4] = PKT3(opcode, count, 0);
state->pm4[state->ndw++] = val;
assert(state->ndw <= SI_PM4_MAX_DW);
}
void si_pm4_add_bo(struct si_pm4_state *state,
struct r600_resource *bo,
enum radeon_bo_usage usage)
{
unsigned idx = state->nbo++;
assert(idx < SI_PM4_MAX_BO);
pipe_resource_reference((struct pipe_resource**)&state->bo[idx],
(struct pipe_resource*)bo);
state->bo_usage[idx] = usage;
}
void si_pm4_inval_shader_cache(struct si_pm4_state *state)
{
state->cp_coher_cntl |= S_0085F0_SH_ICACHE_ACTION_ENA(1);
state->cp_coher_cntl |= S_0085F0_SH_KCACHE_ACTION_ENA(1);
}
void si_pm4_inval_texture_cache(struct si_pm4_state *state)
{
state->cp_coher_cntl |= S_0085F0_TC_ACTION_ENA(1);
}
void si_pm4_inval_vertex_cache(struct si_pm4_state *state)
{
/* Some GPUs don't have the vertex cache and must use the texture cache instead. */
state->cp_coher_cntl |= S_0085F0_TC_ACTION_ENA(1);
}
void si_pm4_inval_fb_cache(struct si_pm4_state *state, unsigned nr_cbufs)
{
state->cp_coher_cntl |= S_0085F0_CB_ACTION_ENA(1);
state->cp_coher_cntl |= ((1 << nr_cbufs) - 1) << S_0085F0_CB0_DEST_BASE_ENA_SHIFT;
}
void si_pm4_inval_zsbuf_cache(struct si_pm4_state *state)
{
state->cp_coher_cntl |= S_0085F0_DB_ACTION_ENA(1) | S_0085F0_DB_DEST_BASE_ENA(1);
}
void si_pm4_free_state(struct r600_context *rctx,
struct si_pm4_state *state,
unsigned idx)
{
if (state == NULL)
return;
if (rctx->emitted.array[idx] == state) {
rctx->emitted.array[idx] = NULL;
}
for (int i = 0; i < state->nbo; ++i) {
pipe_resource_reference((struct pipe_resource**)&state->bo[idx],
NULL);
}
FREE(state);
}
unsigned si_pm4_dirty_dw(struct r600_context *rctx)
{
unsigned count = 0;
uint32_t cp_coher_cntl = 0;
for (int i = 0; i < NUMBER_OF_STATES; ++i) {
struct si_pm4_state *state = rctx->queued.array[i];
if (!state || rctx->emitted.array[i] == state)
continue;
count += state->ndw;
cp_coher_cntl |= state->cp_coher_cntl;
}
//TODO
rctx->atom_surface_sync.flush_flags |= cp_coher_cntl;
r600_atom_dirty(rctx, &rctx->atom_surface_sync.atom);
return count;
}
void si_pm4_emit_dirty(struct r600_context *rctx)
{
struct radeon_winsys_cs *cs = rctx->cs;
for (int i = 0; i < NUMBER_OF_STATES; ++i) {
struct si_pm4_state *state = rctx->queued.array[i];
if (!state || rctx->emitted.array[i] == state)
continue;
for (int j = 0; j < state->nbo; ++j) {
r600_context_bo_reloc(rctx, state->bo[j],
state->bo_usage[j]);
}
memcpy(&cs->buf[cs->cdw], state->pm4, state->ndw * 4);
cs->cdw += state->ndw;
rctx->emitted.array[i] = state;
}
}
void si_pm4_reset_emitted(struct r600_context *rctx)
{
memset(&rctx->emitted, 0, sizeof(rctx->emitted));
}
/*
* Copyright 2012 Advanced Micro Devices, Inc.
*
* 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.
*
* Authors:
* Christian König <christian.koenig@amd.com>
*/
#ifndef RADEONSI_PM4_H
#define RADEONSI_PM4_H
#include "../../winsys/radeon/drm/radeon_winsys.h"
#define SI_PM4_MAX_DW 128
#define SI_PM4_MAX_BO 32
// forward defines
struct r600_context;
struct si_pm4_state
{
/* PKT3_SET_*_REG handling */
unsigned last_opcode;
unsigned last_reg;
unsigned last_pm4;
/* flush flags for SURFACE_SYNC */
uint32_t cp_coher_cntl;
/* commands for the DE */
unsigned ndw;
uint32_t pm4[SI_PM4_MAX_DW];
/* BO's referenced by this state */
unsigned nbo;
struct r600_resource *bo[SI_PM4_MAX_BO];
enum radeon_bo_usage bo_usage[SI_PM4_MAX_BO];
};
void si_pm4_set_reg(struct si_pm4_state *state, unsigned reg, uint32_t val);
void si_pm4_add_bo(struct si_pm4_state *state,
struct r600_resource *bo,
enum radeon_bo_usage usage);
void si_pm4_inval_shader_cache(struct si_pm4_state *state);
void si_pm4_inval_texture_cache(struct si_pm4_state *state);
void si_pm4_inval_vertex_cache(struct si_pm4_state *state);
void si_pm4_inval_fb_cache(struct si_pm4_state *state, unsigned nr_cbufs);
void si_pm4_inval_zsbuf_cache(struct si_pm4_state *state);
void si_pm4_free_state(struct r600_context *rctx,
struct si_pm4_state *state,
unsigned idx);
unsigned si_pm4_dirty_dw(struct r600_context *rctx);
void si_pm4_emit_dirty(struct r600_context *rctx);
void si_pm4_reset_emitted(struct r600_context *rctx);
#endif
/*
* Copyright 2012 Advanced Micro Devices, Inc.
*
* 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.
*
* Authors:
* Christian König <christian.koenig@amd.com>
*
*/
#include "si_state.h"
/*
* Copyright 2012 Advanced Micro Devices, Inc.
*
* 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.
*
* Authors:
* Christian König <christian.koenig@amd.com>
*/
#ifndef SI_STATE_H
#define SI_STATE_H
#include "radeonsi_pm4.h"
union si_state {
struct {
} named;
struct si_pm4_state *array[0];
};
#define si_pm4_block_idx(member) \
(offsetof(union si_state, named.member) / sizeof(struct si_pm4_state *))
#define si_pm4_bind_state(rctx, member, value) \
do { \
(rctx)->queued.named.member = (value); \
} while(0);
#define si_pm4_delete_state(rctx, member, value) \
do { \
if ((rctx)->queued.named.member == (value)) { \
(rctx)->queued.named.member = NULL; \
} \
si_pm4_free_state(rctx, (struct si_pm4_state *)(value), \
si_pm4_block_idx(member)); \
} while(0);
#define si_pm4_set_state(rctx, member, value) \
do { \
if ((rctx)->queued.named.member != (value)) { \
si_pm4_free_state(rctx, \
(struct si_pm4_state *)(rctx)->queued.named.member, \
si_pm4_block_idx(member)); \
(rctx)->queued.named.member = (value); \
} \
} while(0);
#endif
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment