From c3ebac68900de5ad461a7b5a279621a435f5bcec Mon Sep 17 00:00:00 2001 From: =?utf8?q?Marek=20Ol=C5=A1=C3=A1k?= Date: Tue, 29 Aug 2017 03:58:22 +0200 Subject: [PATCH] radeonsi/gfx9: implement primitive binning MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit This increases performance, but it was tuned for Raven, not Vega. We don't know yet how Vega will perform, hopefully not worse. Reviewed-by: Nicolai Hähnle --- src/gallium/drivers/radeon/r600_pipe_common.c | 2 + src/gallium/drivers/radeon/r600_pipe_common.h | 2 + src/gallium/drivers/radeonsi/Makefile.sources | 1 + src/gallium/drivers/radeonsi/si_hw_context.c | 2 + src/gallium/drivers/radeonsi/si_pipe.c | 5 + src/gallium/drivers/radeonsi/si_pipe.h | 2 + src/gallium/drivers/radeonsi/si_state.c | 26 +- src/gallium/drivers/radeonsi/si_state.h | 6 +- src/gallium/drivers/radeonsi/si_state_binning.c | 448 ++++++++++++++++++++++++ src/gallium/drivers/radeonsi/si_state_shaders.c | 2 + 10 files changed, 489 insertions(+), 7 deletions(-) create mode 100644 src/gallium/drivers/radeonsi/si_state_binning.c diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c b/src/gallium/drivers/radeon/r600_pipe_common.c index 99172c6..1302e11 100644 --- a/src/gallium/drivers/radeon/r600_pipe_common.c +++ b/src/gallium/drivers/radeon/r600_pipe_common.c @@ -823,6 +823,8 @@ static const struct debug_named_value common_debug_options[] = { { "mono", DBG_MONOLITHIC_SHADERS, "Use old-style monolithic shaders compiled on demand" }, { "unsafemath", DBG_UNSAFE_MATH, "Enable unsafe math shader optimizations" }, { "nodccfb", DBG_NO_DCC_FB, "Disable separate DCC on the main framebuffer" }, + { "nodpbb", DBG_NO_DPBB, "Disable DPBB." }, + { "nodfsm", DBG_NO_DFSM, "Disable DFSM." }, DEBUG_NAMED_VALUE_END /* must be last */ }; diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h index 9f9b42b..42480ec 100644 --- a/src/gallium/drivers/radeon/r600_pipe_common.h +++ b/src/gallium/drivers/radeon/r600_pipe_common.h @@ -116,6 +116,8 @@ struct u_log_context; #define DBG_TEST_VMFAULT_CP (1ull << 51) #define DBG_TEST_VMFAULT_SDMA (1ull << 52) #define DBG_TEST_VMFAULT_SHADER (1ull << 53) +#define DBG_NO_DPBB (1ull << 54) +#define DBG_NO_DFSM (1ull << 55) #define R600_MAP_BUFFER_ALIGNMENT 64 #define R600_MAX_VIEWPORTS 16 diff --git a/src/gallium/drivers/radeonsi/Makefile.sources b/src/gallium/drivers/radeonsi/Makefile.sources index 74df4d7..3795bed 100644 --- a/src/gallium/drivers/radeonsi/Makefile.sources +++ b/src/gallium/drivers/radeonsi/Makefile.sources @@ -27,6 +27,7 @@ C_SOURCES := \ si_shader_tgsi_mem.c \ si_shader_tgsi_setup.c \ si_state.c \ + si_state_binning.c \ si_state_draw.c \ si_state_shaders.c \ si_state.h \ diff --git a/src/gallium/drivers/radeonsi/si_hw_context.c b/src/gallium/drivers/radeonsi/si_hw_context.c index f3af8dc..d090eea 100644 --- a/src/gallium/drivers/radeonsi/si_hw_context.c +++ b/src/gallium/drivers/radeonsi/si_hw_context.c @@ -239,6 +239,8 @@ void si_begin_new_cs(struct si_context *ctx) if (!has_clear_state || ctx->blend_color.any_nonzeros) si_mark_atom_dirty(ctx, &ctx->blend_color.atom); si_mark_atom_dirty(ctx, &ctx->db_render_state); + if (ctx->b.chip_class >= GFX9) + si_mark_atom_dirty(ctx, &ctx->dpbb_state); si_mark_atom_dirty(ctx, &ctx->stencil_ref.atom); si_mark_atom_dirty(ctx, &ctx->spi_map); si_mark_atom_dirty(ctx, &ctx->b.streamout.enable_atom); diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c index 2a3af6f..ca2e055 100644 --- a/src/gallium/drivers/radeonsi/si_pipe.c +++ b/src/gallium/drivers/radeonsi/si_pipe.c @@ -1049,6 +1049,11 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws, sscreen->b.family <= CHIP_POLARIS12) || sscreen->b.family == CHIP_VEGA10 || sscreen->b.family == CHIP_RAVEN; + sscreen->dpbb_allowed = sscreen->b.chip_class >= GFX9 && + !(sscreen->b.debug_flags & DBG_NO_DPBB); + sscreen->dfsm_allowed = sscreen->dpbb_allowed && + !(sscreen->b.debug_flags & DBG_NO_DFSM); + /* While it would be nice not to have this flag, we are constrained * by the reality that LLVM 5.0 doesn't have working VGPR indexing * on GFX9. diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h index 0228b32..fbf8db9 100644 --- a/src/gallium/drivers/radeonsi/si_pipe.h +++ b/src/gallium/drivers/radeonsi/si_pipe.h @@ -96,6 +96,7 @@ struct si_screen { bool has_draw_indirect_multi; bool has_ds_bpermute; bool has_msaa_sample_loc_bug; + bool dpbb_allowed; bool dfsm_allowed; bool llvm_has_working_vgpr_indexing; @@ -309,6 +310,7 @@ struct si_context { struct si_framebuffer framebuffer; struct si_sample_locs msaa_sample_locs; struct r600_atom db_render_state; + struct r600_atom dpbb_state; struct r600_atom msaa_config; struct si_sample_mask sample_mask; struct r600_atom cb_render_state; diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c index 076b583..5bbfbdd 100644 --- a/src/gallium/drivers/radeonsi/si_state.c +++ b/src/gallium/drivers/radeonsi/si_state.c @@ -623,6 +623,13 @@ static void si_bind_blend_state(struct pipe_context *ctx, void *state) old_blend->blend_enable_4bit != blend->blend_enable_4bit || old_blend->need_src_alpha_4bit != blend->need_src_alpha_4bit) sctx->do_update_shaders = true; + + if (sctx->screen->dpbb_allowed && + (!old_blend || + old_blend->alpha_to_coverage != blend->alpha_to_coverage || + old_blend->blend_enable_4bit != blend->blend_enable_4bit || + old_blend->cb_target_enabled_4bit != blend->cb_target_enabled_4bit)) + si_mark_atom_dirty(sctx, &sctx->dpbb_state); } static void si_delete_blend_state(struct pipe_context *ctx, void *state) @@ -1140,6 +1147,13 @@ static void si_bind_dsa_state(struct pipe_context *ctx, void *state) if (!old_dsa || old_dsa->alpha_func != dsa->alpha_func) sctx->do_update_shaders = true; + + if (sctx->screen->dpbb_allowed && + (!old_dsa || + (old_dsa->depth_enabled != dsa->depth_enabled || + old_dsa->stencil_enabled != dsa->stencil_enabled || + old_dsa->db_can_write != dsa->db_can_write))) + si_mark_atom_dirty(sctx, &sctx->dpbb_state); } static void si_delete_dsa_state(struct pipe_context *ctx, void *state) @@ -2691,6 +2705,9 @@ static void si_set_framebuffer_state(struct pipe_context *ctx, si_mark_atom_dirty(sctx, &sctx->cb_render_state); si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); + if (sctx->screen->dpbb_allowed) + si_mark_atom_dirty(sctx, &sctx->dpbb_state); + if (sctx->framebuffer.any_dst_linear != old_any_dst_linear) si_mark_atom_dirty(sctx, &sctx->msaa_config); @@ -3072,6 +3089,8 @@ static void si_set_min_samples(struct pipe_context *ctx, unsigned min_samples) if (sctx->framebuffer.nr_samples > 1) si_mark_atom_dirty(sctx, &sctx->msaa_config); + if (sctx->screen->dpbb_allowed) + si_mark_atom_dirty(sctx, &sctx->dpbb_state); } /* @@ -4131,6 +4150,7 @@ void si_init_state_functions(struct si_context *sctx) si_init_atom(sctx, &sctx->framebuffer.atom, &sctx->atoms.s.framebuffer, si_emit_framebuffer_state); si_init_atom(sctx, &sctx->msaa_sample_locs.atom, &sctx->atoms.s.msaa_sample_locs, si_emit_msaa_sample_locs); si_init_atom(sctx, &sctx->db_render_state, &sctx->atoms.s.db_render_state, si_emit_db_render_state); + si_init_atom(sctx, &sctx->dpbb_state, &sctx->atoms.s.dpbb_state, si_emit_dpbb_state); si_init_atom(sctx, &sctx->msaa_config, &sctx->atoms.s.msaa_config, si_emit_msaa_config); si_init_atom(sctx, &sctx->sample_mask.atom, &sctx->atoms.s.sample_mask, si_emit_sample_mask); si_init_atom(sctx, &sctx->cb_render_state, &sctx->atoms.s.cb_render_state, si_emit_cb_render_state); @@ -4743,12 +4763,6 @@ static void si_init_config(struct si_context *sctx) assert(0); } - si_pm4_set_reg(pm4, R_028060_DB_DFSM_CONTROL, - S_028060_PUNCHOUT_MODE(V_028060_FORCE_OFF)); - /* TODO: Enable the binner: */ - si_pm4_set_reg(pm4, R_028C44_PA_SC_BINNER_CNTL_0, - S_028C44_BINNING_MODE(V_028C44_DISABLE_BINNING_USE_LEGACY_SC) | - S_028C44_DISABLE_START_OF_PRIM(1)); si_pm4_set_reg(pm4, R_028C48_PA_SC_BINNER_CNTL_1, S_028C48_MAX_ALLOC_COUNT(MIN2(128, pc_lines / (4 * num_se))) | S_028C48_MAX_PRIM_PER_BATCH(1023)); diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h index 9770960..17d210a 100644 --- a/src/gallium/drivers/radeonsi/si_state.h +++ b/src/gallium/drivers/radeonsi/si_state.h @@ -153,6 +153,7 @@ union si_state_atoms { struct r600_atom *framebuffer; struct r600_atom *msaa_sample_locs; struct r600_atom *db_render_state; + struct r600_atom *dpbb_state; struct r600_atom *msaa_config; struct r600_atom *sample_mask; struct r600_atom *cb_render_state; @@ -373,7 +374,10 @@ si_create_sampler_view_custom(struct pipe_context *ctx, unsigned force_level); void si_update_fb_dirtiness_after_rendering(struct si_context *sctx); -/* si_state_shader.c */ +/* si_state_binning.c */ +void si_emit_dpbb_state(struct si_context *sctx, struct r600_atom *state); + +/* si_state_shaders.c */ bool si_update_shaders(struct si_context *sctx); void si_init_shader_functions(struct si_context *sctx); bool si_init_shader_cache(struct si_screen *sscreen); diff --git a/src/gallium/drivers/radeonsi/si_state_binning.c b/src/gallium/drivers/radeonsi/si_state_binning.c new file mode 100644 index 0000000..d75e86e --- /dev/null +++ b/src/gallium/drivers/radeonsi/si_state_binning.c @@ -0,0 +1,448 @@ +/* + * Copyright 2017 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. + */ + +/* This file handles register programming of primitive binning. */ + +#include "si_pipe.h" +#include "sid.h" +#include "gfx9d.h" +#include "radeon/r600_cs.h" + +struct uvec2 { + unsigned x, y; +}; + +struct si_bin_size_map { + unsigned start; + unsigned bin_size_x; + unsigned bin_size_y; +}; + +typedef struct si_bin_size_map si_bin_size_subtable[3][9]; + +/* Find the bin size where sum is >= table[i].start and < table[i + 1].start. */ +static struct uvec2 si_find_bin_size(struct si_screen *sscreen, + const si_bin_size_subtable table[], + unsigned sum) +{ + unsigned log_num_rb_per_se = + util_logbase2_ceil(sscreen->b.info.num_render_backends / + sscreen->b.info.max_se); + unsigned log_num_se = util_logbase2_ceil(sscreen->b.info.max_se); + unsigned i; + + /* Get the chip-specific subtable. */ + const struct si_bin_size_map *subtable = + &table[log_num_rb_per_se][log_num_se][0]; + + for (i = 0; subtable[i].start != UINT_MAX; i++) { + if (sum >= subtable[i].start && sum < subtable[i + 1].start) + break; + } + + struct uvec2 size = {subtable[i].bin_size_x, subtable[i].bin_size_y}; + return size; +} + +static struct uvec2 si_get_color_bin_size(struct si_context *sctx, + unsigned cb_target_enabled_4bit) +{ + unsigned nr_samples = sctx->framebuffer.nr_samples; + unsigned sum = 0; + + /* Compute the sum of all Bpp. */ + for (unsigned i = 0; i < sctx->framebuffer.state.nr_cbufs; i++) { + if (!(cb_target_enabled_4bit & (0xf << (i * 4)))) + continue; + + struct r600_texture *rtex = + (struct r600_texture*)sctx->framebuffer.state.cbufs[i]->texture; + sum += rtex->surface.bpe; + } + + /* Multiply the sum by some function of the number of samples. */ + if (nr_samples >= 2) { + if (sctx->ps_iter_samples >= 2) + sum *= nr_samples; + else + sum *= 2; + } + + static const si_bin_size_subtable table[] = { + { + /* One RB / SE */ + { + /* One shader engine */ + { 0, 128, 128 }, + { 1, 64, 128 }, + { 2, 32, 128 }, + { 3, 16, 128 }, + { 17, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + /* Two shader engines */ + { 0, 128, 128 }, + { 2, 64, 128 }, + { 3, 32, 128 }, + { 5, 16, 128 }, + { 17, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + /* Four shader engines */ + { 0, 128, 128 }, + { 3, 64, 128 }, + { 5, 16, 128 }, + { 17, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + }, + { + /* Two RB / SE */ + { + /* One shader engine */ + { 0, 128, 128 }, + { 2, 64, 128 }, + { 3, 32, 128 }, + { 5, 16, 128 }, + { 33, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + /* Two shader engines */ + { 0, 128, 128 }, + { 3, 64, 128 }, + { 5, 32, 128 }, + { 9, 16, 128 }, + { 33, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + /* Four shader engines */ + { 0, 256, 256 }, + { 2, 128, 256 }, + { 3, 128, 128 }, + { 5, 64, 128 }, + { 9, 16, 128 }, + { 33, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + }, + { + /* Four RB / SE */ + { + /* One shader engine */ + { 0, 128, 256 }, + { 2, 128, 128 }, + { 3, 64, 128 }, + { 5, 32, 128 }, + { 9, 16, 128 }, + { 33, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + /* Two shader engines */ + { 0, 256, 256 }, + { 2, 128, 256 }, + { 3, 128, 128 }, + { 5, 64, 128 }, + { 9, 32, 128 }, + { 17, 16, 128 }, + { 33, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + /* Four shader engines */ + { 0, 256, 512 }, + { 2, 256, 256 }, + { 3, 128, 256 }, + { 5, 128, 128 }, + { 9, 64, 128 }, + { 17, 16, 128 }, + { 33, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + }, + }; + + return si_find_bin_size(sctx->screen, table, sum); +} + +static struct uvec2 si_get_depth_bin_size(struct si_context *sctx) +{ + struct si_state_dsa *dsa = sctx->queued.named.dsa; + + if (!sctx->framebuffer.state.zsbuf || + (!dsa->depth_enabled && !dsa->stencil_enabled)) { + /* Return the max size. */ + struct uvec2 size = {512, 512}; + return size; + } + + struct r600_texture *rtex = + (struct r600_texture*)sctx->framebuffer.state.zsbuf->texture; + unsigned depth_coeff = dsa->depth_enabled ? 5 : 0; + unsigned stencil_coeff = rtex->surface.flags & RADEON_SURF_SBUFFER && + dsa->stencil_enabled ? 1 : 0; + unsigned sum = 4 * (depth_coeff + stencil_coeff) * + sctx->framebuffer.nr_samples; + + static const si_bin_size_subtable table[] = { + { + // One RB / SE + { + // One shader engine + { 0, 128, 256 }, + { 2, 128, 128 }, + { 4, 64, 128 }, + { 7, 32, 128 }, + { 13, 16, 128 }, + { 49, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + // Two shader engines + { 0, 256, 256 }, + { 2, 128, 256 }, + { 4, 128, 128 }, + { 7, 64, 128 }, + { 13, 32, 128 }, + { 25, 16, 128 }, + { 49, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + // Four shader engines + { 0, 256, 512 }, + { 2, 256, 256 }, + { 4, 128, 256 }, + { 7, 128, 128 }, + { 13, 64, 128 }, + { 25, 16, 128 }, + { 49, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + }, + { + // Two RB / SE + { + // One shader engine + { 0, 256, 256 }, + { 2, 128, 256 }, + { 4, 128, 128 }, + { 7, 64, 128 }, + { 13, 32, 128 }, + { 25, 16, 128 }, + { 97, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + // Two shader engines + { 0, 256, 512 }, + { 2, 256, 256 }, + { 4, 128, 256 }, + { 7, 128, 128 }, + { 13, 64, 128 }, + { 25, 32, 128 }, + { 49, 16, 128 }, + { 97, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + { + // Four shader engines + { 0, 512, 512 }, + { 2, 256, 512 }, + { 4, 256, 256 }, + { 7, 128, 256 }, + { 13, 128, 128 }, + { 25, 64, 128 }, + { 49, 16, 128 }, + { 97, 0, 0 }, + { UINT_MAX, 0, 0 }, + }, + }, + { + // Four RB / SE + { + // One shader engine + { 0, 256, 512 }, + { 2, 256, 256 }, + { 4, 128, 256 }, + { 7, 128, 128 }, + { 13, 64, 128 }, + { 25, 32, 128 }, + { 49, 16, 128 }, + { UINT_MAX, 0, 0 }, + }, + { + // Two shader engines + { 0, 512, 512 }, + { 2, 256, 512 }, + { 4, 256, 256 }, + { 7, 128, 256 }, + { 13, 128, 128 }, + { 25, 64, 128 }, + { 49, 32, 128 }, + { 97, 16, 128 }, + { UINT_MAX, 0, 0 }, + }, + { + // Four shader engines + { 0, 512, 512 }, + { 4, 256, 512 }, + { 7, 256, 256 }, + { 13, 128, 256 }, + { 25, 128, 128 }, + { 49, 64, 128 }, + { 97, 16, 128 }, + { UINT_MAX, 0, 0 }, + }, + }, + }; + + return si_find_bin_size(sctx->screen, table, sum); +} + +static void si_emit_dpbb_disable(struct si_context *sctx) +{ + struct radeon_winsys_cs *cs = sctx->b.gfx.cs; + + radeon_set_context_reg(cs, R_028C44_PA_SC_BINNER_CNTL_0, + S_028C44_BINNING_MODE(V_028C44_DISABLE_BINNING_USE_LEGACY_SC) | + S_028C44_DISABLE_START_OF_PRIM(1)); + radeon_set_context_reg(cs, R_028060_DB_DFSM_CONTROL, + S_028060_PUNCHOUT_MODE(V_028060_FORCE_OFF)); +} + +void si_emit_dpbb_state(struct si_context *sctx, struct r600_atom *state) +{ + struct si_screen *sscreen = sctx->screen; + struct si_state_blend *blend = sctx->queued.named.blend; + struct si_state_dsa *dsa = sctx->queued.named.dsa; + unsigned db_shader_control = sctx->ps_db_shader_control; + + assert(sctx->b.chip_class >= GFX9); + + if (!sscreen->dpbb_allowed || !blend || !dsa) { + si_emit_dpbb_disable(sctx); + return; + } + + bool ps_can_kill = G_02880C_KILL_ENABLE(db_shader_control) || + G_02880C_MASK_EXPORT_ENABLE(db_shader_control) || + G_02880C_COVERAGE_TO_MASK_ENABLE(db_shader_control) || + blend->alpha_to_coverage; + + /* This is ported from Vulkan, but it doesn't make much sense to me. + * Maybe it's for RE-Z? But Vulkan doesn't use RE-Z. TODO: Clarify this. + */ + bool ps_can_reject_z_trivially = + !G_02880C_Z_EXPORT_ENABLE(db_shader_control) || + G_02880C_CONSERVATIVE_Z_EXPORT(db_shader_control); + + /* Disable binning if PS can kill trivially with DB writes. + * Ported from Vulkan. (heuristic?) + */ + if (ps_can_kill && + ps_can_reject_z_trivially && + sctx->framebuffer.state.zsbuf && + dsa->db_can_write) { + si_emit_dpbb_disable(sctx); + return; + } + + /* Compute the bin size. */ + /* TODO: We could also look at enabled pixel shader outputs. */ + unsigned cb_target_enabled_4bit = sctx->framebuffer.colorbuf_enabled_4bit & + blend->cb_target_enabled_4bit; + struct uvec2 color_bin_size = + si_get_color_bin_size(sctx, cb_target_enabled_4bit); + struct uvec2 depth_bin_size = si_get_depth_bin_size(sctx); + + unsigned color_area = color_bin_size.x * color_bin_size.y; + unsigned depth_area = depth_bin_size.x * depth_bin_size.y; + + struct uvec2 bin_size = color_area < depth_area ? color_bin_size + : depth_bin_size; + + if (!bin_size.x || !bin_size.y) { + si_emit_dpbb_disable(sctx); + return; + } + + /* Enable DFSM if it's preferred. */ + unsigned punchout_mode = V_028060_FORCE_OFF; + bool disable_start_of_prim = true; + + if (sscreen->dfsm_allowed && + cb_target_enabled_4bit && + !G_02880C_KILL_ENABLE(db_shader_control) && + /* These two also imply that DFSM is disabled when PS writes to memory. */ + !G_02880C_EXEC_ON_HIER_FAIL(db_shader_control) && + !G_02880C_EXEC_ON_NOOP(db_shader_control) && + G_02880C_Z_ORDER(db_shader_control) == V_02880C_EARLY_Z_THEN_LATE_Z) { + punchout_mode = V_028060_AUTO; + disable_start_of_prim = (cb_target_enabled_4bit & + blend->blend_enable_4bit) != 0; + } + + /* Tunable parameters. Also test with DFSM enabled/disabled. */ + unsigned context_states_per_bin; /* allowed range: [0, 5] */ + unsigned persistent_states_per_bin; /* allowed range: [0, 31] */ + unsigned fpovs_per_batch; /* allowed range: [0, 255], 0 = unlimited */ + + switch (sctx->b.family) { + case CHIP_VEGA10: + case CHIP_RAVEN: + /* Tuned for Raven. Vega might need different values. */ + context_states_per_bin = 5; + persistent_states_per_bin = 31; + fpovs_per_batch = 63; + break; + default: + assert(0); + } + + /* Emit registers. */ + struct uvec2 bin_size_extend = {}; + if (bin_size.x >= 32) + bin_size_extend.x = util_logbase2(bin_size.x) - 5; + if (bin_size.y >= 32) + bin_size_extend.y = util_logbase2(bin_size.y) - 5; + + struct radeon_winsys_cs *cs = sctx->b.gfx.cs; + radeon_set_context_reg(cs, R_028C44_PA_SC_BINNER_CNTL_0, + S_028C44_BINNING_MODE(V_028C44_BINNING_ALLOWED) | + S_028C44_BIN_SIZE_X(bin_size.x == 16) | + S_028C44_BIN_SIZE_Y(bin_size.y == 16) | + S_028C44_BIN_SIZE_X_EXTEND(bin_size_extend.x) | + S_028C44_BIN_SIZE_Y_EXTEND(bin_size_extend.y) | + S_028C44_CONTEXT_STATES_PER_BIN(context_states_per_bin) | + S_028C44_PERSISTENT_STATES_PER_BIN(persistent_states_per_bin) | + S_028C44_DISABLE_START_OF_PRIM(disable_start_of_prim) | + S_028C44_FPOVS_PER_BATCH(fpovs_per_batch) | + S_028C44_OPTIMAL_BIN_SELECTION(1)); + radeon_set_context_reg(cs, R_028060_DB_DFSM_CONTROL, + S_028060_PUNCHOUT_MODE(punchout_mode)); +} diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c index 061b3d2..4f04bbd 100644 --- a/src/gallium/drivers/radeonsi/si_state_shaders.c +++ b/src/gallium/drivers/radeonsi/si_state_shaders.c @@ -3285,6 +3285,8 @@ bool si_update_shaders(struct si_context *sctx) if (sctx->ps_db_shader_control != db_shader_control) { sctx->ps_db_shader_control = db_shader_control; si_mark_atom_dirty(sctx, &sctx->db_render_state); + if (sctx->screen->dpbb_allowed) + si_mark_atom_dirty(sctx, &sctx->dpbb_state); } if (sctx->smoothing_enabled != sctx->ps_shader.current->key.part.ps.epilog.poly_line_smoothing) { -- 2.7.4