#include "util/u_memory.h"
#include "util/u_resource.h"
#include "util/u_upload_mgr.h"
+#include "util/u_blend.h"
#include "gfx10_format_table.h"
}
}
-static bool si_blend_factor_uses_dst(unsigned factor)
-{
- return factor == PIPE_BLENDFACTOR_DST_COLOR || factor == PIPE_BLENDFACTOR_DST_ALPHA ||
- factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
- factor == PIPE_BLENDFACTOR_INV_DST_ALPHA || factor == PIPE_BLENDFACTOR_INV_DST_COLOR;
-}
-
static void *si_create_blend_state_mode(struct pipe_context *ctx,
const struct pipe_blend_state *state, unsigned mode)
{
dstA_opt = si_translate_blend_opt_factor(dstA, true);
/* Handle interdependencies. */
- if (si_blend_factor_uses_dst(srcRGB))
+ if (util_blend_factor_uses_dest(srcRGB, false))
dstRGB_opt = V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE;
- if (si_blend_factor_uses_dst(srcA))
+ if (util_blend_factor_uses_dest(srcA, false))
dstA_opt = V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE;
if (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE &&
struct si_shader_selector *vs_sel = vs->selector;
struct si_shader_info *info = &vs_sel->info;
struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
- unsigned window_space = info->properties[TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION];
+ bool window_space = info->stage == MESA_SHADER_VERTEX ?
+ info->base.vs.window_space_position : 0;
unsigned clipdist_mask = vs_sel->clipdist_mask;
unsigned ucp_mask = clipdist_mask ? 0 : rs->clip_plane_enable & SIX_BITS;
unsigned culldist_mask = vs_sel->culldist_mask;
unsigned total_mask;
if (vs->key.opt.clip_disable) {
- assert(!info->culldist_writemask);
+ assert(!info->base.cull_distance_array_size);
clipdist_mask = 0;
culldist_mask = 0;
}
unsigned initial_cdw = sctx->gfx_cs->current.cdw;
unsigned pa_cl_cntl = S_02881C_VS_OUT_CCDIST0_VEC_ENA((total_mask & 0x0F) != 0) |
S_02881C_VS_OUT_CCDIST1_VEC_ENA((total_mask & 0xF0) != 0) |
- S_02881C_BYPASS_PRIM_RATE_COMBINER_GFX103(sctx->chip_class >= GFX10_3) |
+ S_02881C_BYPASS_VTX_RATE_COMBINER(sctx->chip_class >= GFX10_3) |
+ S_02881C_BYPASS_PRIM_RATE_COMBINER(sctx->chip_class >= GFX10_3) |
clipdist_mask | (culldist_mask << 8);
if (sctx->chip_class >= GFX10) {
S_028010_DISABLE_ZMASK_EXPCLEAR_OPTIMIZATION(sctx->db_depth_disable_expclear) |
S_028010_DISABLE_SMEM_EXPCLEAR_OPTIMIZATION(sctx->db_stencil_disable_expclear) |
S_028010_DECOMPRESS_Z_ON_FLUSH(sctx->framebuffer.nr_samples >= 4) |
- S_028010_CENTROID_COMPUTATION_MODE_GFX103(sctx->chip_class >= GFX10_3 ? 2 : 0));
+ S_028010_CENTROID_COMPUTATION_MODE(sctx->chip_class >= GFX10_3 ? 2 : 0));
db_shader_control = sctx->ps_db_shader_control;
if (desc->layout == UTIL_FORMAT_LAYOUT_ETC &&
(sscreen->info.family == CHIP_STONEY || sscreen->info.family == CHIP_VEGA10 ||
- sscreen->info.family == CHIP_RAVEN)) {
+ sscreen->info.family == CHIP_RAVEN || sscreen->info.family == CHIP_RAVEN2)) {
switch (format) {
case PIPE_FORMAT_ETC1_RGB8:
case PIPE_FORMAT_ETC2_RGB8:
case 4:
switch (desc->nr_channels) {
#if 0 /* Not supported for render targets */
- case 2:
- return V_008F14_IMG_DATA_FORMAT_4_4;
+ case 2:
+ return V_008F14_IMG_DATA_FORMAT_4_4;
#endif
case 4:
return V_008F14_IMG_DATA_FORMAT_4_4_4_4;
case 2:
return V_008F14_IMG_DATA_FORMAT_32_32;
#if 0 /* Not supported for render targets */
- case 3:
- return V_008F14_IMG_DATA_FORMAT_32_32_32;
+ case 3:
+ return V_008F14_IMG_DATA_FORMAT_32_32_32;
#endif
case 4:
return V_008F14_IMG_DATA_FORMAT_32_32_32_32;
meta_write_policy = V_02807C_CACHE_LRU_WR; /* cache writes */
meta_read_policy = V_02807C_CACHE_LRU_RD; /* cache reads */
} else {
- meta_write_policy = V_02807C_CACHE_STREAM_WR; /* write combine */
- meta_read_policy = V_02807C_CACHE_NOA_RD; /* don't cache reads */
+ meta_write_policy = V_02807C_CACHE_STREAM; /* write combine */
+ meta_read_policy = V_02807C_CACHE_NOA; /* don't cache reads */
}
/* Colorbuffers. */
radeon_emit(cs, zb->db_stencil_base >> 32); /* DB_STENCIL_WRITE_BASE_HI */
radeon_emit(cs, zb->db_htile_data_base >> 32); /* DB_HTILE_DATA_BASE_HI */
radeon_emit(cs, /* DB_RMI_L2_CACHE_CONTROL */
- S_02807C_Z_WR_POLICY(V_02807C_CACHE_STREAM_WR) |
- S_02807C_S_WR_POLICY(V_02807C_CACHE_STREAM_WR) |
+ S_02807C_Z_WR_POLICY(V_02807C_CACHE_STREAM) |
+ S_02807C_S_WR_POLICY(V_02807C_CACHE_STREAM) |
S_02807C_HTILE_WR_POLICY(meta_write_policy) |
- S_02807C_ZPCPSD_WR_POLICY(V_02807C_CACHE_STREAM_WR) |
- S_02807C_Z_RD_POLICY(V_02807C_CACHE_NOA_RD) |
- S_02807C_S_RD_POLICY(V_02807C_CACHE_NOA_RD) |
+ S_02807C_ZPCPSD_WR_POLICY(V_02807C_CACHE_STREAM) |
+ S_02807C_Z_RD_POLICY(V_02807C_CACHE_NOA) |
+ S_02807C_S_RD_POLICY(V_02807C_CACHE_NOA) |
S_02807C_HTILE_RD_POLICY(meta_read_policy) |
S_02807C_Z_BIG_PAGE(zs_big_page) |
S_02807C_S_BIG_PAGE(zs_big_page));
}
/* Framebuffer dimensions. */
- /* PA_SC_WINDOW_SCISSOR_TL is set in si_init_config() */
+ /* PA_SC_WINDOW_SCISSOR_TL is set in si_init_cs_preamble_state */
radeon_set_context_reg(cs, R_028208_PA_SC_WINDOW_SCISSOR_BR,
S_028208_BR_X(state->width) | S_028208_BR_Y(state->height));
S_028410_CMASK_WR_POLICY(meta_write_policy) |
S_028410_FMASK_WR_POLICY(meta_write_policy) |
S_028410_DCC_WR_POLICY(meta_write_policy) |
- S_028410_COLOR_WR_POLICY(V_028410_CACHE_STREAM_WR) |
+ S_028410_COLOR_WR_POLICY(V_028410_CACHE_STREAM) |
S_028410_CMASK_RD_POLICY(meta_read_policy) |
S_028410_FMASK_RD_POLICY(meta_read_policy) |
S_028410_DCC_RD_POLICY(meta_read_policy) |
- S_028410_COLOR_RD_POLICY(V_028410_CACHE_NOA_RD) |
+ S_028410_COLOR_RD_POLICY(V_028410_CACHE_NOA) |
S_028410_FMASK_BIG_PAGE(color_big_page) |
S_028410_COLOR_BIG_PAGE(color_big_page));
}
/* The set of PS invocations is always order invariant,
* except when early Z/S tests are requested. */
- if (sctx->ps_shader.cso && sctx->ps_shader.cso->info.writes_memory &&
- sctx->ps_shader.cso->info.properties[TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL] &&
+ if (sctx->ps_shader.cso && sctx->ps_shader.cso->info.base.writes_memory &&
+ sctx->ps_shader.cso->info.base.fs.early_fragment_tests &&
!dsa_order_invariant.pass_set)
return false;
sc_aa_config = S_028BE0_MSAA_NUM_SAMPLES(log_samples) |
S_028BE0_MAX_SAMPLE_DIST(max_dist[log_samples]) |
S_028BE0_MSAA_EXPOSED_SAMPLES(log_samples) |
- S_028BE0_COVERED_CENTROID_IS_CENTER_GFX103(sctx->chip_class >= GFX10_3);
+ S_028BE0_COVERED_CENTROID_IS_CENTER(sctx->chip_class >= GFX10_3);
if (sctx->framebuffer.nr_samples > 1) {
db_eqaa |= S_028804_MAX_ANCHOR_SAMPLES(log_z_samples) |
S_00A00C_BASE_LEVEL(res->nr_samples > 1 ? 0 : first_level) |
S_00A00C_LAST_LEVEL(res->nr_samples > 1 ? util_logbase2(res->nr_samples) : last_level) |
S_00A00C_BC_SWIZZLE(gfx9_border_color_swizzle(desc->swizzle)) | S_00A00C_TYPE(type);
-
- if (res->target == PIPE_TEXTURE_1D ||
- res->target == PIPE_TEXTURE_2D) {
- /* 1D, 2D, and 2D_MSAA can set a custom pitch for shader resources
- * starting with gfx10.3 (ignored if pitch <= width). Other texture
- * targets can't. CB and DB can't set a custom pitch for any target.
- */
- if (screen->info.chip_class >= GFX10_3)
- state[4] = S_00A010_DEPTH(tex->surface.u.gfx9.surf_pitch - 1);
- else
- state[4] = 0;
- } else {
- /* Depth is the last accessible layer on gfx9+. The hw doesn't need
- * to know the total number of layers.
- */
- state[4] = S_00A010_DEPTH((type == V_008F1C_SQ_RSRC_IMG_3D && sampler) ?
- depth - 1 : last_layer) |
- S_00A010_BASE_ARRAY(first_layer);
- }
-
+ /* Depth is the the last accessible layer on gfx9+. The hw doesn't need
+ * to know the total number of layers.
+ */
+ state[4] =
+ S_00A010_DEPTH((type == V_008F1C_SQ_RSRC_IMG_3D && sampler) ? depth - 1 : last_layer) |
+ S_00A010_BASE_ARRAY(first_layer);
state[5] = S_00A014_ARRAY_PITCH(!!(type == V_008F1C_SQ_RSRC_IMG_3D && !sampler)) |
S_00A014_MAX_MIP(res->nr_samples > 1 ? util_logbase2(res->nr_samples)
: tex->buffer.b.b.last_level) |
data_format = V_008F14_IMG_DATA_FORMAT_FMASK;
switch (FMASK(res->nr_samples, res->nr_storage_samples)) {
case FMASK(2, 1):
- num_format = V_008F14_IMG_FMASK_8_2_1;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_8_2_1;
break;
case FMASK(2, 2):
- num_format = V_008F14_IMG_FMASK_8_2_2;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_8_2_2;
break;
case FMASK(4, 1):
- num_format = V_008F14_IMG_FMASK_8_4_1;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_8_4_1;
break;
case FMASK(4, 2):
- num_format = V_008F14_IMG_FMASK_8_4_2;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_8_4_2;
break;
case FMASK(4, 4):
- num_format = V_008F14_IMG_FMASK_8_4_4;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_8_4_4;
break;
case FMASK(8, 1):
- num_format = V_008F14_IMG_FMASK_8_8_1;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_8_8_1;
break;
case FMASK(8, 2):
- num_format = V_008F14_IMG_FMASK_16_8_2;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_16_8_2;
break;
case FMASK(8, 4):
- num_format = V_008F14_IMG_FMASK_32_8_4;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_32_8_4;
break;
case FMASK(8, 8):
- num_format = V_008F14_IMG_FMASK_32_8_8;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_32_8_8;
break;
case FMASK(16, 1):
- num_format = V_008F14_IMG_FMASK_16_16_1;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_16_16_1;
break;
case FMASK(16, 2):
- num_format = V_008F14_IMG_FMASK_32_16_2;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_32_16_2;
break;
case FMASK(16, 4):
- num_format = V_008F14_IMG_FMASK_64_16_4;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_64_16_4;
break;
case FMASK(16, 8):
- num_format = V_008F14_IMG_FMASK_64_16_8;
+ num_format = V_008F14_IMG_NUM_FORMAT_FMASK_64_16_8;
break;
default:
unreachable("invalid nr_samples");
} else {
rstate->val[2] |= S_008F38_DISABLE_LSB_CEIL(sctx->chip_class <= GFX8) |
S_008F38_FILTER_PREC_FIX(1) |
- S_008F38_ANISO_OVERRIDE_GFX6(sctx->chip_class >= GFX8);
+ S_008F38_ANISO_OVERRIDE_GFX8(sctx->chip_class >= GFX8);
}
/* Create sampler resource for integer textures. */
unsigned mask = sctx->sample_mask;
/* Needed for line and polygon smoothing as well as for the Polaris
- * small primitive filter. We expect the gallium frontend to take care of
+ * small primitive filter. We expect the gallium frontend to take care of
* this for us.
*/
assert(mask == 0xffff || sctx->framebuffer.nr_samples > 1 ||
return si_create_blend_state_mode(&sctx->b, &blend, mode);
}
-static void si_init_config(struct si_context *sctx);
-
void si_init_state_compute_functions(struct si_context *sctx)
{
sctx->b.create_sampler_state = si_create_sampler_state;
sctx->b.set_tess_state = si_set_tess_state;
sctx->b.set_active_query_state = si_set_active_query_state;
-
- si_init_config(sctx);
}
void si_init_screen_state_functions(struct si_screen *sscreen)
}
}
-static void si_init_config(struct si_context *sctx)
+void si_init_cs_preamble_state(struct si_context *sctx, bool uses_reg_shadowing)
{
struct si_screen *sscreen = sctx->screen;
uint64_t border_color_va = sctx->border_color_buffer->gpu_address;
if (!pm4)
return;
- si_pm4_cmd_begin(pm4, PKT3_CONTEXT_CONTROL);
- si_pm4_cmd_add(pm4, CC0_UPDATE_LOAD_ENABLES(1));
- si_pm4_cmd_add(pm4, CC1_UPDATE_SHADOW_ENABLES(1));
- si_pm4_cmd_end(pm4, false);
-
- if (has_clear_state) {
- si_pm4_cmd_begin(pm4, PKT3_CLEAR_STATE);
- si_pm4_cmd_add(pm4, 0);
- si_pm4_cmd_end(pm4, false);
- }
-
- if (sctx->chip_class <= GFX8)
- si_set_raster_config(sctx, pm4);
-
- si_pm4_set_reg(pm4, R_028A18_VGT_HOS_MAX_TESS_LEVEL, fui(64));
- if (!has_clear_state)
- si_pm4_set_reg(pm4, R_028A1C_VGT_HOS_MIN_TESS_LEVEL, fui(0));
-
- /* FIXME calculate these values somehow ??? */
- if (sctx->chip_class <= GFX8) {
- si_pm4_set_reg(pm4, R_028A54_VGT_GS_PER_ES, SI_GS_PER_ES);
- si_pm4_set_reg(pm4, R_028A58_VGT_ES_PER_GS, 0x40);
- }
+ if (!uses_reg_shadowing) {
+ si_pm4_cmd_add(pm4, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
+ si_pm4_cmd_add(pm4, CC0_UPDATE_LOAD_ENABLES(1));
+ si_pm4_cmd_add(pm4, CC1_UPDATE_SHADOW_ENABLES(1));
- if (!has_clear_state) {
- si_pm4_set_reg(pm4, R_028A5C_VGT_GS_PER_VS, 0x2);
- si_pm4_set_reg(pm4, R_028A8C_VGT_PRIMITIVEID_RESET, 0x0);
- si_pm4_set_reg(pm4, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0x0);
+ if (has_clear_state) {
+ si_pm4_cmd_add(pm4, PKT3(PKT3_CLEAR_STATE, 0, 0));
+ si_pm4_cmd_add(pm4, 0);
+ }
}
- if (sscreen->info.chip_class <= GFX9)
- si_pm4_set_reg(pm4, R_028AA0_VGT_INSTANCE_STEP_RATE_0, 1);
- if (!has_clear_state)
- si_pm4_set_reg(pm4, R_028AB8_VGT_VTX_CNT_EN, 0x0);
- if (sctx->chip_class < GFX7)
- si_pm4_set_reg(pm4, R_008A14_PA_CL_ENHANCE,
- S_008A14_NUM_CLIP_SEQ(3) | S_008A14_CLIP_VTX_REORDER_ENA(1));
-
/* CLEAR_STATE doesn't restore these correctly. */
si_pm4_set_reg(pm4, R_028240_PA_SC_GENERIC_SCISSOR_TL, S_028240_WINDOW_OFFSET_DISABLE(1));
si_pm4_set_reg(pm4, R_028244_PA_SC_GENERIC_SCISSOR_BR,
S_028244_BR_X(16384) | S_028244_BR_Y(16384));
- /* CLEAR_STATE doesn't clear these correctly on certain generations.
- * I don't know why. Deduced by trial and error.
- */
- if (sctx->chip_class <= GFX7 || !has_clear_state) {
- si_pm4_set_reg(pm4, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
- si_pm4_set_reg(pm4, R_028204_PA_SC_WINDOW_SCISSOR_TL, S_028204_WINDOW_OFFSET_DISABLE(1));
- si_pm4_set_reg(pm4, R_028030_PA_SC_SCREEN_SCISSOR_TL, 0);
- si_pm4_set_reg(pm4, R_028034_PA_SC_SCREEN_SCISSOR_BR,
- S_028034_BR_X(16384) | S_028034_BR_Y(16384));
- }
+ si_pm4_set_reg(pm4, R_028A18_VGT_HOS_MAX_TESS_LEVEL, fui(64));
+ if (!has_clear_state)
+ si_pm4_set_reg(pm4, R_028A1C_VGT_HOS_MIN_TESS_LEVEL, fui(0));
if (!has_clear_state) {
si_pm4_set_reg(pm4, R_028230_PA_SC_EDGERULE,
si_pm4_set_reg(pm4, R_028AC4_DB_SRESULTS_COMPARE_STATE1, 0x0);
si_pm4_set_reg(pm4, R_028AC8_DB_PRELOAD_CONTROL, 0x0);
si_pm4_set_reg(pm4, R_02800C_DB_RENDER_OVERRIDE, 0);
+ si_pm4_set_reg(pm4, R_028A5C_VGT_GS_PER_VS, 0x2);
+ si_pm4_set_reg(pm4, R_028A8C_VGT_PRIMITIVEID_RESET, 0x0);
+ si_pm4_set_reg(pm4, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0x0);
+ si_pm4_set_reg(pm4, R_028AB8_VGT_VTX_CNT_EN, 0x0);
}
- if (sctx->chip_class >= GFX10) {
- si_pm4_set_reg(pm4, R_028A98_VGT_DRAW_PAYLOAD_CNTL, 0);
- si_pm4_set_reg(pm4, R_030964_GE_MAX_VTX_INDX, ~0);
- si_pm4_set_reg(pm4, R_030924_GE_MIN_VTX_INDX, 0);
- si_pm4_set_reg(pm4, R_030928_GE_INDX_OFFSET, 0);
- si_pm4_set_reg(pm4, R_03097C_GE_STEREO_CNTL, 0);
- si_pm4_set_reg(pm4, R_030988_GE_USER_VGPR_EN, 0);
- } else if (sctx->chip_class == GFX9) {
- si_pm4_set_reg(pm4, R_030920_VGT_MAX_VTX_INDX, ~0);
- si_pm4_set_reg(pm4, R_030924_VGT_MIN_VTX_INDX, 0);
- si_pm4_set_reg(pm4, R_030928_VGT_INDX_OFFSET, 0);
- } else {
- /* These registers, when written, also overwrite the CLEAR_STATE
- * context, so we can't rely on CLEAR_STATE setting them.
- * It would be an issue if there was another UMD changing them.
- */
- si_pm4_set_reg(pm4, R_028400_VGT_MAX_VTX_INDX, ~0);
- si_pm4_set_reg(pm4, R_028404_VGT_MIN_VTX_INDX, 0);
- si_pm4_set_reg(pm4, R_028408_VGT_INDX_OFFSET, 0);
+ si_pm4_set_reg(pm4, R_028080_TA_BC_BASE_ADDR, border_color_va >> 8);
+ if (sctx->chip_class >= GFX7)
+ si_pm4_set_reg(pm4, R_028084_TA_BC_BASE_ADDR_HI, S_028084_ADDRESS(border_color_va >> 40));
+
+ if (sctx->chip_class == GFX6) {
+ si_pm4_set_reg(pm4, R_008A14_PA_CL_ENHANCE,
+ S_008A14_NUM_CLIP_SEQ(3) | S_008A14_CLIP_VTX_REORDER_ENA(1));
}
- if (sctx->chip_class >= GFX7) {
- if (sctx->chip_class >= GFX10) {
- /* Logical CUs 16 - 31 */
- si_pm4_set_reg(pm4, R_00B404_SPI_SHADER_PGM_RSRC4_HS, S_00B404_CU_EN(0xffff));
- si_pm4_set_reg(pm4, R_00B104_SPI_SHADER_PGM_RSRC4_VS, S_00B104_CU_EN(0xffff));
- si_pm4_set_reg(pm4, R_00B004_SPI_SHADER_PGM_RSRC4_PS, S_00B004_CU_EN(0xffff));
- }
+ if (sctx->chip_class <= GFX7 || !has_clear_state) {
+ si_pm4_set_reg(pm4, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
+ si_pm4_set_reg(pm4, R_028C5C_VGT_OUT_DEALLOC_CNTL, 16);
- if (sctx->chip_class >= GFX9) {
- si_pm4_set_reg(pm4, R_00B41C_SPI_SHADER_PGM_RSRC3_HS,
- S_00B41C_CU_EN(0xffff) | S_00B41C_WAVE_LIMIT(0x3F));
- } else {
- si_pm4_set_reg(pm4, R_00B51C_SPI_SHADER_PGM_RSRC3_LS,
- S_00B51C_CU_EN(0xffff) | S_00B51C_WAVE_LIMIT(0x3F));
- si_pm4_set_reg(pm4, R_00B41C_SPI_SHADER_PGM_RSRC3_HS, S_00B41C_WAVE_LIMIT(0x3F));
- si_pm4_set_reg(pm4, R_00B31C_SPI_SHADER_PGM_RSRC3_ES,
- S_00B31C_CU_EN(0xffff) | S_00B31C_WAVE_LIMIT(0x3F));
-
- /* If this is 0, Bonaire can hang even if GS isn't being used.
- * Other chips are unaffected. These are suboptimal values,
- * but we don't use on-chip GS.
- */
- si_pm4_set_reg(pm4, R_028A44_VGT_GS_ONCHIP_CNTL,
- S_028A44_ES_VERTS_PER_SUBGRP(64) | S_028A44_GS_PRIMS_PER_SUBGRP(4));
- }
+ /* CLEAR_STATE doesn't clear these correctly on certain generations.
+ * I don't know why. Deduced by trial and error.
+ */
+ si_pm4_set_reg(pm4, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
+ si_pm4_set_reg(pm4, R_028204_PA_SC_WINDOW_SCISSOR_TL, S_028204_WINDOW_OFFSET_DISABLE(1));
+ si_pm4_set_reg(pm4, R_028030_PA_SC_SCREEN_SCISSOR_TL, 0);
+ si_pm4_set_reg(pm4, R_028034_PA_SC_SCREEN_SCISSOR_BR,
+ S_028034_BR_X(16384) | S_028034_BR_Y(16384));
+ }
+ if (sctx->chip_class >= GFX7) {
/* Compute LATE_ALLOC_VS.LIMIT. */
unsigned num_cu_per_sh = sscreen->info.min_good_cu_per_sa;
unsigned late_alloc_wave64 = 0; /* The limit is per SA. */
si_pm4_set_reg(pm4, R_00B118_SPI_SHADER_PGM_RSRC3_VS,
S_00B118_CU_EN(cu_mask_vs) | S_00B118_WAVE_LIMIT(0x3F));
si_pm4_set_reg(pm4, R_00B11C_SPI_SHADER_LATE_ALLOC_VS, S_00B11C_LIMIT(late_alloc_wave64));
-
si_pm4_set_reg(pm4, R_00B21C_SPI_SHADER_PGM_RSRC3_GS,
S_00B21C_CU_EN(cu_mask_gs) | S_00B21C_WAVE_LIMIT(0x3F));
-
si_pm4_set_reg(pm4, R_00B01C_SPI_SHADER_PGM_RSRC3_PS,
S_00B01C_CU_EN(0xffff) | S_00B01C_WAVE_LIMIT(0x3F));
}
- if (sctx->chip_class >= GFX10) {
- /* Break up a pixel wave if it contains deallocs for more than
- * half the parameter cache.
- *
- * To avoid a deadlock where pixel waves aren't launched
- * because they're waiting for more pixels while the frontend
- * is stuck waiting for PC space, the maximum allowed value is
- * the size of the PC minus the largest possible allocation for
- * a single primitive shader subgroup.
- */
- si_pm4_set_reg(pm4, R_028C50_PA_SC_NGG_MODE_CNTL, S_028C50_MAX_DEALLOCS_IN_WAVE(512));
- /* Reuse for legacy (non-NGG) only. */
- si_pm4_set_reg(pm4, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
-
- if (!has_clear_state) {
- si_pm4_set_reg(pm4, R_02835C_PA_SC_TILE_STEERING_OVERRIDE,
- sscreen->info.pa_sc_tile_steering_override);
- }
+ if (sctx->chip_class <= GFX8) {
+ si_set_raster_config(sctx, pm4);
- si_pm4_set_reg(pm4, R_028428_CB_COVERAGE_OUT_CONTROL, 0);
+ /* FIXME calculate these values somehow ??? */
+ si_pm4_set_reg(pm4, R_028A54_VGT_GS_PER_ES, SI_GS_PER_ES);
+ si_pm4_set_reg(pm4, R_028A58_VGT_ES_PER_GS, 0x40);
- si_pm4_set_reg(pm4, R_00B0C0_SPI_SHADER_REQ_CTRL_PS,
- S_00B0C0_SOFT_GROUPING_EN(1) | S_00B0C0_NUMBER_OF_REQUESTS_PER_CU(4 - 1));
- si_pm4_set_reg(pm4, R_00B1C0_SPI_SHADER_REQ_CTRL_VS, 0);
+ /* These registers, when written, also overwrite the CLEAR_STATE
+ * context, so we can't rely on CLEAR_STATE setting them.
+ * It would be an issue if there was another UMD changing them.
+ */
+ si_pm4_set_reg(pm4, R_028400_VGT_MAX_VTX_INDX, ~0);
+ si_pm4_set_reg(pm4, R_028404_VGT_MIN_VTX_INDX, 0);
+ si_pm4_set_reg(pm4, R_028408_VGT_INDX_OFFSET, 0);
}
- if (sctx->chip_class >= GFX10_3) {
- si_pm4_set_reg(pm4, R_028750_SX_PS_DOWNCONVERT_CONTROL_GFX103, 0xff);
+
+ if (sctx->chip_class >= GFX7 && sctx->chip_class <= GFX8) {
+ si_pm4_set_reg(pm4, R_00B51C_SPI_SHADER_PGM_RSRC3_LS,
+ S_00B51C_CU_EN(0xffff) | S_00B51C_WAVE_LIMIT(0x3F));
+ si_pm4_set_reg(pm4, R_00B41C_SPI_SHADER_PGM_RSRC3_HS, S_00B41C_WAVE_LIMIT(0x3F));
+ si_pm4_set_reg(pm4, R_00B31C_SPI_SHADER_PGM_RSRC3_ES,
+ S_00B31C_CU_EN(0xffff) | S_00B31C_WAVE_LIMIT(0x3F));
+
+ /* If this is 0, Bonaire can hang even if GS isn't being used.
+ * Other chips are unaffected. These are suboptimal values,
+ * but we don't use on-chip GS.
+ */
+ si_pm4_set_reg(pm4, R_028A44_VGT_GS_ONCHIP_CNTL,
+ S_028A44_ES_VERTS_PER_SUBGRP(64) | S_028A44_GS_PRIMS_PER_SUBGRP(4));
}
- if (sctx->chip_class >= GFX9) {
- si_pm4_set_reg(pm4, R_028B50_VGT_TESS_DISTRIBUTION,
- S_028B50_ACCUM_ISOLINE(40) | S_028B50_ACCUM_TRI(30) | S_028B50_ACCUM_QUAD(24) |
- S_028B50_DONUT_SPLIT(24) | S_028B50_TRAP_SPLIT(6));
- } else if (sctx->chip_class >= GFX8) {
+ if (sctx->chip_class == GFX8) {
unsigned vgt_tess_distribution;
vgt_tess_distribution = S_028B50_ACCUM_ISOLINE(32) | S_028B50_ACCUM_TRI(11) |
vgt_tess_distribution |= S_028B50_TRAP_SPLIT(3);
si_pm4_set_reg(pm4, R_028B50_VGT_TESS_DISTRIBUTION, vgt_tess_distribution);
- } else if (!has_clear_state) {
- si_pm4_set_reg(pm4, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
- si_pm4_set_reg(pm4, R_028C5C_VGT_OUT_DEALLOC_CNTL, 16);
}
- si_pm4_set_reg(pm4, R_028080_TA_BC_BASE_ADDR, border_color_va >> 8);
- if (sctx->chip_class >= GFX7) {
- si_pm4_set_reg(pm4, R_028084_TA_BC_BASE_ADDR_HI, S_028084_ADDRESS(border_color_va >> 40));
+ if (sscreen->info.chip_class <= GFX9) {
+ si_pm4_set_reg(pm4, R_028AA0_VGT_INSTANCE_STEP_RATE_0, 1);
+ }
+
+ if (sctx->chip_class == GFX9) {
+ si_pm4_set_reg(pm4, R_030920_VGT_MAX_VTX_INDX, ~0);
+ si_pm4_set_reg(pm4, R_030924_VGT_MIN_VTX_INDX, 0);
+ si_pm4_set_reg(pm4, R_030928_VGT_INDX_OFFSET, 0);
}
- si_pm4_add_bo(pm4, sctx->border_color_buffer, RADEON_USAGE_READ, RADEON_PRIO_BORDER_COLORS);
if (sctx->chip_class >= GFX9) {
+ si_pm4_set_reg(pm4, R_00B41C_SPI_SHADER_PGM_RSRC3_HS,
+ S_00B41C_CU_EN(0xffff) | S_00B41C_WAVE_LIMIT(0x3F));
+
+ si_pm4_set_reg(pm4, R_028B50_VGT_TESS_DISTRIBUTION,
+ S_028B50_ACCUM_ISOLINE(40) | S_028B50_ACCUM_TRI(30) | S_028B50_ACCUM_QUAD(24) |
+ S_028B50_DONUT_SPLIT(24) | S_028B50_TRAP_SPLIT(6));
si_pm4_set_reg(pm4, R_028C48_PA_SC_BINNER_CNTL_1,
S_028C48_MAX_ALLOC_COUNT(sscreen->info.pbb_max_alloc_count - 1) |
- S_028C48_MAX_PRIM_PER_BATCH(1023));
+ S_028C48_MAX_PRIM_PER_BATCH(1023));
si_pm4_set_reg(pm4, R_028C4C_PA_SC_CONSERVATIVE_RASTERIZATION_CNTL,
S_028C4C_NULL_SQUAD_AA_MASK_ENABLE(1));
+
si_pm4_set_reg(pm4, R_030968_VGT_INSTANCE_BASE_ID, 0);
+ si_pm4_set_reg(pm4, R_0301EC_CP_COHER_START_DELAY,
+ sctx->chip_class >= GFX10 ? 0x20 : 0);
+ }
+
+ if (sctx->chip_class >= GFX10) {
+ /* Logical CUs 16 - 31 */
+ si_pm4_set_reg(pm4, R_00B004_SPI_SHADER_PGM_RSRC4_PS, S_00B004_CU_EN(0xffff));
+ si_pm4_set_reg(pm4, R_00B104_SPI_SHADER_PGM_RSRC4_VS, S_00B104_CU_EN(0xffff));
+ si_pm4_set_reg(pm4, R_00B404_SPI_SHADER_PGM_RSRC4_HS, S_00B404_CU_EN(0xffff));
+
+ si_pm4_set_reg(pm4, R_00B0C8_SPI_SHADER_USER_ACCUM_PS_0, 0);
+ si_pm4_set_reg(pm4, R_00B0CC_SPI_SHADER_USER_ACCUM_PS_1, 0);
+ si_pm4_set_reg(pm4, R_00B0D0_SPI_SHADER_USER_ACCUM_PS_2, 0);
+ si_pm4_set_reg(pm4, R_00B0D4_SPI_SHADER_USER_ACCUM_PS_3, 0);
+ si_pm4_set_reg(pm4, R_00B1C8_SPI_SHADER_USER_ACCUM_VS_0, 0);
+ si_pm4_set_reg(pm4, R_00B1CC_SPI_SHADER_USER_ACCUM_VS_1, 0);
+ si_pm4_set_reg(pm4, R_00B1D0_SPI_SHADER_USER_ACCUM_VS_2, 0);
+ si_pm4_set_reg(pm4, R_00B1D4_SPI_SHADER_USER_ACCUM_VS_3, 0);
+ si_pm4_set_reg(pm4, R_00B2C8_SPI_SHADER_USER_ACCUM_ESGS_0, 0);
+ si_pm4_set_reg(pm4, R_00B2CC_SPI_SHADER_USER_ACCUM_ESGS_1, 0);
+ si_pm4_set_reg(pm4, R_00B2D0_SPI_SHADER_USER_ACCUM_ESGS_2, 0);
+ si_pm4_set_reg(pm4, R_00B2D4_SPI_SHADER_USER_ACCUM_ESGS_3, 0);
+ si_pm4_set_reg(pm4, R_00B4C8_SPI_SHADER_USER_ACCUM_LSHS_0, 0);
+ si_pm4_set_reg(pm4, R_00B4CC_SPI_SHADER_USER_ACCUM_LSHS_1, 0);
+ si_pm4_set_reg(pm4, R_00B4D0_SPI_SHADER_USER_ACCUM_LSHS_2, 0);
+ si_pm4_set_reg(pm4, R_00B4D4_SPI_SHADER_USER_ACCUM_LSHS_3, 0);
+
+ si_pm4_set_reg(pm4, R_00B0C0_SPI_SHADER_REQ_CTRL_PS,
+ S_00B0C0_SOFT_GROUPING_EN(1) |
+ S_00B0C0_NUMBER_OF_REQUESTS_PER_CU(4 - 1));
+ si_pm4_set_reg(pm4, R_00B1C0_SPI_SHADER_REQ_CTRL_VS, 0);
+
+ si_pm4_set_reg(pm4, R_028428_CB_COVERAGE_OUT_CONTROL, 0);
+ si_pm4_set_reg(pm4, R_028A98_VGT_DRAW_PAYLOAD_CNTL, 0);
+
+ /* Break up a pixel wave if it contains deallocs for more than
+ * half the parameter cache.
+ *
+ * To avoid a deadlock where pixel waves aren't launched
+ * because they're waiting for more pixels while the frontend
+ * is stuck waiting for PC space, the maximum allowed value is
+ * the size of the PC minus the largest possible allocation for
+ * a single primitive shader subgroup.
+ */
+ si_pm4_set_reg(pm4, R_028C50_PA_SC_NGG_MODE_CNTL, S_028C50_MAX_DEALLOCS_IN_WAVE(512));
+ /* Reuse for legacy (non-NGG) only. */
+ si_pm4_set_reg(pm4, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
+
+ if (!has_clear_state) {
+ si_pm4_set_reg(pm4, R_02835C_PA_SC_TILE_STEERING_OVERRIDE,
+ sscreen->info.pa_sc_tile_steering_override);
+ }
+
+
+ si_pm4_set_reg(pm4, R_030964_GE_MAX_VTX_INDX, ~0);
+ si_pm4_set_reg(pm4, R_030924_GE_MIN_VTX_INDX, 0);
+ si_pm4_set_reg(pm4, R_030928_GE_INDX_OFFSET, 0);
+ si_pm4_set_reg(pm4, R_03097C_GE_STEREO_CNTL, 0);
+ si_pm4_set_reg(pm4, R_030988_GE_USER_VGPR_EN, 0);
+ }
+
+ if (sctx->chip_class >= GFX10_3) {
+ si_pm4_set_reg(pm4, R_028750_SX_PS_DOWNCONVERT_CONTROL, 0xff);
+ si_pm4_set_reg(pm4, 0x28848, 1 << 9); /* This fixes sample shading. */
}
- sctx->init_config = pm4;
+ sctx->cs_preamble_state = pm4;
}
projects / mesa.git / blobdiff