return shader->pm4;
}
+static unsigned si_get_num_vs_user_sgprs(unsigned num_always_on_user_sgprs)
+{
+ /* Add the pointer to VBO descriptors. */
+ if (HAVE_32BIT_POINTERS) {
+ return num_always_on_user_sgprs + 1;
+ } else {
+ assert(num_always_on_user_sgprs % 2 == 0);
+ return num_always_on_user_sgprs + 2;
+ }
+}
+
static void si_shader_ls(struct si_screen *sscreen, struct si_shader *shader)
{
struct si_pm4_state *pm4;
vgpr_comp_cnt = shader->info.uses_instanceid ? 2 : 1;
si_pm4_set_reg(pm4, R_00B520_SPI_SHADER_PGM_LO_LS, va >> 8);
- si_pm4_set_reg(pm4, R_00B524_SPI_SHADER_PGM_HI_LS, va >> 40);
+ si_pm4_set_reg(pm4, R_00B524_SPI_SHADER_PGM_HI_LS, S_00B524_MEM_BASE(va >> 40));
shader->config.rsrc1 = S_00B528_VGPRS((shader->config.num_vgprs - 1) / 4) |
S_00B528_SGPRS((shader->config.num_sgprs - 1) / 8) |
S_00B528_VGPR_COMP_CNT(vgpr_comp_cnt) |
S_00B528_DX10_CLAMP(1) |
S_00B528_FLOAT_MODE(shader->config.float_mode);
- shader->config.rsrc2 = S_00B52C_USER_SGPR(SI_VS_NUM_USER_SGPR) |
+ shader->config.rsrc2 = S_00B52C_USER_SGPR(si_get_num_vs_user_sgprs(SI_VS_NUM_USER_SGPR)) |
S_00B52C_SCRATCH_EN(shader->config.scratch_bytes_per_wave > 0);
}
if (sscreen->info.chip_class >= GFX9) {
si_pm4_set_reg(pm4, R_00B410_SPI_SHADER_PGM_LO_LS, va >> 8);
- si_pm4_set_reg(pm4, R_00B414_SPI_SHADER_PGM_HI_LS, va >> 40);
+ si_pm4_set_reg(pm4, R_00B414_SPI_SHADER_PGM_HI_LS, S_00B414_MEM_BASE(va >> 40));
/* We need at least 2 components for LS.
* VGPR0-3: (VertexID, RelAutoindex, InstanceID / StepRate0, InstanceID).
*/
ls_vgpr_comp_cnt = shader->info.uses_instanceid ? 2 : 1;
+ unsigned num_user_sgprs =
+ si_get_num_vs_user_sgprs(GFX9_TCS_NUM_USER_SGPR);
+
shader->config.rsrc2 =
- S_00B42C_USER_SGPR(GFX9_TCS_NUM_USER_SGPR) |
- S_00B42C_USER_SGPR_MSB(GFX9_TCS_NUM_USER_SGPR >> 5) |
+ S_00B42C_USER_SGPR(num_user_sgprs) |
+ S_00B42C_USER_SGPR_MSB(num_user_sgprs >> 5) |
S_00B42C_SCRATCH_EN(shader->config.scratch_bytes_per_wave > 0);
} else {
si_pm4_set_reg(pm4, R_00B420_SPI_SHADER_PGM_LO_HS, va >> 8);
- si_pm4_set_reg(pm4, R_00B424_SPI_SHADER_PGM_HI_HS, va >> 40);
+ si_pm4_set_reg(pm4, R_00B424_SPI_SHADER_PGM_HI_HS, S_00B424_MEM_BASE(va >> 40));
shader->config.rsrc2 =
S_00B42C_USER_SGPR(GFX6_TCS_NUM_USER_SGPR) |
if (shader->selector->type == PIPE_SHADER_VERTEX) {
/* VGPR0-3: (VertexID, InstanceID / StepRate0, ...) */
vgpr_comp_cnt = shader->info.uses_instanceid ? 1 : 0;
- num_user_sgprs = SI_VS_NUM_USER_SGPR;
+ num_user_sgprs = si_get_num_vs_user_sgprs(SI_VS_NUM_USER_SGPR);
} else if (shader->selector->type == PIPE_SHADER_TESS_EVAL) {
vgpr_comp_cnt = shader->selector->info.uses_primid ? 3 : 2;
num_user_sgprs = SI_TES_NUM_USER_SGPR;
si_pm4_set_reg(pm4, R_028AAC_VGT_ESGS_RING_ITEMSIZE,
shader->selector->esgs_itemsize / 4);
si_pm4_set_reg(pm4, R_00B320_SPI_SHADER_PGM_LO_ES, va >> 8);
- si_pm4_set_reg(pm4, R_00B324_SPI_SHADER_PGM_HI_ES, va >> 40);
+ si_pm4_set_reg(pm4, R_00B324_SPI_SHADER_PGM_HI_ES, S_00B324_MEM_BASE(va >> 40));
si_pm4_set_reg(pm4, R_00B328_SPI_SHADER_PGM_RSRC1_ES,
S_00B328_VGPRS((shader->config.num_vgprs - 1) / 4) |
S_00B328_SGPRS((shader->config.num_sgprs - 1) / 8) |
else
gs_vgpr_comp_cnt = 0; /* VGPR0 contains offsets 0, 1 */
+ unsigned num_user_sgprs;
+ if (es_type == PIPE_SHADER_VERTEX)
+ num_user_sgprs = si_get_num_vs_user_sgprs(GFX9_VSGS_NUM_USER_SGPR);
+ else
+ num_user_sgprs = GFX9_TESGS_NUM_USER_SGPR;
+
gfx9_get_gs_info(shader->key.part.gs.es, sel, &gs_info);
si_pm4_set_reg(pm4, R_00B210_SPI_SHADER_PGM_LO_ES, va >> 8);
- si_pm4_set_reg(pm4, R_00B214_SPI_SHADER_PGM_HI_ES, va >> 40);
+ si_pm4_set_reg(pm4, R_00B214_SPI_SHADER_PGM_HI_ES, S_00B214_MEM_BASE(va >> 40));
si_pm4_set_reg(pm4, R_00B228_SPI_SHADER_PGM_RSRC1_GS,
S_00B228_VGPRS((shader->config.num_vgprs - 1) / 4) |
S_00B228_FLOAT_MODE(shader->config.float_mode) |
S_00B228_GS_VGPR_COMP_CNT(gs_vgpr_comp_cnt));
si_pm4_set_reg(pm4, R_00B22C_SPI_SHADER_PGM_RSRC2_GS,
- S_00B22C_USER_SGPR(GFX9_GS_NUM_USER_SGPR) |
- S_00B22C_USER_SGPR_MSB(GFX9_GS_NUM_USER_SGPR >> 5) |
+ S_00B22C_USER_SGPR(num_user_sgprs) |
+ S_00B22C_USER_SGPR_MSB(num_user_sgprs >> 5) |
S_00B22C_ES_VGPR_COMP_CNT(es_vgpr_comp_cnt) |
S_00B22C_OC_LDS_EN(es_type == PIPE_SHADER_TESS_EVAL) |
S_00B22C_LDS_SIZE(gs_info.lds_size) |
NULL, pm4);
} else {
si_pm4_set_reg(pm4, R_00B220_SPI_SHADER_PGM_LO_GS, va >> 8);
- si_pm4_set_reg(pm4, R_00B224_SPI_SHADER_PGM_HI_GS, va >> 40);
+ si_pm4_set_reg(pm4, R_00B224_SPI_SHADER_PGM_HI_GS, S_00B224_MEM_BASE(va >> 40));
si_pm4_set_reg(pm4, R_00B228_SPI_SHADER_PGM_RSRC1_GS,
S_00B228_VGPRS((shader->config.num_vgprs - 1) / 4) |
num_user_sgprs = SI_SGPR_VS_BLIT_DATA +
info->properties[TGSI_PROPERTY_VS_BLIT_SGPRS];
} else {
- num_user_sgprs = SI_VS_NUM_USER_SGPR;
+ num_user_sgprs = si_get_num_vs_user_sgprs(SI_VS_NUM_USER_SGPR);
}
} else if (shader->selector->type == PIPE_SHADER_TESS_EVAL) {
vgpr_comp_cnt = enable_prim_id ? 3 : 2;
oc_lds_en = shader->selector->type == PIPE_SHADER_TESS_EVAL ? 1 : 0;
si_pm4_set_reg(pm4, R_00B120_SPI_SHADER_PGM_LO_VS, va >> 8);
- si_pm4_set_reg(pm4, R_00B124_SPI_SHADER_PGM_HI_VS, va >> 40);
+ si_pm4_set_reg(pm4, R_00B124_SPI_SHADER_PGM_HI_VS, S_00B124_MEM_BASE(va >> 40));
si_pm4_set_reg(pm4, R_00B128_SPI_SHADER_PGM_RSRC1_VS,
S_00B128_VGPRS((shader->config.num_vgprs - 1) / 4) |
S_00B128_SGPRS((shader->config.num_sgprs - 1) / 8) |
va = shader->bo->gpu_address;
si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
si_pm4_set_reg(pm4, R_00B020_SPI_SHADER_PGM_LO_PS, va >> 8);
- si_pm4_set_reg(pm4, R_00B024_SPI_SHADER_PGM_HI_PS, va >> 40);
+ si_pm4_set_reg(pm4, R_00B024_SPI_SHADER_PGM_HI_PS, S_00B024_MEM_BASE(va >> 40));
si_pm4_set_reg(pm4, R_00B028_SPI_SHADER_PGM_RSRC1_PS,
S_00B028_VGPRS((shader->config.num_vgprs - 1) / 4) |
ps->info.writes_samplemask ||
alpha_to_coverage ||
si_get_alpha_test_func(sctx) != PIPE_FUNC_ALWAYS;
-
- unsigned ps_colormask = sctx->framebuffer.colorbuf_enabled_4bit &
- sctx->queued.named.blend->cb_target_mask;
- if (!ps->info.properties[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS])
- ps_colormask &= ps->colors_written_4bit;
+ unsigned ps_colormask = si_get_total_colormask(sctx);
ps_disabled = sctx->queued.named.rasterizer->rasterizer_discard ||
(!ps_colormask &&
si_shader_selector_key_vs(sctx, sctx->vs_shader.cso,
key, &key->part.gs.vs_prolog);
key->part.gs.es = sctx->vs_shader.cso;
+ key->part.gs.prolog.gfx9_prev_is_vs = 1;
}
/* Merged ES-GS can have unbalanced wave usage.
}
key->part.ps.epilog.alpha_func = si_get_alpha_test_func(sctx);
+
+ /* ps_uses_fbfetch is true only if the color buffer is bound. */
+ if (sctx->ps_uses_fbfetch) {
+ struct pipe_surface *cb0 = sctx->framebuffer.state.cbufs[0];
+ struct pipe_resource *tex = cb0->texture;
+
+ /* 1D textures are allocated and used as 2D on GFX9. */
+ key->mono.u.ps.fbfetch_msaa = sctx->framebuffer.nr_samples > 1;
+ key->mono.u.ps.fbfetch_is_1D = sctx->b.chip_class != GFX9 &&
+ (tex->target == PIPE_TEXTURE_1D ||
+ tex->target == PIPE_TEXTURE_1D_ARRAY);
+ key->mono.u.ps.fbfetch_layered = tex->target == PIPE_TEXTURE_1D_ARRAY ||
+ tex->target == PIPE_TEXTURE_2D_ARRAY ||
+ tex->target == PIPE_TEXTURE_CUBE ||
+ tex->target == PIPE_TEXTURE_CUBE_ARRAY ||
+ tex->target == PIPE_TEXTURE_3D;
+ }
break;
}
default:
si_mark_atom_dirty(sctx, &sctx->msaa_config);
}
si_set_active_descriptors_for_shader(sctx, sel);
+ si_update_ps_colorbuf0_slot(sctx);
}
static void si_delete_shader(struct si_context *sctx, struct si_shader *shader)
/* Flush the context to re-emit both init_config states. */
sctx->b.initial_gfx_cs_size = 0; /* force flush */
- si_context_gfx_flush(sctx, PIPE_FLUSH_ASYNC, NULL);
+ si_flush_gfx_cs(sctx, PIPE_FLUSH_ASYNC, NULL);
/* Set ring bindings. */
if (sctx->esgs_ring) {
static void si_init_tess_factor_ring(struct si_context *sctx)
{
- bool double_offchip_buffers = sctx->b.chip_class >= CIK &&
- sctx->b.family != CHIP_CARRIZO &&
- sctx->b.family != CHIP_STONEY;
- /* This must be one less than the maximum number due to a hw limitation.
- * Various hardware bugs in SI, CIK, and GFX9 need this.
- */
- unsigned max_offchip_buffers_per_se = double_offchip_buffers ? 127 : 63;
- unsigned max_offchip_buffers = max_offchip_buffers_per_se *
- sctx->screen->info.max_se;
- unsigned offchip_granularity;
-
- switch (sctx->screen->tess_offchip_block_dw_size) {
- default:
- assert(0);
- /* fall through */
- case 8192:
- offchip_granularity = V_03093C_X_8K_DWORDS;
- break;
- case 4096:
- offchip_granularity = V_03093C_X_4K_DWORDS;
- break;
- }
+ assert(!sctx->tess_rings);
- assert(!sctx->tf_ring);
- /* Use 64K alignment for both rings, so that we can pass the address
- * to shaders as one SGPR containing bits [16:47].
+ /* The address must be aligned to 2^19, because the shader only
+ * receives the high 13 bits.
*/
- sctx->tf_ring = si_aligned_buffer_create(sctx->b.b.screen,
- R600_RESOURCE_FLAG_UNMAPPABLE,
- PIPE_USAGE_DEFAULT,
- 32768 * sctx->screen->info.max_se,
- 64 * 1024);
- if (!sctx->tf_ring)
- return;
-
- assert(((sctx->tf_ring->width0 / 4) & C_030938_SIZE) == 0);
-
- sctx->tess_offchip_ring =
- si_aligned_buffer_create(sctx->b.b.screen,
- R600_RESOURCE_FLAG_UNMAPPABLE,
- PIPE_USAGE_DEFAULT,
- max_offchip_buffers *
- sctx->screen->tess_offchip_block_dw_size * 4,
- 64 * 1024);
- if (!sctx->tess_offchip_ring)
+ sctx->tess_rings = si_aligned_buffer_create(sctx->b.b.screen,
+ R600_RESOURCE_FLAG_32BIT,
+ PIPE_USAGE_DEFAULT,
+ sctx->screen->tess_offchip_ring_size +
+ sctx->screen->tess_factor_ring_size,
+ 1 << 19);
+ if (!sctx->tess_rings)
return;
si_init_config_add_vgt_flush(sctx);
- uint64_t offchip_va = r600_resource(sctx->tess_offchip_ring)->gpu_address;
- uint64_t factor_va = r600_resource(sctx->tf_ring)->gpu_address;
- assert((offchip_va & 0xffff) == 0);
- assert((factor_va & 0xffff) == 0);
-
- si_pm4_add_bo(sctx->init_config, r600_resource(sctx->tess_offchip_ring),
- RADEON_USAGE_READWRITE, RADEON_PRIO_SHADER_RINGS);
- si_pm4_add_bo(sctx->init_config, r600_resource(sctx->tf_ring),
+ si_pm4_add_bo(sctx->init_config, r600_resource(sctx->tess_rings),
RADEON_USAGE_READWRITE, RADEON_PRIO_SHADER_RINGS);
+ uint64_t factor_va = r600_resource(sctx->tess_rings)->gpu_address +
+ sctx->screen->tess_offchip_ring_size;
+
/* Append these registers to the init config state. */
if (sctx->b.chip_class >= CIK) {
- if (sctx->b.chip_class >= VI)
- --max_offchip_buffers;
-
si_pm4_set_reg(sctx->init_config, R_030938_VGT_TF_RING_SIZE,
- S_030938_SIZE(sctx->tf_ring->width0 / 4));
+ S_030938_SIZE(sctx->screen->tess_factor_ring_size / 4));
si_pm4_set_reg(sctx->init_config, R_030940_VGT_TF_MEMORY_BASE,
factor_va >> 8);
if (sctx->b.chip_class >= GFX9)
si_pm4_set_reg(sctx->init_config, R_030944_VGT_TF_MEMORY_BASE_HI,
- factor_va >> 40);
+ S_030944_BASE_HI(factor_va >> 40));
si_pm4_set_reg(sctx->init_config, R_03093C_VGT_HS_OFFCHIP_PARAM,
- S_03093C_OFFCHIP_BUFFERING(max_offchip_buffers) |
- S_03093C_OFFCHIP_GRANULARITY(offchip_granularity));
+ sctx->screen->vgt_hs_offchip_param);
} else {
- assert(offchip_granularity == V_03093C_X_8K_DWORDS);
si_pm4_set_reg(sctx->init_config, R_008988_VGT_TF_RING_SIZE,
- S_008988_SIZE(sctx->tf_ring->width0 / 4));
+ S_008988_SIZE(sctx->screen->tess_factor_ring_size / 4));
si_pm4_set_reg(sctx->init_config, R_0089B8_VGT_TF_MEMORY_BASE,
factor_va >> 8);
si_pm4_set_reg(sctx->init_config, R_0089B0_VGT_HS_OFFCHIP_PARAM,
- S_0089B0_OFFCHIP_BUFFERING(max_offchip_buffers));
- }
-
- if (sctx->b.chip_class >= GFX9) {
- si_pm4_set_reg(sctx->init_config,
- R_00B430_SPI_SHADER_USER_DATA_LS_0 +
- GFX9_SGPR_TCS_OFFCHIP_ADDR_BASE64K * 4,
- offchip_va >> 16);
- si_pm4_set_reg(sctx->init_config,
- R_00B430_SPI_SHADER_USER_DATA_LS_0 +
- GFX9_SGPR_TCS_FACTOR_ADDR_BASE64K * 4,
- factor_va >> 16);
- } else {
- si_pm4_set_reg(sctx->init_config,
- R_00B430_SPI_SHADER_USER_DATA_HS_0 +
- GFX6_SGPR_TCS_OFFCHIP_ADDR_BASE64K * 4,
- offchip_va >> 16);
- si_pm4_set_reg(sctx->init_config,
- R_00B430_SPI_SHADER_USER_DATA_HS_0 +
- GFX6_SGPR_TCS_FACTOR_ADDR_BASE64K * 4,
- factor_va >> 16);
+ sctx->screen->vgt_hs_offchip_param);
}
/* Flush the context to re-emit the init_config state.
*/
si_pm4_upload_indirect_buffer(sctx, sctx->init_config);
sctx->b.initial_gfx_cs_size = 0; /* force flush */
- si_context_gfx_flush(sctx, PIPE_FLUSH_ASYNC, NULL);
+ si_flush_gfx_cs(sctx, PIPE_FLUSH_ASYNC, NULL);
}
/**
/* Update stages before GS. */
if (sctx->tes_shader.cso) {
- if (!sctx->tf_ring) {
+ if (!sctx->tess_rings) {
si_init_tess_factor_ring(sctx);
- if (!sctx->tf_ring)
+ if (!sctx->tess_rings)
return false;
}
projects / mesa.git / blobdiff