radeon_opt_set_context_reg(sctx, R_028AAC_VGT_ESGS_RING_ITEMSIZE,
SI_TRACKED_VGT_ESGS_RING_ITEMSIZE,
shader->ctx_reg.ngg.vgt_esgs_ring_itemsize);
- radeon_opt_set_context_reg(sctx, R_028AB4_VGT_REUSE_OFF,
- SI_TRACKED_VGT_REUSE_OFF,
- shader->ctx_reg.ngg.vgt_reuse_off);
radeon_opt_set_context_reg(sctx, R_0286C4_SPI_VS_OUT_CONFIG,
SI_TRACKED_SPI_VS_OUT_CONFIG,
shader->ctx_reg.ngg.spi_vs_out_config);
gfx10_emit_shader_ngg_tail(sctx, shader, initial_cdw);
}
-static void si_set_ge_pc_alloc(struct si_screen *sscreen,
- struct si_pm4_state *pm4, bool culling)
-{
- si_pm4_set_reg(pm4, R_030980_GE_PC_ALLOC,
- S_030980_OVERSUB_EN(1) |
- S_030980_NUM_PC_LINES((culling ? 256 : 128) * sscreen->info.max_se - 1));
-}
-
unsigned si_get_input_prim(const struct si_shader_selector *gs)
{
if (gs->type == PIPE_SHADER_GEOMETRY)
S_00B22C_USER_SGPR_MSB_GFX10(num_user_sgprs >> 5) |
S_00B22C_OC_LDS_EN(es_type == PIPE_SHADER_TESS_EVAL) |
S_00B22C_LDS_SIZE(shader->config.lds_size));
- si_set_ge_pc_alloc(sscreen, pm4, false);
nparams = MAX2(shader->info.nr_param_exports, 1);
shader->ctx_reg.ngg.spi_vs_out_config =
S_03096C_VERT_GRP_SIZE(shader->ngg.hw_max_esverts) |
S_03096C_BREAK_WAVE_AT_EOI(break_wave_at_eoi);
+ /* Bug workaround for a possible hang with non-tessellation cases.
+ * Tessellation always sets GE_CNTL.VERT_GRP_SIZE = 0
+ *
+ * Requirement: GE_CNTL.VERT_GRP_SIZE = VGT_GS_ONCHIP_CNTL.ES_VERTS_PER_SUBGRP - 5
+ */
+ if ((sscreen->info.family == CHIP_NAVI10 ||
+ sscreen->info.family == CHIP_NAVI12 ||
+ sscreen->info.family == CHIP_NAVI14) &&
+ (es_type == PIPE_SHADER_VERTEX || gs_type == PIPE_SHADER_VERTEX) && /* = no tess */
+ shader->ngg.hw_max_esverts != 256) {
+ shader->ge_cntl &= C_03096C_VERT_GRP_SIZE;
+
+ if (shader->ngg.hw_max_esverts > 5) {
+ shader->ge_cntl |=
+ S_03096C_VERT_GRP_SIZE(shader->ngg.hw_max_esverts - 5);
+ }
+ }
+
if (window_space) {
shader->ctx_reg.ngg.pa_cl_vte_cntl =
S_028818_VTX_XY_FMT(1) | S_028818_VTX_Z_FMT(1);
S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) |
S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1);
}
-
- shader->ctx_reg.ngg.vgt_reuse_off =
- S_028AB4_REUSE_OFF(sscreen->info.family == CHIP_NAVI10 &&
- sscreen->info.chip_external_rev == 0x1 &&
- es_type == PIPE_SHADER_TESS_EVAL);
}
static void si_emit_shader_vs(struct si_context *sctx)
vgpr_comp_cnt = 0; /* only VertexID is needed for GS-COPY. */
num_user_sgprs = SI_GSCOPY_NUM_USER_SGPR;
} else if (shader->selector->type == PIPE_SHADER_VERTEX) {
- /* VGPR0-3: (VertexID, InstanceID / StepRate0, PrimID, InstanceID)
- * If PrimID is disabled. InstanceID / StepRate1 is loaded instead.
- * StepRate0 is set to 1. so that VGPR3 doesn't have to be loaded.
- */
- vgpr_comp_cnt = enable_prim_id ? 2 : (shader->info.uses_instanceid ? 1 : 0);
+ if (sscreen->info.chip_class >= GFX10) {
+ vgpr_comp_cnt = shader->info.uses_instanceid ? 3 : (enable_prim_id ? 2 : 0);
+ } else {
+ /* VGPR0-3: (VertexID, InstanceID / StepRate0, PrimID, InstanceID)
+ * If PrimID is disabled. InstanceID / StepRate1 is loaded instead.
+ * StepRate0 is set to 1. so that VGPR3 doesn't have to be loaded.
+ */
+ vgpr_comp_cnt = enable_prim_id ? 2 : (shader->info.uses_instanceid ? 1 : 0);
+ }
if (info->properties[TGSI_PROPERTY_VS_BLIT_SGPRS]) {
num_user_sgprs = SI_SGPR_VS_BLIT_DATA +
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, S_00B124_MEM_BASE(va >> 40));
- if (sscreen->info.chip_class >= GFX10)
- si_set_ge_pc_alloc(sscreen, pm4, false);
uint32_t rsrc1 = S_00B128_VGPRS((shader->config.num_vgprs - 1) /
(sscreen->ge_wave_size == 32 ? 8 : 4)) |
static unsigned si_get_alpha_test_func(struct si_context *sctx)
{
/* Alpha-test should be disabled if colorbuffer 0 is integer. */
- if (sctx->queued.named.dsa)
- return sctx->queued.named.dsa->alpha_func;
-
- return PIPE_FUNC_ALWAYS;
+ return sctx->queued.named.dsa->alpha_func;
}
void si_shader_selector_key_vs(struct si_context *sctx,
new_ngg = false;
if (new_ngg != sctx->ngg) {
+ /* Transitioning from NGG to legacy GS requires VGT_FLUSH on Navi10-14.
+ * VGT_FLUSH is also emitted at the beginning of IBs when legacy GS ring
+ * pointers are set.
+ */
+ if ((sctx->family == CHIP_NAVI10 ||
+ sctx->family == CHIP_NAVI12 ||
+ sctx->family == CHIP_NAVI14) &&
+ !new_ngg)
+ sctx->flags |= SI_CONTEXT_VGT_FLUSH;
+
sctx->ngg = new_ngg;
sctx->last_rast_prim = -1; /* reset this so that it gets updated */
return true;
if (sctx->gs_shader.cso)
key.u.gs = 1;
- if (sctx->chip_class >= GFX10) {
- key.u.ngg = sctx->ngg;
-
- if (sctx->gs_shader.cso)
- key.u.streamout = !!sctx->gs_shader.cso->so.num_outputs;
- else if (sctx->tes_shader.cso)
- key.u.streamout = !!sctx->tes_shader.cso->so.num_outputs;
- else
- key.u.streamout = !!sctx->vs_shader.cso->so.num_outputs;
+ if (sctx->ngg) {
+ key.u.ngg = 1;
+ key.u.streamout = !!si_get_vs(sctx)->cso->so.num_outputs;
}
/* Update TCS and TES. */
projects / mesa.git / blobdiff