si_emit_graphics(struct radv_physical_device *physical_device,
struct radeon_cmdbuf *cs)
{
+ bool has_clear_state = physical_device->rad_info.has_clear_state;
int i;
- radeon_emit(cs, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
- radeon_emit(cs, CONTEXT_CONTROL_LOAD_ENABLE(1));
- radeon_emit(cs, CONTEXT_CONTROL_SHADOW_ENABLE(1));
+ /* Since amdgpu version 3.6.0, CONTEXT_CONTROL is emitted by the kernel */
+ if (physical_device->rad_info.drm_minor < 6) {
+ radeon_emit(cs, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
+ radeon_emit(cs, CONTEXT_CONTROL_LOAD_ENABLE(1));
+ radeon_emit(cs, CONTEXT_CONTROL_SHADOW_ENABLE(1));
+ }
- if (physical_device->has_clear_state) {
+ if (has_clear_state) {
radeon_emit(cs, PKT3(PKT3_CLEAR_STATE, 0, 0));
radeon_emit(cs, 0);
}
si_set_raster_config(physical_device, cs);
radeon_set_context_reg(cs, R_028A18_VGT_HOS_MAX_TESS_LEVEL, fui(64));
- if (!physical_device->has_clear_state)
+ if (!has_clear_state)
radeon_set_context_reg(cs, R_028A1C_VGT_HOS_MIN_TESS_LEVEL, fui(0));
/* FIXME calculate these values somehow ??? */
radeon_set_context_reg(cs, R_028A58_VGT_ES_PER_GS, 0x40);
}
- if (!physical_device->has_clear_state) {
+ if (!has_clear_state) {
radeon_set_context_reg(cs, R_028A5C_VGT_GS_PER_VS, 0x2);
radeon_set_context_reg(cs, R_028A8C_VGT_PRIMITIVEID_RESET, 0x0);
radeon_set_context_reg(cs, R_028B98_VGT_STRMOUT_BUFFER_CONFIG, 0x0);
}
- radeon_set_context_reg(cs, R_028AA0_VGT_INSTANCE_STEP_RATE_0, 1);
- if (!physical_device->has_clear_state)
+ if (physical_device->rad_info.chip_class <= GFX9)
+ radeon_set_context_reg(cs, R_028AA0_VGT_INSTANCE_STEP_RATE_0, 1);
+ if (!has_clear_state)
radeon_set_context_reg(cs, R_028AB8_VGT_VTX_CNT_EN, 0x0);
if (physical_device->rad_info.chip_class < GFX7)
radeon_set_config_reg(cs, R_008A14_PA_CL_ENHANCE, S_008A14_NUM_CLIP_SEQ(3) |
S_008A14_CLIP_VTX_REORDER_ENA(1));
- if (!physical_device->has_clear_state)
+ if (!has_clear_state)
radeon_set_context_reg(cs, R_02882C_PA_SU_PRIM_FILTER_CNTL, 0);
/* CLEAR_STATE doesn't clear these correctly on certain generations.
* I don't know why. Deduced by trial and error.
*/
- if (physical_device->rad_info.chip_class <= GFX7 || !physical_device->has_clear_state) {
+ if (physical_device->rad_info.chip_class <= GFX7 || !has_clear_state) {
radeon_set_context_reg(cs, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET, 0);
radeon_set_context_reg(cs, R_028204_PA_SC_WINDOW_SCISSOR_TL,
S_028204_WINDOW_OFFSET_DISABLE(1));
S_028034_BR_X(16384) | S_028034_BR_Y(16384));
}
- if (!physical_device->has_clear_state) {
+ if (!has_clear_state) {
for (i = 0; i < 16; i++) {
radeon_set_context_reg(cs, R_0282D0_PA_SC_VPORT_ZMIN_0 + i*8, 0);
radeon_set_context_reg(cs, R_0282D4_PA_SC_VPORT_ZMAX_0 + i*8, fui(1.0));
}
}
- if (!physical_device->has_clear_state) {
+ if (!has_clear_state) {
radeon_set_context_reg(cs, R_02820C_PA_SC_CLIPRECT_RULE, 0xFFFF);
radeon_set_context_reg(cs, R_028230_PA_SC_EDGERULE, 0xAAAAAAAA);
/* PA_SU_HARDWARE_SCREEN_OFFSET must be 0 due to hw bug on GFX6 */
late_alloc_limit = (num_cu_per_sh - 2) * 4;
}
+ unsigned late_alloc_limit_gs = late_alloc_limit;
unsigned cu_mask_vs = 0xffff;
unsigned cu_mask_gs = 0xffff;
}
}
+ /* Don't use late alloc for NGG on Navi14 due to a hw bug. */
+ if (physical_device->rad_info.family == CHIP_NAVI14) {
+ late_alloc_limit_gs = 0;
+ cu_mask_gs = 0xffff;
+ }
+
radeon_set_sh_reg_idx(physical_device, cs, R_00B118_SPI_SHADER_PGM_RSRC3_VS,
3, S_00B118_CU_EN(cu_mask_vs) |
S_00B118_WAVE_LIMIT(0x3F));
if (physical_device->rad_info.chip_class >= GFX10) {
radeon_set_sh_reg_idx(physical_device, cs, R_00B204_SPI_SHADER_PGM_RSRC4_GS,
3, S_00B204_CU_EN(0xffff) |
- S_00B204_SPI_SHADER_LATE_ALLOC_GS_GFX10(late_alloc_limit));
+ S_00B204_SPI_SHADER_LATE_ALLOC_GS_GFX10(late_alloc_limit_gs));
}
radeon_set_sh_reg_idx(physical_device, cs, R_00B01C_SPI_SHADER_PGM_RSRC3_PS,
radeon_set_context_reg(cs, R_028C50_PA_SC_NGG_MODE_CNTL,
S_028C50_MAX_DEALLOCS_IN_WAVE(512));
radeon_set_context_reg(cs, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
- radeon_set_context_reg(cs, R_02835C_PA_SC_TILE_STEERING_OVERRIDE,
- physical_device->rad_info.pa_sc_tile_steering_override);
radeon_set_context_reg(cs, R_02807C_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_00B0C0_SOFT_GROUPING_EN(1) |
S_00B0C0_NUMBER_OF_REQUESTS_PER_CU(4 - 1));
radeon_set_sh_reg(cs, R_00B1C0_SPI_SHADER_REQ_CTRL_VS, 0);
+
+ if (physical_device->rad_info.family == CHIP_NAVI10 ||
+ physical_device->rad_info.family == CHIP_NAVI12 ||
+ physical_device->rad_info.family == CHIP_NAVI14) {
+ /* SQ_NON_EVENT must be emitted before GE_PC_ALLOC is written. */
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, EVENT_TYPE(V_028A90_SQ_NON_EVENT) | EVENT_INDEX(0));
+ }
+
+ /* TODO: For culling, replace 128 with 256. */
+ radeon_set_uconfig_reg(cs, R_030980_GE_PC_ALLOC,
+ S_030980_OVERSUB_EN(1) |
+ S_030980_NUM_PC_LINES(128 * physical_device->rad_info.max_se - 1));
}
if (physical_device->rad_info.chip_class >= GFX8) {
radeon_set_context_reg(cs, R_028B50_VGT_TESS_DISTRIBUTION,
vgt_tess_distribution);
- } else if (!physical_device->has_clear_state) {
+ } else if (!has_clear_state) {
radeon_set_context_reg(cs, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
radeon_set_context_reg(cs, R_028C5C_VGT_OUT_DEALLOC_CNTL, 16);
}
if (physical_device->rad_info.chip_class >= GFX9) {
- unsigned num_se = physical_device->rad_info.max_se;
- unsigned pc_lines = 0;
- unsigned max_alloc_count = 0;
-
- switch (physical_device->rad_info.family) {
- case CHIP_VEGA10:
- case CHIP_VEGA12:
- case CHIP_VEGA20:
- pc_lines = 4096;
- break;
- case CHIP_RAVEN:
- case CHIP_RAVEN2:
- case CHIP_NAVI10:
- case CHIP_NAVI12:
- pc_lines = 1024;
- break;
- case CHIP_NAVI14:
- pc_lines = 512;
- break;
- default:
- assert(0);
- }
-
- if (physical_device->rad_info.chip_class >= GFX10) {
- max_alloc_count = pc_lines / 3;
- } else {
- max_alloc_count = MIN2(128, pc_lines / (4 * num_se));
- }
-
radeon_set_context_reg(cs, R_028C48_PA_SC_BINNER_CNTL_1,
- S_028C48_MAX_ALLOC_COUNT(max_alloc_count) |
+ S_028C48_MAX_ALLOC_COUNT(physical_device->rad_info.pbb_max_alloc_count - 1) |
S_028C48_MAX_PRIM_PER_BATCH(1023));
radeon_set_context_reg(cs, R_028C4C_PA_SC_CONSERVATIVE_RASTERIZATION_CNTL,
S_028C4C_NULL_SQUAD_AA_MASK_ENABLE(1));
radeon_emit(cs, S_028A04_MIN_SIZE(radv_pack_float_12p4(0)) |
S_028A04_MAX_SIZE(radv_pack_float_12p4(8192/2)));
- if (!physical_device->has_clear_state) {
+ if (!has_clear_state) {
radeon_set_context_reg(cs, R_028004_DB_COUNT_CONTROL,
S_028004_ZPASS_INCREMENT_DISABLE(1));
}
unsigned cb_db_event = 0;
/* We don't need these. */
- assert(!(flush_bits & (RADV_CMD_FLAG_VGT_FLUSH |
- RADV_CMD_FLAG_VGT_STREAMOUT_SYNC)));
+ assert(!(flush_bits & (RADV_CMD_FLAG_VGT_STREAMOUT_SYNC)));
if (flush_bits & RADV_CMD_FLAG_INV_ICACHE)
gcr_cntl |= S_586_GLI_INV(V_586_GLI_ALL);
gcr_cntl |= S_586_GL1_INV(1) | S_586_GLV_INV(1);
if (flush_bits & RADV_CMD_FLAG_INV_L2) {
/* Writeback and invalidate everything in L2. */
- gcr_cntl |= S_586_GL2_INV(1) | S_586_GLM_INV(1);
+ gcr_cntl |= S_586_GL2_INV(1) | S_586_GL2_WB(1) |
+ S_586_GLM_INV(1) | S_586_GLM_WB(1);
} else if (flush_bits & RADV_CMD_FLAG_WB_L2) {
- /* Writeback but do not invalidate. */
- gcr_cntl |= S_586_GL2_WB(1);
+ /* Writeback but do not invalidate.
+ * GLM doesn't support WB alone. If WB is set, INV must be set too.
+ */
+ gcr_cntl |= S_586_GL2_WB(1) |
+ S_586_GLM_WB(1) | S_586_GLM_INV(1);
}
/* TODO: Implement this new flag for GFX9+.
- if (flush_bits & RADV_CMD_FLAG_INV_L2_METADATA)
- gcr_cntl |= S_586_GLM_INV(1);
+ else if (flush_bits & RADV_CMD_FLAG_INV_L2_METADATA)
+ gcr_cntl |= S_586_GLM_INV(1) | S_586_GLM_WB(1);
*/
if (flush_bits & (RADV_CMD_FLAG_FLUSH_AND_INV_CB | RADV_CMD_FLAG_FLUSH_AND_INV_DB)) {
*flush_cnt, 0xffffffff);
}
+ /* VGT state sync */
+ if (flush_bits & RADV_CMD_FLAG_VGT_FLUSH) {
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, EVENT_TYPE(V_028A90_VGT_FLUSH) | EVENT_INDEX(0));
+ }
+
/* Ignore fields that only modify the behavior of other fields. */
if (gcr_cntl & C_586_GL1_RANGE & C_586_GL2_RANGE & C_586_SEQ) {
/* Flush caches and wait for the caches to assert idle.
unsigned dma_flags = 0;
unsigned byte_count = MIN2(size, cp_dma_max_byte_count(cmd_buffer));
+ if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX10) {
+ /* DMA operations via L2 are coherent and faster.
+ * TODO: GFX7-GFX9 should also support this but it
+ * requires tests/benchmarks.
+ */
+ dma_flags |= CP_DMA_USE_L2;
+ }
+
si_cp_dma_prepare(cmd_buffer, byte_count,
size + skipped_size + realign_size,
&dma_flags);
unsigned byte_count = MIN2(size, cp_dma_max_byte_count(cmd_buffer));
unsigned dma_flags = CP_DMA_CLEAR;
+ if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX10) {
+ /* DMA operations via L2 are coherent and faster.
+ * TODO: GFX7-GFX9 should also support this but it
+ * requires tests/benchmarks.
+ */
+ dma_flags |= CP_DMA_USE_L2;
+ }
+
si_cp_dma_prepare(cmd_buffer, byte_count, size, &dma_flags);
/* Emit the clear packet. */
projects / mesa.git / blobdiff