slab->size = 256 * 1024;
slab->bo = device->ws->buffer_create(device->ws, slab->size, 256,
- RADEON_DOMAIN_VRAM, RADEON_FLAG_NO_INTERPROCESS_SHARING);
+ RADEON_DOMAIN_VRAM,
+ RADEON_FLAG_NO_INTERPROCESS_SHARING |
+ device->physical_device->cpdma_prefetch_writes_memory ?
+ 0 : RADEON_FLAG_READ_ONLY);
slab->ptr = (char*)device->ws->buffer_map(slab->bo);
list_inithead(&slab->shaders);
case MESA_SHADER_COMPUTE: {
struct ac_shader_info *info = &variant->info.info;
variant->rsrc2 |=
- S_00B84C_TGID_X_EN(1) | S_00B84C_TGID_Y_EN(1) |
- S_00B84C_TGID_Z_EN(1) | S_00B84C_TIDIG_COMP_CNT(2) |
+ S_00B84C_TGID_X_EN(info->cs.uses_block_id[0]) |
+ S_00B84C_TGID_Y_EN(info->cs.uses_block_id[1]) |
+ S_00B84C_TGID_Z_EN(info->cs.uses_block_id[2]) |
+ S_00B84C_TIDIG_COMP_CNT(info->cs.uses_thread_id[2] ? 2 :
+ info->cs.uses_thread_id[1] ? 1 : 0) |
S_00B84C_TG_SIZE_EN(info->cs.uses_local_invocation_idx) |
S_00B84C_LDS_SIZE(variant->config.lds_size);
break;
if (device->physical_device->rad_info.chip_class >= GFX9 &&
stage == MESA_SHADER_GEOMETRY) {
+ struct ac_shader_info *info = &variant->info.info;
+ unsigned es_type = variant->info.gs.es_type;
+ unsigned gs_vgpr_comp_cnt;
+
+ /* If offsets 4, 5 are used, GS_VGPR_COMP_CNT is ignored and
+ * VGPR[0:4] are always loaded.
+ */
+ if (info->uses_invocation_id)
+ gs_vgpr_comp_cnt = 3; /* VGPR3 contains InvocationID. */
+ else if (info->uses_prim_id)
+ gs_vgpr_comp_cnt = 2; /* VGPR2 contains PrimitiveID. */
+ else if (variant->info.gs.vertices_in >= 3)
+ gs_vgpr_comp_cnt = 1; /* VGPR1 contains offsets 2, 3 */
+ else
+ gs_vgpr_comp_cnt = 0; /* VGPR0 contains offsets 0, 1 */
+
/* TODO: Figure out how many we actually need. */
- variant->rsrc1 |= S_00B228_GS_VGPR_COMP_CNT(3);
+ variant->rsrc1 |= S_00B228_GS_VGPR_COMP_CNT(gs_vgpr_comp_cnt);
variant->rsrc2 |= S_00B22C_ES_VGPR_COMP_CNT(3) |
- S_00B22C_OC_LDS_EN(1);
+ S_00B22C_OC_LDS_EN(es_type == MESA_SHADER_TESS_EVAL);
} else if (device->physical_device->rad_info.chip_class >= GFX9 &&
stage == MESA_SHADER_TESS_CTRL)
variant->rsrc1 |= S_00B428_LS_VGPR_COMP_CNT(vgpr_comp_cnt);