-
-static void
-brw_upload_cs_state(struct brw_context *brw)
-{
- if (!brw->cs.prog_data)
- return;
-
- uint32_t offset;
- uint32_t *desc = (uint32_t*) brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
- 8 * 4, 64, &offset);
- struct brw_stage_state *stage_state = &brw->cs.base;
- struct brw_cs_prog_data *cs_prog_data = brw->cs.prog_data;
- struct brw_stage_prog_data *prog_data = &cs_prog_data->base;
-
- if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
- brw->vtbl.emit_buffer_surface_state(
- brw, &stage_state->surf_offset[
- prog_data->binding_table.shader_time_start],
- brw->shader_time.bo, 0, BRW_SURFACEFORMAT_RAW,
- brw->shader_time.bo->size, 1, true);
- }
-
- uint32_t *bind = (uint32_t*) brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
- prog_data->binding_table.size_bytes,
- 32, &stage_state->bind_bo_offset);
-
- uint32_t dwords = brw->gen < 8 ? 8 : 9;
- BEGIN_BATCH(dwords);
- OUT_BATCH(MEDIA_VFE_STATE << 16 | (dwords - 2));
-
- if (prog_data->total_scratch) {
- if (brw->gen >= 8)
- OUT_RELOC64(stage_state->scratch_bo,
- I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
- ffs(prog_data->total_scratch) - 11);
- else
- OUT_RELOC(stage_state->scratch_bo,
- I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
- ffs(prog_data->total_scratch) - 11);
- } else {
- OUT_BATCH(0);
- if (brw->gen >= 8)
- OUT_BATCH(0);
- }
-
- const uint32_t vfe_num_urb_entries = brw->gen >= 8 ? 2 : 0;
- const uint32_t vfe_gpgpu_mode =
- brw->gen == 7 ? SET_FIELD(1, GEN7_MEDIA_VFE_STATE_GPGPU_MODE) : 0;
- OUT_BATCH(SET_FIELD(brw->max_cs_threads - 1, MEDIA_VFE_STATE_MAX_THREADS) |
- SET_FIELD(vfe_num_urb_entries, MEDIA_VFE_STATE_URB_ENTRIES) |
- SET_FIELD(1, MEDIA_VFE_STATE_RESET_GTW_TIMER) |
- SET_FIELD(1, MEDIA_VFE_STATE_BYPASS_GTW) |
- vfe_gpgpu_mode);
-
- OUT_BATCH(0);
- const uint32_t vfe_urb_allocation = brw->gen >= 8 ? 2 : 0;
-
- /* We are uploading duplicated copies of push constant uniforms for each
- * thread. Although the local id data needs to vary per thread, it won't
- * change for other uniform data. Unfortunately this duplication is
- * required for gen7. As of Haswell, this duplication can be avoided, but
- * this older mechanism with duplicated data continues to work.
- *
- * FINISHME: As of Haswell, we could make use of the
- * INTERFACE_DESCRIPTOR_DATA "Cross-Thread Constant Data Read Length" field
- * to only store one copy of uniform data.
- *
- * FINISHME: Broadwell adds a new alternative "Indirect Payload Storage"
- * which is described in the GPGPU_WALKER command and in the Broadwell PRM
- * Volume 7: 3D Media GPGPU, under Media GPGPU Pipeline => Mode of
- * Operations => GPGPU Mode => Indirect Payload Storage.
- *
- * Note: The constant data is built in brw_upload_cs_push_constants below.
- */
- const uint32_t vfe_curbe_allocation =
- ALIGN(cs_prog_data->push.per_thread.regs * cs_prog_data->threads +
- cs_prog_data->push.cross_thread.regs, 2);
- OUT_BATCH(SET_FIELD(vfe_urb_allocation, MEDIA_VFE_STATE_URB_ALLOC) |
- SET_FIELD(vfe_curbe_allocation, MEDIA_VFE_STATE_CURBE_ALLOC));
- OUT_BATCH(0);
- OUT_BATCH(0);
- OUT_BATCH(0);
- ADVANCE_BATCH();
-
- if (cs_prog_data->push.total.size > 0) {
- BEGIN_BATCH(4);
- OUT_BATCH(MEDIA_CURBE_LOAD << 16 | (4 - 2));
- OUT_BATCH(0);
- OUT_BATCH(ALIGN(cs_prog_data->push.total.size, 64));
- OUT_BATCH(stage_state->push_const_offset);
- ADVANCE_BATCH();
- }
-
- /* BRW_NEW_SURFACES and BRW_NEW_*_CONSTBUF */
- memcpy(bind, stage_state->surf_offset,
- prog_data->binding_table.size_bytes);
-
- memset(desc, 0, 8 * 4);
-
- int dw = 0;
- desc[dw++] = brw->cs.base.prog_offset;
- if (brw->gen >= 8)
- desc[dw++] = 0; /* Kernel Start Pointer High */
- desc[dw++] = 0;
- desc[dw++] = stage_state->sampler_offset |
- ((stage_state->sampler_count + 3) / 4);
- desc[dw++] = stage_state->bind_bo_offset;
- desc[dw++] = SET_FIELD(cs_prog_data->push.per_thread.regs,
- MEDIA_CURBE_READ_LENGTH);
- const uint32_t media_threads =
- brw->gen >= 8 ?
- SET_FIELD(cs_prog_data->threads, GEN8_MEDIA_GPGPU_THREAD_COUNT) :
- SET_FIELD(cs_prog_data->threads, MEDIA_GPGPU_THREAD_COUNT);
- assert(cs_prog_data->threads <= brw->max_cs_threads);
-
- assert(prog_data->total_shared <= 64 * 1024);
- uint32_t slm_size = 0;
- if (prog_data->total_shared > 0) {
- /* slm_size is in 4k increments, but must be a power of 2. */
- slm_size = 4 * 1024;
- while (slm_size < prog_data->total_shared)
- slm_size <<= 1;
- slm_size /= 4 * 1024;
- }
-
- desc[dw++] =
- SET_FIELD(cs_prog_data->uses_barrier, MEDIA_BARRIER_ENABLE) |
- SET_FIELD(slm_size, MEDIA_SHARED_LOCAL_MEMORY_SIZE) |
- media_threads;
-
- desc[dw++] =
- SET_FIELD(cs_prog_data->push.cross_thread.regs, CROSS_THREAD_READ_LENGTH);
-
- BEGIN_BATCH(4);
- OUT_BATCH(MEDIA_INTERFACE_DESCRIPTOR_LOAD << 16 | (4 - 2));
- OUT_BATCH(0);
- OUT_BATCH(8 * 4);
- OUT_BATCH(offset);
- ADVANCE_BATCH();
-}
-
-const struct brw_tracked_state brw_cs_state = {
- .dirty = {
- .mesa = _NEW_PROGRAM_CONSTANTS,
- .brw = BRW_NEW_BATCH |
- BRW_NEW_BLORP |
- BRW_NEW_CS_PROG_DATA |
- BRW_NEW_PUSH_CONSTANT_ALLOCATION |
- BRW_NEW_SAMPLER_STATE_TABLE |
- BRW_NEW_SURFACES,
- },
- .emit = brw_upload_cs_state
-};
-