{
for (int i = 0; i < state->nbo; ++i)
r600_resource_reference(&state->bo[i], NULL);
+ r600_resource_reference(&state->indirect_buffer, NULL);
FREE(state);
}
state->bo_usage[i], state->bo_priority[i]);
}
- radeon_emit_array(cs, state->pm4, state->ndw);
+ if (!state->indirect_buffer) {
+ radeon_emit_array(cs, state->pm4, state->ndw);
+ } else {
+ struct r600_resource *ib = state->indirect_buffer;
+
+ radeon_add_to_buffer_list(&sctx->b, &sctx->b.rings.gfx, ib,
+ RADEON_USAGE_READ, RADEON_PRIO_MIN);
+
+ radeon_emit(cs, PKT3(PKT3_INDIRECT_BUFFER_CIK, 2, 0));
+ radeon_emit(cs, ib->gpu_address);
+ radeon_emit(cs, (ib->gpu_address >> 32) & 0xffff);
+ radeon_emit(cs, (ib->b.b.width0 >> 2) & 0xfffff);
+ }
}
void si_pm4_emit_dirty(struct si_context *sctx)
{
memset(&sctx->emitted, 0, sizeof(sctx->emitted));
}
+
+void si_pm4_upload_indirect_buffer(struct si_context *sctx,
+ struct si_pm4_state *state)
+{
+ struct pipe_screen *screen = sctx->b.b.screen;
+ unsigned aligned_ndw = align(state->ndw, 8);
+
+ /* only supported on CIK and later */
+ if (sctx->b.chip_class < CIK)
+ return;
+
+ assert(state->ndw);
+ assert(aligned_ndw <= SI_PM4_MAX_DW);
+
+ r600_resource_reference(&state->indirect_buffer, NULL);
+ state->indirect_buffer = (struct r600_resource*)
+ pipe_buffer_create(screen, PIPE_BIND_CUSTOM,
+ PIPE_USAGE_DEFAULT, aligned_ndw * 4);
+ if (!state->indirect_buffer)
+ return;
+
+ /* Pad the IB to 8 DWs to meet CP fetch alignment requirements. */
+ if (sctx->screen->b.info.gfx_ib_pad_with_type2) {
+ for (int i = state->ndw; i < aligned_ndw; i++)
+ state->pm4[i] = 0x80000000; /* type2 nop packet */
+ } else {
+ for (int i = state->ndw; i < aligned_ndw; i++)
+ state->pm4[i] = 0xffff1000; /* type3 nop packet */
+ }
+
+ pipe_buffer_write(&sctx->b.b, &state->indirect_buffer->b.b,
+ 0, aligned_ndw *4, state->pm4);
+}
struct si_pm4_state
{
+ /* optional indirect buffer */
+ struct r600_resource *indirect_buffer;
+
/* PKT3_SET_*_REG handling */
unsigned last_opcode;
unsigned last_reg;
struct r600_resource *bo,
enum radeon_bo_usage usage,
enum radeon_bo_priority priority);
+void si_pm4_upload_indirect_buffer(struct si_context *sctx,
+ struct si_pm4_state *state);
void si_pm4_free_state_simple(struct si_pm4_state *state);
void si_pm4_free_state(struct si_context *sctx,
#define PKT3_DRAW_INDEX_IMMD 0x2E /* not on CIK */
#define PKT3_NUM_INSTANCES 0x2F
#define PKT3_DRAW_INDEX_MULTI_AUTO 0x30
-#define PKT3_INDIRECT_BUFFER 0x32
+#define PKT3_INDIRECT_BUFFER_SI 0x32 /* not on CIK */
#define PKT3_STRMOUT_BUFFER_UPDATE 0x34
#define PKT3_DRAW_INDEX_OFFSET_2 0x35
#define PKT3_DRAW_PREAMBLE 0x36 /* new on CIK, required on GFX7.2 and later */
#define PKT3_WAIT_REG_MEM 0x3C
#define WAIT_REG_MEM_EQUAL 3
#define PKT3_MEM_WRITE 0x3D /* not on CIK */
+#define PKT3_INDIRECT_BUFFER_CIK 0x3F /* new on CIK */
#define PKT3_COPY_DATA 0x40
#define COPY_DATA_SRC_SEL(x) ((x) & 0xf)
#define COPY_DATA_REG 0