r300/compiler: Implement hardware assisted loops for vertex shaders.

Single loops work, but nested loops do not.

3 files changed, 40 insertions, 1 deletions

diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 7bd43b6eb5d..98958d1a2e0 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -936,6 +936,22 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
             OUT_CS_TABLE(data, 4);
         }
     }
+
+    /* Emit flow control instructions. */
+    if (code->num_fc_ops) {
+
+        OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops);
+        if (r300screen->caps.is_r500) {
+            OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, code->num_fc_ops * 2);
+            OUT_CS_TABLE(code->fc_op_addrs.r500, code->num_fc_ops * 2);
+        } else {
+            OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, code->num_fc_ops);
+            OUT_CS_TABLE(code->fc_op_addrs.r300, code->num_fc_ops);
+        }
+        OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, code->num_fc_ops);
+        OUT_CS_TABLE(code->fc_loop_index, code->num_fc_ops);
+    }
+
     END_CS;
 }
 
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 99a9d650551..60d3b600cb7 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -496,6 +496,12 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R300_VAP_GB_HORZ_CLIP_ADJ                   0x2228
 #define R300_VAP_GB_HORZ_DISC_ADJ                   0x222c
 
+#define R300_VAP_PVS_FLOW_CNTL_ADDRS_0      0x2230
+#define R300_PVS_FC_ACT_ADRS(x)             ((x) << 0)
+#define R300_PVS_FC_LOOP_CNT_JMP_INST(x)    ((x) << 8)
+#define R300_PVS_FC_LAST_INST(x)            ((x) << 16)
+#define R300_PVS_FC_RTN_INST(x)             ((x) << 24)
+
 /* gap */
 
 /* Sometimes, END_OF_PKT and 0x2284=0 are the only commands sent between
@@ -514,6 +520,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_2288_R300                    0x00750000 /* -- nh */
 #       define R300_2288_RV350                   0x0000FFFF /* -- Vladimir */
 
+#define R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0 0x2290
+#define R300_PVS_FC_LOOP_INIT_VAL(x)        ((x) << 0)
+#define R300_PVS_FC_LOOP_STEP_VAL(x)        ((x) << 8)
+
 /* gap */
 
 /* Addresses are relative to the vertex program instruction area of the
@@ -548,6 +558,9 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R300_VAP_PVS_CODE_CNTL_1	    0x22D8
 #       define R300_PVS_LAST_VTX_SRC_INST_SHIFT  0
 #define R300_VAP_PVS_FLOW_CNTL_OPC          0x22DC
+#define R300_VAP_PVS_FC_OPC_JUMP(x)         (1 << (2 * (x)))
+#define R300_VAP_PVS_FC_OPC_LOOP(x)         (2 << (2 * (x)))
+#define R300_VAP_PVS_FC_OPC_JSR(x)          (3 << (2 * (x)))
 
 /* The entire range from 0x2300 to 0x2AC inclusive seems to be used for
  * immediate vertices
@@ -564,6 +577,14 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 /* write 0 to indicate end of packet? */
 #define R300_VAP_VTX_END_OF_PKT             0x24AC
 
+#define R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0   0x2500
+#define R500_PVS_FC_ACT_ADRS(x)             ((x) << 0)
+#define R500_PVS_FC_LOOP_CNT_JMP_INST(x)    ((x) << 16)
+
+#define R500_VAP_PVS_FLOW_CNTL_ADDRS_UW_0   0x2504
+#define R500_PVS_FC_LAST_INST(x)            ((x) << 0)
+#define R500_PVS_FC_RTN_INST(x)             ((x) << 16)
+
 /* gap */
 
 /* These are values from r300_reg/r300_reg.h - they are known to be correct
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 9db5e9e0545..e62a33daebd 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -1758,10 +1758,12 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
     r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
 
     if (r300->screen->caps.has_tcl) {
+        unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2;
         r300->vs_state.dirty = TRUE;
         r300->vs_state.size =
                 vs->code.length + 9 +
-                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0);
+                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0) +
+        (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0);
 
         if (vs->externals_count) {
             r300->vs_constants.dirty = TRUE;