v3d: Fold comparisons for IF conditions into the flags for the IF.

[mesa.git] / src / broadcom / compiler / nir_to_vir.c

diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index ac3e103353675cb627376b2f35517346b8e1f4fb..9d20853061252b57bb8430ab83fa96b484a9f5a7 100644 (file)
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -559,12 +559,28 @@ ntq_emit_comparison(struct v3d_compile *c,
         return true;
 }
 
+/* Finds an ALU instruction that generates our src value that could
+ * (potentially) be greedily emitted in the consuming instruction.
+ */
 static struct nir_alu_instr *
 ntq_get_alu_parent(nir_src src)
 {
         if (!src.is_ssa || src.ssa->parent_instr->type != nir_instr_type_alu)
                 return NULL;
-        return nir_instr_as_alu(src.ssa->parent_instr);
+        nir_alu_instr *instr = nir_instr_as_alu(src.ssa->parent_instr);
+        if (!instr)
+                return NULL;
+
+        /* If the ALU instr's srcs are non-SSA, then we would have to avoid
+         * moving emission of the ALU instr down past another write of the
+         * src.
+         */
+        for (int i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
+                if (!instr->src[i].src.is_ssa)
+                        return NULL;
+        }
+
+        return instr;
 }
 
 /**
@@ -1720,19 +1736,33 @@ ntq_emit_if(struct v3d_compile *c, nir_if *if_stmt)
                 was_top_level = true;
         }
 
-        /* Set A for executing (execute == 0) and jumping (if->condition ==
-         * 0) channels, and then update execute flags for those to point to
-         * the ELSE block.
-         *
-         * XXX perf: we could reuse ntq_emit_comparison() to generate our if
-         * condition, and the .uf field to ignore non-executing channels, to
-         * reduce the overhead of if statements.
+        /* Set up the flags for the IF condition (taking the THEN branch). */
+        nir_alu_instr *if_condition_alu = ntq_get_alu_parent(if_stmt->condition);
+        enum v3d_qpu_cond cond;
+        if (!if_condition_alu ||
+            !ntq_emit_comparison(c, if_condition_alu, &cond)) {
+                vir_PF(c, ntq_get_src(c, if_stmt->condition, 0),
+                       V3D_QPU_PF_PUSHZ);
+                cond = V3D_QPU_COND_IFNA;
+        }
+
+        /* Update the flags+cond to mean "Taking the ELSE branch (!cond) and
+         * was previously active (execute Z) for updating the exec flags.
          */
-        vir_PF(c, vir_OR(c,
-                         c->execute,
-                         ntq_get_src(c, if_stmt->condition, 0)),
-                V3D_QPU_PF_PUSHZ);
-        vir_MOV_cond(c, V3D_QPU_COND_IFA,
+        if (was_top_level) {
+                cond = v3d_qpu_cond_invert(cond);
+        } else {
+                struct qinst *inst = vir_MOV_dest(c, vir_reg(QFILE_NULL, 0),
+                                                  c->execute);
+                if (cond == V3D_QPU_COND_IFA) {
+                        vir_set_uf(inst, V3D_QPU_UF_NORNZ);
+                } else {
+                        vir_set_uf(inst, V3D_QPU_UF_ANDZ);
+                        cond = V3D_QPU_COND_IFA;
+                }
+        }
+
+        vir_MOV_cond(c, cond,
                      c->execute,
                      vir_uniform_ui(c, else_block->index));