vc4: Add support for ARB_draw_elements_base_vertex.

[mesa.git] / src / gallium / drivers / vc4 / vc4_register_allocate.c

diff --git a/src/gallium/drivers/vc4/vc4_register_allocate.c b/src/gallium/drivers/vc4/vc4_register_allocate.c
index 9eae7fca758f5bc7cd001e10ee36c7111e90eb15..73964b48dca0f0ab40788ad6fcf2a96fa2cec1ac 100644 (file)
--- a/src/gallium/drivers/vc4/vc4_register_allocate.c
+++ b/src/gallium/drivers/vc4/vc4_register_allocate.c
@@ -117,10 +117,10 @@ vc4_alloc_reg_set(struct vc4_context *vc4)
 
         vc4->reg_class_any = ra_alloc_reg_class(vc4->regs);
         for (uint32_t i = 0; i < ARRAY_SIZE(vc4_regs); i++) {
-                /* Reserve rb31 for spilling fixup_raddr_conflict() in
+                /* Reserve ra31/rb31 for spilling fixup_raddr_conflict() in
                  * vc4_qpu_emit.c
                  */
-                if (vc4_regs[i].mux == QPU_MUX_B && vc4_regs[i].addr == 31)
+                if (vc4_regs[i].addr == 31)
                         continue;
 
                 /* R4 can't be written as a general purpose register. (it's
@@ -161,7 +161,6 @@ node_to_temp_priority(const void *in_a, const void *in_b)
 struct qpu_reg *
 vc4_register_allocate(struct vc4_context *vc4, struct vc4_compile *c)
 {
-        struct simple_node *node;
         struct node_to_temp_map map[c->num_temps];
         uint32_t temp_to_node[c->num_temps];
         uint32_t def[c->num_temps];
@@ -189,9 +188,7 @@ vc4_register_allocate(struct vc4_context *vc4, struct vc4_compile *c)
         /* Compute the live ranges so we can figure out interference.
          */
         uint32_t ip = 0;
-        foreach(node, &c->instructions) {
-                struct qinst *inst = (struct qinst *)node;
-
+        list_for_each_entry(struct qinst, inst, &c->instructions, link) {
                 if (inst->dst.file == QFILE_TEMP) {
                         def[inst->dst.index] = ip;
                         use[inst->dst.index] = ip;
@@ -227,9 +224,7 @@ vc4_register_allocate(struct vc4_context *vc4, struct vc4_compile *c)
         }
 
         /* Figure out our register classes and preallocated registers*/
-        foreach(node, &c->instructions) {
-                struct qinst *inst = (struct qinst *)node;
-
+        list_for_each_entry(struct qinst, inst, &c->instructions, link) {
                 switch (inst->op) {
                 case QOP_FRAG_Z:
                         ra_set_node_reg(g, temp_to_node[inst->dst.index],
@@ -254,26 +249,14 @@ vc4_register_allocate(struct vc4_context *vc4, struct vc4_compile *c)
                                           vc4->reg_class_a);
                         break;
 
-                case QOP_UNPACK_8A_F:
-                case QOP_UNPACK_8B_F:
-                case QOP_UNPACK_8C_F:
-                case QOP_UNPACK_8D_F:
-                case QOP_UNPACK_16A_F:
-                case QOP_UNPACK_16B_F:
-                case QOP_UNPACK_8A_I:
-                case QOP_UNPACK_8B_I:
-                case QOP_UNPACK_8C_I:
-                case QOP_UNPACK_8D_I:
-                case QOP_UNPACK_16A_I:
-                case QOP_UNPACK_16B_I:
-                        /* The unpack flags require an A-file src register. */
-                        ra_set_node_class(g, temp_to_node[inst->src[0].index],
-                                          vc4->reg_class_a);
-                        break;
-
                 default:
                         break;
                 }
+
+                if (qir_src_needs_a_file(inst)) {
+                        ra_set_node_class(g, temp_to_node[inst->src[0].index],
+                                          vc4->reg_class_a);
+                }
         }
 
         for (uint32_t i = 0; i < c->num_temps; i++) {
@@ -287,7 +270,11 @@ vc4_register_allocate(struct vc4_context *vc4, struct vc4_compile *c)
         }
 
         bool ok = ra_allocate(g);
-        assert(ok);
+        if (!ok) {
+                fprintf(stderr, "Failed to register allocate:\n");
+                qir_dump(c);
+                abort();
+        }
 
         for (uint32_t i = 0; i < c->num_temps; i++) {
                 temp_registers[i] = vc4_regs[ra_get_node_reg(g, temp_to_node[i])];