r600/sb: fall back to un-optimized byte code when ra_init fails

Some emulated fp64 piglit create code that seems to make the register
allocation with sb worse than the original shader created from NIR, so
fall back to using the un-optimized shader.

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8563>

diff --git a/src/gallium/drivers/r600/sb/sb_pass.h b/src/gallium/drivers/r600/sb/sb_pass.h
index a21b0bf..179eab4 100644 (file)
--- a/src/gallium/drivers/r600/sb/sb_pass.h
+++ b/src/gallium/drivers/r600/sb/sb_pass.h
@@ -546,10 +546,10 @@ private:
        void add_prev_chan(unsigned chan);
        unsigned get_preferable_chan_mask();
 
-       void ra_node(container_node *c);
-       void process_op(node *n);
+       bool ra_node(container_node *c);
+       bool process_op(node *n);
 
-       void color(value *v);
+       bool color(value *v);
 
        void color_bs_constraint(ra_constraint *c);
 

diff --git a/src/gallium/drivers/r600/sb/sb_ra_init.cpp b/src/gallium/drivers/r600/sb/sb_ra_init.cpp
index c557b86..e14b187 100644 (file)
--- a/src/gallium/drivers/r600/sb/sb_ra_init.cpp
+++ b/src/gallium/drivers/r600/sb/sb_ra_init.cpp
@@ -313,24 +313,26 @@ int ra_init::run() {
 
        alloc_arrays();
 
-       ra_node(sh.root);
-       return 0;
+       return ra_node(sh.root) ? 0 : 1;
 }
 
-void ra_init::ra_node(container_node* c) {
+bool ra_init::ra_node(container_node* c) {
 
        for (node_iterator I = c->begin(), E = c->end(); I != E; ++I) {
                node *n = *I;
                if (n->type == NT_OP) {
-                       process_op(n);
+                       if (!process_op(n))
+                           return false;
                }
                if (n->is_container() && !n->is_alu_packed()) {
-                       ra_node(static_cast<container_node*>(n));
+                       if (!ra_node(static_cast<container_node*>(n)))
+                           return false;
                }
        }
+        return true;
 }
 
-void ra_init::process_op(node* n) {
+bool ra_init::process_op(node* n) {
 
        bool copy = n->is_copy_mov();
 
@@ -355,7 +357,8 @@ void ra_init::process_op(node* n) {
                for (vvec::iterator I = n->src.begin(), E = n->src.end(); I != E; ++I) {
                        value *v = *I;
                        if (v && v->is_sgpr())
-                               color(v);
+                               if (!color(v))
+                                       return false;
                }
        }
 
@@ -372,10 +375,12 @@ void ra_init::process_op(node* n) {
                                                assign_color(v, s->gpr);
                                        }
                                } else
-                                       color(v);
+                                   if (!color(v))
+                                          return false;
                        }
                }
        }
+        return true;
 }
 
 void ra_init::color_bs_constraint(ra_constraint* c) {
@@ -476,15 +481,15 @@ void ra_init::color_bs_constraint(ra_constraint* c) {
        }
 }
 
-void ra_init::color(value* v) {
+bool ra_init::color(value* v) {
 
        if (v->constraint && v->constraint->kind == CK_PACKED_BS) {
                color_bs_constraint(v->constraint);
-               return;
+               return true;
        }
 
        if (v->chunk && v->chunk->is_fixed())
-               return;
+               return true;
 
        RA_DUMP(
                sblog << "coloring ";
@@ -497,24 +502,24 @@ void ra_init::color(value* v) {
        if (v->is_reg_pinned()) {
                assert(v->is_chan_pinned());
                assign_color(v, v->pin_gpr);
-               return;
+               return true;
        }
 
        regbits rb(sh, v->interferences);
        sel_chan c;
 
        if (v->is_chan_pinned()) {
-               RA_DUMP( sblog << "chan_pinned = " << v->pin_gpr.chan() << "  ";        );
                unsigned mask = 1 << v->pin_gpr.chan();
                c = rb.find_free_chans(mask) + v->pin_gpr.chan();
        } else {
                unsigned cm = get_preferable_chan_mask();
-               RA_DUMP( sblog << "pref chan mask: " << cm << "\n"; );
                c = rb.find_free_chan_by_mask(cm);
-       }
+       }    
 
-       assert(c && c.sel() < 128 - ctx.alu_temp_gprs && "color failed");
+        if (!c || c.sel() >= 128 - ctx.alu_temp_gprs)
+           return false;
        assign_color(v, c);
+        return true;
 }
 
 void ra_init::assign_color(value* v, sel_chan c) {