r600g/sb: relax register allocation for compute shaders

We have to assume that all GPRs in compute shader can be indirectly
addressed because LLVM backend doesn't provide any indirect array info.
That's why for compute shaders GPR array is created that covers all used
GPRs (0..r600_bytecode::ngpr-1), but this seriously restricts register
allocation in sb.

This patch checks for actual use of indirect access in the shader and
if it's not used then GPR array is not created, so that regalloc is not
unnecessarily restricted.

Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>

diff --git a/src/gallium/drivers/r600/sb/sb_bc.h b/src/gallium/drivers/r600/sb/sb_bc.h
index 6e73b418d23fc939b67cb716ef2e650fb7dbcbd8..25255a71d32f179f2316be02b27a31d761954861 100644 (file)
--- a/src/gallium/drivers/r600/sb/sb_bc.h
+++ b/src/gallium/drivers/r600/sb/sb_bc.h
@@ -789,13 +789,15 @@ class bc_parser {
        typedef std::stack<region_node*> region_stack;
        region_stack loop_stack;
 
+       bool gpr_reladdr;
+
 public:
 
        bc_parser(sb_context &sctx, r600_bytecode *bc, r600_shader* pshader) :
                ctx(sctx), dec(), bc(bc), pshader(pshader),
                dw(), bc_ndw(), max_cf(),
                sh(), error(), slots(), cgroup(),
-               cf_map(), loop_stack() { }
+               cf_map(), loop_stack(), gpr_reladdr() { }
 
        int decode();
        int prepare();

diff --git a/src/gallium/drivers/r600/sb/sb_bc_parser.cpp b/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
index f8c39ce246bbe292d0893b5b600ba0612a1951fa..3ea09cb73b438e495209e66cfd8be84fcfc5dadc 100644 (file)
--- a/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
+++ b/src/gallium/drivers/r600/sb/sb_bc_parser.cpp
@@ -109,7 +109,8 @@ int bc_parser::prepare() {
 int bc_parser::parse_decls() {
 
        if (!pshader) {
-               sh->add_gpr_array(0, bc->ngpr, 0x0F);
+               if (gpr_reladdr)
+                       sh->add_gpr_array(0, bc->ngpr, 0x0F);
                return 0;
        }
 
@@ -192,8 +193,12 @@ int bc_parser::decode_cf(unsigned &i, bool &eop) {
                if ((r = decode_fetch_clause(cf)))
                        return r;;
        } else if (flags & CF_EXP) {
+               if (cf->bc.rw_rel)
+                       gpr_reladdr = true;
                assert(!cf->bc.rw_rel);
        } else if (flags & (CF_STRM | CF_RAT)) {
+               if (cf->bc.rw_rel)
+                       gpr_reladdr = true;
                assert(!cf->bc.rw_rel);
        } else if (flags & CF_BRANCH) {
                if (cf->bc.addr > max_cf)
@@ -257,8 +262,13 @@ int bc_parser::decode_alu_group(cf_node* cf, unsigned &i, unsigned &gcnt) {
        for (node_iterator I = g->begin(), E = g->end(); I != E; ++I) {
                n = static_cast<alu_node*>(*I);
 
+               if (n->bc.dst_rel)
+                       gpr_reladdr = true;
+
                for (int k = 0; k < n->bc.op_ptr->src_count; ++k) {
                        bc_alu_src &src = n->bc.src[k];
+                       if (src.rel)
+                               gpr_reladdr = true;
                        if (src.sel == ALU_SRC_LITERAL) {
                                literal_mask |= (1 << src.chan);
                                src.value.u = dw[i + src.chan];
@@ -483,6 +493,8 @@ int bc_parser::decode_fetch_clause(cf_node* cf) {
                cf->push_back(n);
                if ((r = dec->decode_fetch(i, n->bc)))
                        return r;
+               if (n->bc.src_rel || n->bc.dst_rel)
+                       gpr_reladdr = true;
 
        }
        return 0;