pan: Use bitset instead of bool array in bi_find_loop_blocks

Reviewed-by: Eric R. Smith <eric.smith@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38393>

src/panfrost/compiler/bi_ra.c

@@ -1149,25 +1149,26 @@ op_is_load_store(enum bi_opcode op)
 static uint64_t
 compute_spill_cost(bi_context *ctx)
 {
+   void *mctx = ralloc_context(NULL);
+
    /* Required for finding blocks belonging to loops. */
    bi_calc_dominance(ctx);
 
    /* The cost of a spill/fill is just 10*block_depth for now. */
 
-   uint32_t *block_depth = calloc(ctx->num_blocks, sizeof(uint32_t));
+   uint32_t *block_depth = rzalloc_array(mctx, uint32_t, ctx->num_blocks);
+   BITSET_WORD *loop_block = BITSET_RZALLOC(mctx, ctx->num_blocks);
 
    bi_foreach_block(ctx, block) {
       if (!block->loop_header)
          continue;
 
-      bool *loop_block = bi_find_loop_blocks(ctx, block);
+      bi_find_loop_blocks(ctx, block, loop_block);
 
       for (uint32_t b = 0; b < ctx->num_blocks; ++b) {
-         if (loop_block[b])
+         if (BITSET_SET(loop_block, b))
             block_depth[b] += 1;
       }
-
-      free(loop_block);
    }
 
    uint64_t cost = 0;
@@ -1178,7 +1179,7 @@ compute_spill_cost(bi_context *ctx)
       }
    }
 
-   free(block_depth);
+   ralloc_free(mctx);
 
    return cost;
 }

src/panfrost/compiler/bi_spill_ssa.c

@@ -767,10 +767,11 @@ compute_w_entry_loop_header(struct spill_ctx *ctx)
    uint32_t n_ca = 0;
 
    uint32_t max_loop_pressure = 0;
-   const bool *loop_block = bi_find_loop_blocks(ctx->shader, ctx->block);
+   BITSET_WORD *loop_block = BITSET_RZALLOC(NULL, ctx->shader->num_blocks);
+   bi_find_loop_blocks(ctx->shader, ctx->block, loop_block);
 
    bi_foreach_block(ctx->shader, block) {
-      if (loop_block[block->index]) {
+      if (BITSET_TEST(loop_block, block->index)) {
          bi_foreach_instr_in_block(block, I) {
             max_loop_pressure = MAX2(max_loop_pressure, block->ssa_max_live);
 
@@ -790,6 +791,8 @@ compute_w_entry_loop_header(struct spill_ctx *ctx)
       }
    }
 
+   ralloc_free(loop_block);
+
    /* Sort by next-use distance. */
    util_qsort_r(candidates, n_ca, sizeof(struct candidate), cmp_dist, ctx);
 
@@ -833,7 +836,6 @@ compute_w_entry_loop_header(struct spill_ctx *ctx)
 
    assert(ctx->nW <= ctx->k && "invariant");
 
-   free((void *)loop_block);
    free(candidates);
    free(flag_mem);
 }

src/panfrost/compiler/bifrost_compile.c

@@ -6914,37 +6914,44 @@ bifrost_compile_shader_nir(nir_shader *nir,
 }
 
 static void
-find_all_predecessors(const bi_context *ctx, bi_block *b, bool *out)
+find_all_predecessors(const bi_context *ctx, bi_block *b, BITSET_WORD *out)
 {
    assert(out);
    assert(b);
 
+   BITSET_CLEAR_RANGE(out, 0, ctx->num_blocks);
+
    /* If the CFG was one long chain, we would require |blocks|-1 iters to
     * propagate the in_loop info all the way through.
     */
    for (uint32_t iter = 0; iter < ctx->num_blocks - 1; ++iter) {
       bi_foreach_block(ctx, block) {
          bi_foreach_successor(block, succ) {
-            out[block->index] |= out[succ->index] || succ->index == b->index;
+            if (succ == b || BITSET_TEST(out, succ->index)) {
+               BITSET_SET(out, block->index);
+               break;
+            }
          }
       }
    }
 }
 
-bool *
-bi_find_loop_blocks(const bi_context *ctx, bi_block *header)
+void
+bi_find_loop_blocks(const bi_context *ctx, bi_block *header, BITSET_WORD *out)
 {
-   bool *h_as_suc = (bool *)calloc(ctx->num_blocks, sizeof(bool));
-   assert(h_as_suc);
+   find_all_predecessors(ctx, header, out);
 
-   find_all_predecessors(ctx, header, h_as_suc);
+   BITSET_WORD *dominators = BITSET_RZALLOC(NULL, ctx->num_blocks);
+   bi_foreach_block(ctx, b) {
+      if (bi_block_dominates(header, b)) {
+         BITSET_SET(dominators, b->index);
+      }
+   }
 
    /* If the header dominates b and is also the successor of b, then b
     * is in the loop of that header.
     */
-   bi_foreach_block(ctx, b) {
-      h_as_suc[b->index] &= bi_block_dominates(header, b);
-   }
+   __bitset_and(out, out, dominators, BITSET_WORDS(ctx->num_blocks));
 
-   return h_as_suc;
+   ralloc_free(dominators);
 }

src/panfrost/compiler/compiler.h

@@ -1769,7 +1769,8 @@ bi_record_use(bi_instr **uses, BITSET_WORD *multiple, bi_instr *I, unsigned s)
 
 bool bi_lower_divergent_indirects(nir_shader *shader, unsigned lanes);
 
-bool *bi_find_loop_blocks(const bi_context *ctx, bi_block *header);
+void bi_find_loop_blocks(const bi_context *ctx, bi_block *header,
+                         BITSET_WORD *out);
 
 #ifdef __cplusplus
 } /* extern C */