Mesa (main): pan/bi: Shrink live array to 8-bits

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: 7de5293b897080a2a954d54e076b2c1e55f08c53
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=7de5293b897080a2a954d54e076b2c1e55f08c53

Author: Alyssa Rosenzweig <alyssa at collabora.com>
Date:   Fri Jul 16 19:17:53 2021 -0400

pan/bi: Shrink live array to 8-bits

We only actually use 4-bits, so we could shrink again. But this by
itself means 1/2 the memory usage for liveness analysis and 1/2 the
copying/alloc/free.

Signed-off-by: Alyssa Rosenzweig <alyssa at collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11936>

---

 src/panfrost/bifrost/bi_liveness.c | 12 ++++++------
 src/panfrost/bifrost/bi_opt_dce.c  |  2 +-
 src/panfrost/bifrost/bi_ra.c       |  4 ++--
 src/panfrost/bifrost/compiler.h    | 10 ++++------
 4 files changed, 13 insertions(+), 15 deletions(-)

diff --git a/src/panfrost/bifrost/bi_liveness.c b/src/panfrost/bifrost/bi_liveness.c
index 49297081543..6c7a7b5c100 100644
--- a/src/panfrost/bifrost/bi_liveness.c
+++ b/src/panfrost/bifrost/bi_liveness.c
@@ -32,7 +32,7 @@
  * returns whether progress was made. */
 
 void
-bi_liveness_ins_update(uint16_t *live, bi_instr *ins, unsigned max)
+bi_liveness_ins_update(uint8_t *live, bi_instr *ins, unsigned max)
 {
         /* live_in[s] = GEN[s] + (live_out[s] - KILL[s]) */
 
@@ -46,7 +46,7 @@ bi_liveness_ins_update(uint16_t *live, bi_instr *ins, unsigned max)
         bi_foreach_src(ins, src) {
                 unsigned count = bi_count_read_registers(ins, src);
                 unsigned rmask = BITFIELD_MASK(count);
-                uint16_t mask = (rmask << ins->src[src].offset);
+                uint8_t mask = (rmask << ins->src[src].offset);
 
                 unsigned node = bi_get_node(ins->src[src]);
                 if (node < max)
@@ -65,8 +65,8 @@ liveness_block_update(bi_block *blk, unsigned temp_count)
                         blk->live_out[i] |= succ->live_in[i];
         }
 
-        uint16_t *live = ralloc_array(blk, uint16_t, temp_count);
-        memcpy(live, blk->live_out, temp_count * sizeof(uint16_t));
+        uint8_t *live = ralloc_array(blk, uint8_t, temp_count);
+        memcpy(live, blk->live_out, temp_count);
 
         bi_foreach_instr_in_block_rev(blk, ins)
                 bi_liveness_ins_update(live, (bi_instr *) ins, temp_count);
@@ -112,8 +112,8 @@ bi_compute_liveness(bi_context *ctx)
                 if (block->live_out)
                         ralloc_free(block->live_out);
 
-                block->live_in = rzalloc_array(block, uint16_t, temp_count);
-                block->live_out = rzalloc_array(block, uint16_t, temp_count);
+                block->live_in = rzalloc_array(block, uint8_t, temp_count);
+                block->live_out = rzalloc_array(block, uint8_t, temp_count);
         }
 
         /* Initialize the work list with the exit block */
diff --git a/src/panfrost/bifrost/bi_opt_dce.c b/src/panfrost/bifrost/bi_opt_dce.c
index 254cd681c45..141591dc532 100644
--- a/src/panfrost/bifrost/bi_opt_dce.c
+++ b/src/panfrost/bifrost/bi_opt_dce.c
@@ -36,7 +36,7 @@ bi_opt_dead_code_eliminate(bi_context *ctx)
         bi_compute_liveness(ctx);
 
         bi_foreach_block_rev(ctx, block) {
-                uint16_t *live = rzalloc_array(block, uint16_t, temp_count);
+                uint8_t *live = rzalloc_array(block, uint8_t, temp_count);
 
                 bi_foreach_successor(block, succ) {
                         for (unsigned i = 0; i < temp_count; ++i)
diff --git a/src/panfrost/bifrost/bi_ra.c b/src/panfrost/bifrost/bi_ra.c
index f62a6509ecb..135727f112a 100644
--- a/src/panfrost/bifrost/bi_ra.c
+++ b/src/panfrost/bifrost/bi_ra.c
@@ -207,7 +207,7 @@ bi_make_affinity(uint64_t clobber, unsigned count, bool split_file)
 }
 
 static void
-bi_mark_interference(bi_block *block, struct lcra_state *l, uint16_t *live, uint64_t preload_live, unsigned node_count, bool is_blend, bool split_file)
+bi_mark_interference(bi_block *block, struct lcra_state *l, uint8_t *live, uint64_t preload_live, unsigned node_count, bool is_blend, bool split_file)
 {
         bi_foreach_instr_in_block_rev(block, ins) {
                 /* Mark all registers live after the instruction as
@@ -265,7 +265,7 @@ bi_compute_interference(bi_context *ctx, struct lcra_state *l, bool full_regs)
         bi_postra_liveness(ctx);
 
         bi_foreach_block_rev(ctx, blk) {
-                uint16_t *live = mem_dup(blk->live_out, node_count * sizeof(uint16_t));
+                uint8_t *live = mem_dup(blk->live_out, node_count);
 
                 bi_mark_interference(blk, l, live, blk->reg_live_out,
                                 node_count, ctx->inputs->is_blend, !full_regs);
diff --git a/src/panfrost/bifrost/compiler.h b/src/panfrost/bifrost/compiler.h
index 3e1e7e3afb0..85925e6a60c 100644
--- a/src/panfrost/bifrost/compiler.h
+++ b/src/panfrost/bifrost/compiler.h
@@ -537,11 +537,9 @@ typedef struct bi_block {
         struct set *predecessors;
         bool unconditional_jumps;
 
-        /* In liveness analysis, these are live masks (per-component) for
-         * indices for the block. Scalar compilers have the luxury of using
-         * simple bit fields, but for us, liveness is a vector idea. */
-        uint16_t *live_in;
-        uint16_t *live_out;
+        /* Per 32-bit word live masks for the block indexed by node */
+        uint8_t *live_in;
+        uint8_t *live_out;
 
         /* If true, uses clauses; if false, uses instructions */
         bool scheduled;
@@ -842,7 +840,7 @@ int bi_test_packing_formats(void);
 /* Liveness */
 
 void bi_compute_liveness(bi_context *ctx);
-void bi_liveness_ins_update(uint16_t *live, bi_instr *ins, unsigned max);
+void bi_liveness_ins_update(uint8_t *live, bi_instr *ins, unsigned max);
 void bi_invalidate_liveness(bi_context *ctx);
 
 void bi_postra_liveness(bi_context *ctx);