pan/midgard: Implement SIMD-aware dead code elimination
We would like to eliminate not just entire dead instructions, but also dead components, which increases scheduler flexibility (since some vector instructions can become scalar after eliminating dead components). This also will allow better RA in the future. Results are meh. total instructions in shared programs: 3453 -> 3451 (-0.06%) instructions in affected programs: 60 -> 58 (-3.33%) helped: 2 HURT: 0 total bundles in shared programs: 1826 -> 1824 (-0.11%) bundles in affected programs: 33 -> 31 (-6.06%) helped: 2 HURT: 0 total quadwords in shared programs: 3144 -> 3144 (0.00%) quadwords in affected programs: 0 -> 0 helped: 0 HURT: 0 total registers in shared programs: 321 -> 321 (0.00%) registers in affected programs: 45 -> 45 (0.00%) helped: 11 HURT: 11 helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1 helped stats (rel) min: 16.67% max: 50.00% x̄: 39.70% x̃: 50.00% HURT stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1 HURT stats (rel) min: 100.00% max: 100.00% x̄: 100.00% x̃: 100.00% 95% mean confidence interval for registers value: -0.45 0.45 95% mean confidence interval for registers %-change: -1.87% 62.18% Inconclusive result (value mean confidence interval includes 0). total threads in shared programs: 445 -> 447 (0.45%) threads in affected programs: 2 -> 4 (100.00%) helped: 1 HURT: 0 Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
This commit is contained in:
@@ -23,8 +23,41 @@
|
||||
*/
|
||||
|
||||
#include "compiler.h"
|
||||
#include "util/u_memory.h"
|
||||
#include "midgard_ops.h"
|
||||
|
||||
/* Basic dead code elimination on the MIR itself */
|
||||
/* SIMD-aware dead code elimination. Perform liveness analysis step-by-step,
|
||||
* removing dead components. If an instruction ends up with a zero mask, the
|
||||
* instruction in total is dead and should be removed. */
|
||||
|
||||
static bool
|
||||
can_cull_mask(compiler_context *ctx, midgard_instruction *ins)
|
||||
{
|
||||
if (ins->dest >= ctx->temp_count)
|
||||
return false;
|
||||
|
||||
if (ins->type == TAG_LOAD_STORE_4)
|
||||
if (load_store_opcode_props[ins->load_store.op].props & LDST_SPECIAL_MASK)
|
||||
return false;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
static bool
|
||||
can_dce(midgard_instruction *ins)
|
||||
{
|
||||
if (ins->mask)
|
||||
return false;
|
||||
|
||||
if (ins->compact_branch)
|
||||
return false;
|
||||
|
||||
if (ins->type == TAG_LOAD_STORE_4)
|
||||
if (load_store_opcode_props[ins->load_store.op].props & LDST_SIDE_FX)
|
||||
return false;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool
|
||||
midgard_opt_dead_code_eliminate(compiler_context *ctx, midgard_block *block)
|
||||
@@ -32,18 +65,34 @@ midgard_opt_dead_code_eliminate(compiler_context *ctx, midgard_block *block)
|
||||
bool progress = false;
|
||||
|
||||
mir_invalidate_liveness(ctx);
|
||||
mir_compute_liveness(ctx);
|
||||
|
||||
uint16_t *live = mem_dup(block->live_out, ctx->temp_count * sizeof(uint16_t));
|
||||
|
||||
mir_foreach_instr_in_block_rev(block, ins) {
|
||||
if (can_cull_mask(ctx, ins)) {
|
||||
midgard_reg_mode mode = mir_typesize(ins);
|
||||
unsigned oldmask = ins->mask;
|
||||
|
||||
unsigned rounded = mir_round_bytemask_down(live[ins->dest], mode);
|
||||
unsigned cmask = mir_from_bytemask(rounded, mode);
|
||||
|
||||
ins->mask &= cmask;
|
||||
progress |= (ins->mask != oldmask);
|
||||
}
|
||||
|
||||
mir_liveness_ins_update(live, ins, ctx->temp_count);
|
||||
}
|
||||
|
||||
mir_foreach_instr_in_block_safe(block, ins) {
|
||||
if (ins->type != TAG_ALU_4) continue;
|
||||
if (ins->compact_branch) continue;
|
||||
|
||||
if (ins->dest >= SSA_FIXED_MINIMUM) continue;
|
||||
if (mir_is_live_after(ctx, block, ins, ins->dest)) continue;
|
||||
|
||||
mir_remove_instruction(ins);
|
||||
progress = true;
|
||||
if (can_dce(ins)) {
|
||||
mir_remove_instruction(ins);
|
||||
progress = true;
|
||||
}
|
||||
}
|
||||
|
||||
free(live);
|
||||
|
||||
return progress;
|
||||
}
|
||||
|
||||
|
Reference in New Issue
Block a user