namespace aco {
+#ifndef NDEBUG
+void perfwarn(Program *program, bool cond, const char *msg, Instruction *instr)
+{
+ if (cond) {
+ char *out;
+ size_t outsize;
+ FILE *memf = open_memstream(&out, &outsize);
+
+ fprintf(memf, "%s: ", msg);
+ aco_print_instr(instr, memf);
+ fclose(memf);
+
+ aco_perfwarn(program, out);
+ free(out);
+
+ if (debug_flags & DEBUG_PERFWARN)
+ exit(1);
+ }
+}
+#endif
+
/**
* The optimizer works in 4 phases:
* (1) The first pass collects information for each ssa-def,
ASSERTED bool all_const = false;
for (Operand& op : instr->operands)
all_const = all_const && (!op.isTemp() || ctx.info[op.tempId()].is_constant_or_literal(32));
- perfwarn(all_const, "All instruction operands are constant", instr.get());
+ perfwarn(ctx.program, all_const, "All instruction operands are constant", instr.get());
}
for (unsigned i = 0; i < instr->operands.size(); i++)
unsigned bits = get_operand_size(instr, i);
if (info.is_constant(bits) && alu_can_accept_constant(instr->opcode, i)) {
Operand op = get_constant_op(ctx, info, bits);
- perfwarn(instr->opcode == aco_opcode::v_cndmask_b32 && i == 2, "v_cndmask_b32 with a constant selector", instr.get());
+ perfwarn(ctx.program, instr->opcode == aco_opcode::v_cndmask_b32 && i == 2, "v_cndmask_b32 with a constant selector", instr.get());
if (i == 0 || instr->opcode == aco_opcode::v_readlane_b32 || instr->opcode == aco_opcode::v_writelane_b32) {
instr->operands[i] = op;
continue;
instr->opcode == aco_opcode::v_sub_f16 ||
instr->opcode == aco_opcode::v_subrev_f16;
if (mad16 || mad32) {
- bool need_fma = mad32 ? block.fp_mode.denorm32 != 0 :
+ bool need_fma = mad32 ? (block.fp_mode.denorm32 != 0 || ctx.program->chip_class >= GFX10_3) :
(block.fp_mode.denorm16_64 != 0 || ctx.program->chip_class >= GFX10);
if (need_fma && instr->definitions[0].isPrecise())
return;