On 12/18/2013 12:48 PM, Tom Musta wrote: > + tcg_gen_div_i64(cpu_gpr[rD(ctx->opcode)], ra, rb); > \ > + tmp0 = tcg_temp_local_new(); > \ > + /* does the result fit in 32 bits? */ > \ > + tcg_gen_ext32s_i64(tmp0, cpu_gpr[rD(ctx->opcode)]); > \ > + tcg_gen_brcond_i64(TCG_COND_NE, cpu_gpr[rD(ctx->opcode)], tmp0, > \ > + lbl_ov); > \ > + tcg_temp_free(tmp0); > \
tmp0 does not need to be a local temp, as it's not used across basic blocks. I do wonder if this wouldn't be better as a helper, what with all of the branches. TCG essentially won't optimize any of this. r~
