On 07/30/2018 12:12 PM, Aleksandar Markovic wrote:
+static void gen_pool32axf_2_nanomips_insn(DisasContext *ctx, uint32_t opc,
> + int ret, int v1, int v2)
> +{
> + TCGv t0;
> + TCGv t1;
> +
> + TCGv v0_t;
> + TCGv v1_t;
> +
> + t0 = tcg_temp_new();
> + t1 = tcg_temp_new();
> +
> + v0_t = tcg_temp_new();
> + v1_t = tcg_temp_new();
> +
> + gen_load_gpr(v0_t, ret);
> + gen_load_gpr(v1_t, v1);
> +
> + switch (opc) {
> + case NM_POOL32AXF_2_0_7:
> + switch (extract32(ctx->opcode, 9, 3)) {
> + case NM_DPA_W_PH:
> + case NM_DPAQ_S_W_PH:
> + case NM_DPS_W_PH:
> + case NM_DPSQ_S_W_PH:
> + gen_pool32axf_2_multiply(ctx, opc, ret, v1, v2);
This structure, in which you have loaded gpr values, then discard them only to
load them up again in a different helper function, could use some improvement.
> + case NM_BALIGN:
> + gen_load_gpr(t0, v1);
> + v2 &= 3;
> + if (v2 != 0 && v2 != 2) {
> + tcg_gen_shli_tl(cpu_gpr[ret], cpu_gpr[ret], 8 * v2);
More unprotected uses of cpu_gpr[0].
r~
