Skip to content

Commit

Permalink
tcg-aarch64: Prefer unsigned offsets before signed offsets for ldst
Browse files Browse the repository at this point in the history
The assembler seems to prefer them, perhaps we should too.

Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
  • Loading branch information
rth7680 authored and Richard Henderson committed Apr 16, 2014
1 parent 3d4299f commit a056c9f
Showing 1 changed file with 6 additions and 5 deletions.
11 changes: 6 additions & 5 deletions tcg/aarch64/tcg-target.c
Expand Up @@ -669,11 +669,6 @@ static void tcg_out_ldst(TCGContext *s, AArch64Insn insn,
{
TCGMemOp size = (uint32_t)insn >> 30;

if (offset >= -256 && offset < 256) {
tcg_out_insn_3312(s, insn, rd, rn, offset);
return;
}

/* If the offset is naturally aligned and in range, then we can
use the scaled uimm12 encoding */
if (offset >= 0 && !(offset & ((1 << size) - 1))) {
Expand All @@ -684,6 +679,12 @@ static void tcg_out_ldst(TCGContext *s, AArch64Insn insn,
}
}

/* Small signed offsets can use the unscaled encoding. */
if (offset >= -256 && offset < 256) {
tcg_out_insn_3312(s, insn, rd, rn, offset);
return;
}

/* Worst-case scenario, move offset to temp register, use reg offset. */
tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP, offset);
tcg_out_ldst_r(s, insn, rd, rn, TCG_REG_TMP);
Expand Down

0 comments on commit a056c9f

Please sign in to comment.