From: Stephen Long <stepl...@quicinc.com>
The fallback inline expansion for vectorized absolute value,
when the host doesn't support such an insn was flawed.
E.g. when a vector of bytes has all elements negative, mask
will be 0xffff_ffff_ffff_ffff. Subtracting mask only adds 1
to the low element instead of all elements becase -mask is 1
and not 0x0101_0101_0101_0101.
Signed-off-by: Stephen Long <stepl...@quicinc.com>
Message-Id: <20200813161818.190-1-stepl...@quicinc.com>
Signed-off-by: Richard Henderson <richard.hender...@linaro.org>
---
tcg/tcg-op-gvec.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c
index 3707c0effb..793d4ba64c 100644
--- a/tcg/tcg-op-gvec.c
+++ b/tcg/tcg-op-gvec.c
@@ -2264,12 +2264,13 @@ static void gen_absv_mask(TCGv_i64 d, TCGv_i64 b,
unsigned vece)
tcg_gen_muli_i64(t, t, (1 << nbit) - 1);
/*
- * Invert (via xor -1) and add one (via sub -1).
+ * Invert (via xor -1) and add one.
* Because of the ordering the msb is cleared,
* so we never have carry into the next element.
*/
tcg_gen_xor_i64(d, b, t);
- tcg_gen_sub_i64(d, d, t);
+ tcg_gen_andi_i64(t, t, dup_const(vece, 1));
+ tcg_gen_add_i64(d, d, t);
tcg_temp_free_i64(t);
}
--
2.25.1
