Re: [Qemu-devel] [PATCH 1/6] target-arm: Fix rounding constant addition for Neon shift instructions.

From: Peter Maydell <peter.maydell@linaro.org>
To: christophe.lyon@st.com
Cc: qemu-devel@nongnu.org
Subject: Re: [Qemu-devel] [PATCH 1/6] target-arm: Fix rounding constant addition for Neon shift instructions.
Date: Mon, 14 Feb 2011 18:12:59 +0000	[thread overview]
Message-ID: <AANLkTinPVdCstd=tfDugoo7zSMJ2GxrAMyzyR6aqOzSn@mail.gmail.com> (raw)
In-Reply-To: <1297437062-6118-2-git-send-email-christophe.lyon@st.com>

On 11 February 2011 15:10,  <christophe.lyon@st.com> wrote:
> +uint32_t HELPER(neon_rshl_s32)(uint32_t valop, uint32_t shiftop)
> +{
> +    int32_t dest;
> +    int32_t val = (int32_t)valop;
> +    int8_t shift = (int8_t)shiftop;
> +    if (shift >= 32) {
> +        dest = 0;
> +    } else if (shift < -32) {
> +        dest = val >> 31;

This is the wrong answer: large rounding right shifts give zero.

> +    } else if (shift == -32) {
> +        dest = val >> 31;
> +        dest++;
> +        dest >>= 1;

These three lines will always result in dest becoming
0 regardless of the input value.

I'm going to post a patch which fixes these as part
of getting the answers right for VRSHL by large shift
counts in general.

-- PMM