From: Laurent Desnogues <laurent.desnogues@gmail.com>
To: Richard Henderson <richard.henderson@linaro.org>
Cc: Peter Maydell <peter.maydell@linaro.org>,
qemu-arm <qemu-arm@nongnu.org>,
"qemu-devel@nongnu.org" <qemu-devel@nongnu.org>
Subject: Re: [Qemu-devel] [Qemu-arm] [PATCH 6/7] target/arm: Simplify SMMLA, SMMLAR, SMMLS, SMMLSR
Date: Wed, 28 Aug 2019 09:22:14 +0200 [thread overview]
Message-ID: <CABoDooPO37bEjDF77_mU6Z57boRzuYaDP8DcTqtAOJxCjvBkRw@mail.gmail.com> (raw)
In-Reply-To: <20190808202616.13782-7-richard.henderson@linaro.org>
Hi Richard,
On Thu, Aug 8, 2019 at 10:28 PM Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> All of the inputs to these instructions are 32-bits. Rather than
> extend each input to 64-bits and then extract the high 32-bits of
> the output, use tcg_gen_muls2_i32 and other 32-bit generator functions.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
> target/arm/translate.c | 72 +++++++++++++++---------------------------
> 1 file changed, 26 insertions(+), 46 deletions(-)
>
> diff --git a/target/arm/translate.c b/target/arm/translate.c
> index ddc54e77e4..77154be743 100644
> --- a/target/arm/translate.c
> +++ b/target/arm/translate.c
> @@ -391,34 +391,6 @@ static void gen_revsh(TCGv_i32 var)
> tcg_gen_ext16s_i32(var, var);
> }
>
> -/* Return (b << 32) + a. Mark inputs as dead */
> -static TCGv_i64 gen_addq_msw(TCGv_i64 a, TCGv_i32 b)
> -{
> - TCGv_i64 tmp64 = tcg_temp_new_i64();
> -
> - tcg_gen_extu_i32_i64(tmp64, b);
> - tcg_temp_free_i32(b);
> - tcg_gen_shli_i64(tmp64, tmp64, 32);
> - tcg_gen_add_i64(a, tmp64, a);
> -
> - tcg_temp_free_i64(tmp64);
> - return a;
> -}
> -
> -/* Return (b << 32) - a. Mark inputs as dead. */
> -static TCGv_i64 gen_subq_msw(TCGv_i64 a, TCGv_i32 b)
> -{
> - TCGv_i64 tmp64 = tcg_temp_new_i64();
> -
> - tcg_gen_extu_i32_i64(tmp64, b);
> - tcg_temp_free_i32(b);
> - tcg_gen_shli_i64(tmp64, tmp64, 32);
> - tcg_gen_sub_i64(a, tmp64, a);
> -
> - tcg_temp_free_i64(tmp64);
> - return a;
> -}
> -
> /* 32x32->64 multiply. Marks inputs as dead. */
> static TCGv_i64 gen_mulu_i64_i32(TCGv_i32 a, TCGv_i32 b)
> {
> @@ -8872,23 +8844,27 @@ static void disas_arm_insn(DisasContext *s, unsigned int insn)
> (SMMUL, SMMLA, SMMLS) */
> tmp = load_reg(s, rm);
> tmp2 = load_reg(s, rs);
> - tmp64 = gen_muls_i64_i32(tmp, tmp2);
> + tcg_gen_muls2_i32(tmp2, tmp, tmp, tmp2);
>
> if (rd != 15) {
> - tmp = load_reg(s, rd);
> + tmp3 = load_reg(s, rd);
> if (insn & (1 << 6)) {
> - tmp64 = gen_subq_msw(tmp64, tmp);
> + tcg_gen_sub_i32(tmp, tmp, tmp3);
Shouldn't you subtract tmp from tmp3?
> } else {
> - tmp64 = gen_addq_msw(tmp64, tmp);
> + tcg_gen_add_i32(tmp, tmp, tmp3);
> }
> + tcg_temp_free_i32(tmp3);
> }
> if (insn & (1 << 5)) {
> - tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
> + /*
> + * Adding 0x80000000 to the 64-bit quantity
> + * means that we have carry in to the high
> + * word when the low word has the high bit set.
> + */
> + tcg_gen_shri_i32(tmp2, tmp2, 31);
> + tcg_gen_add_i32(tmp, tmp, tmp2);
> }
> - tcg_gen_shri_i64(tmp64, tmp64, 32);
> - tmp = tcg_temp_new_i32();
> - tcg_gen_extrl_i64_i32(tmp, tmp64);
> - tcg_temp_free_i64(tmp64);
> + tcg_temp_free_i32(tmp2);
> store_reg(s, rn, tmp);
> break;
> case 0:
> @@ -10114,22 +10090,26 @@ static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
> }
> break;
> case 5: case 6: /* 32 * 32 -> 32msb (SMMUL, SMMLA, SMMLS) */
> - tmp64 = gen_muls_i64_i32(tmp, tmp2);
> + tcg_gen_muls2_i32(tmp2, tmp, tmp, tmp2);
> if (rs != 15) {
> - tmp = load_reg(s, rs);
> + tmp3 = load_reg(s, rs);
> if (insn & (1 << 20)) {
> - tmp64 = gen_addq_msw(tmp64, tmp);
> + tcg_gen_add_i32(tmp, tmp, tmp3);
> } else {
> - tmp64 = gen_subq_msw(tmp64, tmp);
> + tcg_gen_sub_i32(tmp, tmp, tmp3);
Same here.
Also the way you do the computation means you don't propagate the
borrow from the lower 32-bit of the 64-bit product when doing the
subtraction.
Thanks,
Laurent
> }
> + tcg_temp_free_i32(tmp3);
> }
> if (insn & (1 << 4)) {
> - tcg_gen_addi_i64(tmp64, tmp64, 0x80000000u);
> + /*
> + * Adding 0x80000000 to the 64-bit quantity
> + * means that we have carry in to the high
> + * word when the low word has the high bit set.
> + */
> + tcg_gen_shri_i32(tmp2, tmp2, 31);
> + tcg_gen_add_i32(tmp, tmp, tmp2);
> }
> - tcg_gen_shri_i64(tmp64, tmp64, 32);
> - tmp = tcg_temp_new_i32();
> - tcg_gen_extrl_i64_i32(tmp, tmp64);
> - tcg_temp_free_i64(tmp64);
> + tcg_temp_free_i32(tmp2);
> break;
> case 7: /* Unsigned sum of absolute differences. */
> gen_helper_usad8(tmp, tmp, tmp2);
> --
> 2.17.1
>
>
next prev parent reply other threads:[~2019-08-28 7:23 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-08-08 20:26 [Qemu-devel] [PATCH 0/7] target/arm: Misc cleanups Richard Henderson
2019-08-08 20:26 ` [Qemu-devel] [PATCH 1/7] target/arm: Use tcg_gen_extract_i32 for shifter_out_im Richard Henderson
2019-08-08 20:26 ` [Qemu-devel] [PATCH 2/7] target/arm: Use tcg_gen_deposit_i32 for PKHBT, PKHTB Richard Henderson
2019-08-08 20:26 ` [Qemu-devel] [PATCH 3/7] target/arm: Remove redundant shift tests Richard Henderson
2019-08-08 20:26 ` [Qemu-devel] [PATCH 4/7] target/arm: Use ror32 instead of open-coding the operation Richard Henderson
2019-08-08 20:26 ` [Qemu-devel] [PATCH 5/7] target/arm: Use tcg_gen_rotri_i32 for gen_swap_half Richard Henderson
2019-08-08 20:26 ` [Qemu-devel] [PATCH 6/7] target/arm: Simplify SMMLA, SMMLAR, SMMLS, SMMLSR Richard Henderson
2019-08-28 7:22 ` Laurent Desnogues [this message]
2019-08-08 20:26 ` [Qemu-devel] [PATCH 7/7] target/arm: Use tcg_gen_extrh_i64_i32 to extract the high word Richard Henderson
2019-08-15 10:16 ` Peter Maydell
2019-08-15 11:56 ` Richard Henderson
2019-08-15 12:02 ` Peter Maydell
2019-08-15 10:34 ` [Qemu-devel] [PATCH 0/7] target/arm: Misc cleanups Peter Maydell
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CABoDooPO37bEjDF77_mU6Z57boRzuYaDP8DcTqtAOJxCjvBkRw@mail.gmail.com \
--to=laurent.desnogues@gmail.com \
--cc=peter.maydell@linaro.org \
--cc=qemu-arm@nongnu.org \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).