qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Peter Maydell <peter.maydell@linaro.org>
To: Richard Henderson <richard.henderson@linaro.org>
Cc: qemu-devel@nongnu.org, qemu-arm@nongnu.org
Subject: Re: [PATCH v5 26/45] target/arm: Implement FMOPA, FMOPS (widening)
Date: Thu, 7 Jul 2022 10:50:05 +0100	[thread overview]
Message-ID: <CAFEAcA9pv2xH9mvydFK4zqB1viH1DebLB_s7bCYXDGqsCKatew@mail.gmail.com> (raw)
In-Reply-To: <20220706082411.1664825-27-richard.henderson@linaro.org>

On Wed, 6 Jul 2022 at 10:26, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>


> +static float32 f16_dotadd(float32 sum, uint32_t e1, uint32_t e2,
> +                          float_status *s_std, float_status *s_odd)
> +{
> +    float64 e1r = float16_to_float64(e1 & 0xffff, true, s_std);
> +    float64 e1c = float16_to_float64(e1 >> 16, true, s_std);
> +    float64 e2r = float16_to_float64(e2 & 0xffff, true, s_std);
> +    float64 e2c = float16_to_float64(e2 >> 16, true, s_std);
> +    float64 t64;
> +    float32 t32;
> +
> +    /*
> +     * The ARM pseudocode function FPDot performs both multiplies
> +     * and the add with a single rounding operation.  Emulate this
> +     * by performing the first multiply in round-to-odd, then doing
> +     * the second multiply as fused multiply-add, and rounding to
> +     * float32 all in one step.
> +     */

I guess if we find we're not producing quite bit-accurate results
we can come back and revisit this :-)

> +    t64 = float64_mul(e1r, e2r, s_odd);
> +    t64 = float64r32_muladd(e1c, e2c, t64, 0, s_std);
> +
> +    /* This conversion is exact, because we've already rounded. */
> +    t32 = float64_to_float32(t64, s_std);
> +
> +    /* The final accumulation step is not fused. */
> +    return float32_add(sum, t32, s_std);
> +}
> +
> +void HELPER(sme_fmopa_h)(void *vza, void *vzn, void *vzm, void *vpn,
> +                         void *vpm, void *vst, uint32_t desc)
> +{
> +    intptr_t row, col, oprsz = simd_maxsz(desc);
> +    uint32_t neg = simd_data(desc) << 15;
> +    uint16_t *pn = vpn, *pm = vpm;
> +    float_status fpst_odd, fpst_std = *(float_status *)vst;
> +
> +    set_default_nan_mode(true, &fpst_std);
> +    fpst_odd = fpst_std;
> +    set_float_rounding_mode(float_round_to_odd, &fpst_odd);
> +
> +    for (row = 0; row < oprsz; ) {
> +        uint16_t pa = pn[H2(row >> 4)];
> +        do {
> +            void *vza_row = vza + tile_vslice_offset(row);
> +            uint32_t n = *(uint32_t *)(vzn + row);

More missing H macros.

> +
> +            n = f16mop_adj_pair(n, pa, neg);
> +
> +            for (col = 0; col < oprsz; ) {
> +                uint16_t pb = pm[H2(col >> 4)];
> +                do {
> +                    if ((pa & 0b0101) == 0b0101 || (pb & 0b0101) == 0b0101) {

Wrong condition again?

> +                        uint32_t *a = vza_row + col;
> +                        uint32_t m = *(uint32_t *)(vzm + col);
> +
> +                        m = f16mop_adj_pair(m, pb, neg);
> +                        *a = f16_dotadd(*a, n, m, &fpst_std, &fpst_odd);
> +
> +                        col += 4;
> +                        pb >>= 4;
> +                    }
> +                } while (col & 15);
> +            }
> +            row += 4;
> +            pa >>= 4;
> +        } while (row & 15);
> +    }
> +}

thanks
-- PMM


  reply	other threads:[~2022-07-07  9:52 UTC|newest]

Thread overview: 59+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-07-06  8:23 [PATCH v5 00/45] target/arm: Scalable Matrix Extension Richard Henderson
2022-07-06  8:23 ` [PATCH v5 01/45] target/arm: Handle SME in aarch64_cpu_dump_state Richard Henderson
2022-07-06  8:23 ` [PATCH v5 02/45] target/arm: Add infrastructure for disas_sme Richard Henderson
2022-07-06  8:23 ` [PATCH v5 03/45] target/arm: Trap non-streaming usage when Streaming SVE is active Richard Henderson
2022-07-06 16:14   ` Peter Maydell
2022-07-06  8:23 ` [PATCH v5 04/45] target/arm: Mark ADR as non-streaming Richard Henderson
2022-07-06  8:23 ` [PATCH v5 05/45] target/arm: Mark RDFFR, WRFFR, SETFFR " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 06/45] target/arm: Mark BDEP, BEXT, BGRP, COMPACT, FEXPA, FTSSEL " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 07/45] target/arm: Mark PMULL, FMMLA " Richard Henderson
2022-07-06 16:13   ` Peter Maydell
2022-07-06  8:23 ` [PATCH v5 08/45] target/arm: Mark FTSMUL, FTMAD, FADDA " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 09/45] target/arm: Mark SMMLA, UMMLA, USMMLA " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 10/45] target/arm: Mark string/histo/crypto " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 11/45] target/arm: Mark gather/scatter load/store " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 12/45] target/arm: Mark gather prefetch " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 13/45] target/arm: Mark LDFF1 and LDNF1 " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 14/45] target/arm: Mark LD1RO " Richard Henderson
2022-07-06  8:23 ` [PATCH v5 15/45] target/arm: Add SME enablement checks Richard Henderson
2022-07-06  8:23 ` [PATCH v5 16/45] target/arm: Handle SME in sve_access_check Richard Henderson
2022-07-06  8:23 ` [PATCH v5 17/45] target/arm: Implement SME RDSVL, ADDSVL, ADDSPL Richard Henderson
2022-07-06  8:23 ` [PATCH v5 18/45] target/arm: Implement SME ZERO Richard Henderson
2022-07-06  8:23 ` [PATCH v5 19/45] target/arm: Implement SME MOVA Richard Henderson
2022-07-06 16:47   ` Peter Maydell
2022-07-07  0:56     ` Richard Henderson
2022-07-06  8:23 ` [PATCH v5 20/45] target/arm: Implement SME LD1, ST1 Richard Henderson
2022-07-06 16:50   ` Peter Maydell
2022-07-06  8:23 ` [PATCH v5 21/45] target/arm: Export unpredicated ld/st from translate-sve.c Richard Henderson
2022-07-06  8:23 ` [PATCH v5 22/45] target/arm: Implement SME LDR, STR Richard Henderson
2022-07-06  8:23 ` [PATCH v5 23/45] target/arm: Implement SME ADDHA, ADDVA Richard Henderson
2022-07-06 16:53   ` Peter Maydell
2022-07-06  8:23 ` [PATCH v5 24/45] target/arm: Implement FMOPA, FMOPS (non-widening) Richard Henderson
2022-07-07  9:26   ` Peter Maydell
2022-07-06  8:23 ` [PATCH v5 25/45] target/arm: Implement BFMOPA, BFMOPS Richard Henderson
2022-07-07  9:42   ` Peter Maydell
2022-07-08 14:42     ` Richard Henderson
2022-07-06  8:23 ` [PATCH v5 26/45] target/arm: Implement FMOPA, FMOPS (widening) Richard Henderson
2022-07-07  9:50   ` Peter Maydell [this message]
2022-07-06  8:23 ` [PATCH v5 27/45] target/arm: Implement SME integer outer product Richard Henderson
2022-07-06  8:23 ` [PATCH v5 28/45] target/arm: Implement PSEL Richard Henderson
2022-07-06  8:23 ` [PATCH v5 29/45] target/arm: Implement REVD Richard Henderson
2022-07-06  8:23 ` [PATCH v5 30/45] target/arm: Implement SCLAMP, UCLAMP Richard Henderson
2022-07-06  8:23 ` [PATCH v5 31/45] target/arm: Reset streaming sve state on exception boundaries Richard Henderson
2022-07-06  8:23 ` [PATCH v5 32/45] target/arm: Enable SME for -cpu max Richard Henderson
2022-07-06  8:23 ` [PATCH v5 33/45] linux-user/aarch64: Clear tpidr2_el0 if CLONE_SETTLS Richard Henderson
2022-07-06  8:24 ` [PATCH v5 34/45] linux-user/aarch64: Reset PSTATE.SM on syscalls Richard Henderson
2022-07-06  8:24 ` [PATCH v5 35/45] linux-user/aarch64: Add SM bit to SVE signal context Richard Henderson
2022-07-06 16:54   ` Peter Maydell
2022-07-06  8:24 ` [PATCH v5 36/45] linux-user/aarch64: Tidy target_restore_sigframe error return Richard Henderson
2022-07-06  8:24 ` [PATCH v5 37/45] linux-user/aarch64: Do not allow duplicate or short sve records Richard Henderson
2022-07-06 16:55   ` Peter Maydell
2022-07-06  8:24 ` [PATCH v5 38/45] linux-user/aarch64: Verify extra record lock succeeded Richard Henderson
2022-07-06  8:24 ` [PATCH v5 39/45] linux-user/aarch64: Move sve record checks into restore Richard Henderson
2022-07-06  8:24 ` [PATCH v5 40/45] linux-user/aarch64: Implement SME signal handling Richard Henderson
2022-07-06  8:24 ` [PATCH v5 41/45] linux-user: Rename sve prctls Richard Henderson
2022-07-06  8:24 ` [PATCH v5 42/45] linux-user/aarch64: Implement PR_SME_GET_VL, PR_SME_SET_VL Richard Henderson
2022-07-06  8:24 ` [PATCH v5 43/45] target/arm: Only set ZEN in reset if SVE present Richard Henderson
2022-07-06  8:24 ` [PATCH v5 44/45] target/arm: Enable SME for user-only Richard Henderson
2022-07-06  8:24 ` [PATCH v5 45/45] linux-user/aarch64: Add SME related hwcap entries Richard Henderson
2022-07-07  9:52 ` [PATCH v5 00/45] target/arm: Scalable Matrix Extension 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=CAFEAcA9pv2xH9mvydFK4zqB1viH1DebLB_s7bCYXDGqsCKatew@mail.gmail.com \
    --to=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).