qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Chinmay Rath <rathc@linux.vnet.ibm.com>
To: Richard Henderson <richard.henderson@linaro.org>,
	Chinmay Rath <rathc@linux.ibm.com>,
	qemu-ppc@nongnu.org
Cc: qemu-devel@nongnu.org, npiggin@gmail.com, danielhb413@gmail.com,
	harshpb@linux.ibm.com
Subject: Re: [PATCH 7/8] target/ppc: Move cmp{rb, eqb}, tw[i], td[i], isel instructions to decodetree.
Date: Fri, 19 Apr 2024 14:58:53 +0530	[thread overview]
Message-ID: <0bc9e291-f072-42c2-b001-cf059f94291b@linux.vnet.ibm.com> (raw)
In-Reply-To: <1f39a25e-6c64-4ff2-8521-7059c5ed2214@linaro.org>

Hi Richard,

On 4/17/24 00:50, Richard Henderson wrote:
> On 4/15/24 23:39, Chinmay Rath wrote:
>> Moving the following instructions to decodetree specification :
>>
>>     cmp{rb, eqb}, t{w, d}    : X-form
>>     t{w, d}i        : D-form
>>     isel            : A-form
>>
>> The changes were verified by validating that the tcg ops generated by 
>> those
>> instructions remain the same, which were captured using the '-d 
>> in_asm,op' flag.
>>
>> Signed-off-by: Chinmay Rath <rathc@linux.ibm.com>
>
> A faithful reorg of the existing code, so,
> Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
>
Thank you.
> Notes for improvement:
>
>> +static bool trans_CMPRB(DisasContext *ctx, arg_CMPRB *a)
>> +{
>> +    TCGv_i32 src1 = tcg_temp_new_i32();
>> +    TCGv_i32 src2 = tcg_temp_new_i32();
>> +    TCGv_i32 src2lo = tcg_temp_new_i32();
>> +    TCGv_i32 src2hi = tcg_temp_new_i32();
>> +    TCGv_i32 crf = cpu_crf[a->bf];
>> +
>> +    REQUIRE_INSNS_FLAGS2(ctx, ISA300);
>> +    tcg_gen_trunc_tl_i32(src1, cpu_gpr[a->ra]);
>> +    tcg_gen_trunc_tl_i32(src2, cpu_gpr[a->rb]);
>> +
>> +    tcg_gen_andi_i32(src1, src1, 0xFF);
>> +    tcg_gen_ext8u_i32(src2lo, src2);
>> +    tcg_gen_shri_i32(src2, src2, 8);
>> +    tcg_gen_ext8u_i32(src2hi, src2);
>
> tcg_gen_extract_i32(src2hi, src2, 8, 8);
>
>> +
>> +    tcg_gen_setcond_i32(TCG_COND_LEU, src2lo, src2lo, src1);
>> +    tcg_gen_setcond_i32(TCG_COND_LEU, src2hi, src1, src2hi);
>> +    tcg_gen_and_i32(crf, src2lo, src2hi);
>> +
>> +    if (a->l) {
>> +        tcg_gen_shri_i32(src2, src2, 8);
>> +        tcg_gen_ext8u_i32(src2lo, src2);
>
> tcg_gen_extract_i32(src2lo, src2, 16, 8);
>
>> +        tcg_gen_shri_i32(src2, src2, 8);
>> +        tcg_gen_ext8u_i32(src2hi, src2);
>
> tcg_gen_extract_i32(src2hi, src2, 24, 8);
>
Will update the above in v2.

Will implement the below improvements for trap insns as a separate patch 
later.

>> +/*
>> + * Fixed-Point Trap Instructions
>> + */
>> +
>> +static bool trans_TW(DisasContext *ctx, arg_TW *a)
>> +{
>> +    TCGv_i32 t0;
>> +
>> +    if (check_unconditional_trap(ctx, a->rt)) {
>> +        return true;
>> +    }
>> +    t0 = tcg_constant_i32(a->rt);
>> +    gen_helper_TW(tcg_env, cpu_gpr[a->ra], cpu_gpr[a->rb], t0);
>> +    return true;
>> +}
>> +
>> +static bool trans_TWI(DisasContext *ctx, arg_TWI *a)
>> +{
>> +    TCGv t0;
>> +    TCGv_i32 t1;
>> +
>> +    if (check_unconditional_trap(ctx, a->rt)) {
>> +        return true;
>> +    }
>> +    t0 = tcg_constant_tl(a->si);
>> +    t1 = tcg_constant_i32(a->rt);
>> +    gen_helper_TW(tcg_env, cpu_gpr[a->ra], t0, t1);
>> +    return true;
>> +}
>> +
>> +static bool trans_TD(DisasContext *ctx, arg_TD *a)
>> +{
>> +    TCGv_i32 t0;
>> +
>> +    REQUIRE_64BIT(ctx);
>> +    if (check_unconditional_trap(ctx, a->rt)) {
>> +        return true;
>> +    }
>> +    t0 = tcg_constant_i32(a->rt);
>> +    gen_helper_TD(tcg_env, cpu_gpr[a->ra], cpu_gpr[a->rb], t0);
>> +    return true;
>> +}
>> +
>> +static bool trans_TDI(DisasContext *ctx, arg_TDI *a)
>> +{
>> +    TCGv t0;
>> +    TCGv_i32 t1;
>> +
>> +    REQUIRE_64BIT(ctx);
>> +    if (check_unconditional_trap(ctx, a->rt)) {
>> +        return true;
>> +    }
>> +    t0 = tcg_constant_tl(a->si);
>> +    t1 = tcg_constant_i32(a->rt);
>> +    gen_helper_TD(tcg_env, cpu_gpr[a->ra], t0, t1);
>> +    return true;
>> +}
>
> See target/sparc/translate.c, delay_exception, for a method of 
> implementing compare-and-trap inline with no inline branch penalty.
>
> static void do_conditional_trap(DisasContext *ctx, unsigned to, TCGv 
> a, TCGv b)
> {
>     static const TCGCond ucond[8] = {
>         TCG_COND_NEVER, TCG_COND_GTU, TCG_COND_LTU, TCG_COND_NE,
>         TCG_COND_EQ,    TCG_COND_GEU, TCG_COND_LEU, TCG_COND_ALWAYS,
>     };
>     static const TCGCond scond[8] = {
>         TCG_COND_NEVER, TCG_COND_EQ,  TCG_COND_GT,  TCG_COND_GE,
>         TCG_COND_LT,    TCG_COND_LE,  TCG_COND_NE, TCG_COND_ALWAYS,
>     };
>
>     TCGCond uc = ucond[to & 7];
>     TCGCond sc = scond[to >> 2];
>
>     /* There is overlap with EQ; we may not need both comparisons. */
>     if (!(to & 0x18)) {
>         sc = TCG_COND_NEVER;
>     } else if (!(to & 0x03)) {
>         uc = TCG_COND_NEVER;
>     }
>
>     if (uc == TCG_COND_ALWAYS || sc == TCG_COND_ALWAYS) {
>         unconditional trap;
>         return true;
>     }
>     if (uc == TCG_COND_NEVER && sc == TCG_COND_NEVER) {
>         return true;
>     }
>
>     e = delay_exception(ctx, POWERPC_EXCP_TRAP);
>
>     if (uc != TCG_COND_NEVER) {
>         tcg_gen_brcond_tl(uc, a, b, e->lab);
>     }
>     if (sc != TCG_COND_NEVER) {
>         tcg_gen_brcond_tl(sc, a, b, e->lab);
>     }
>     return true;
> }
>
> bool trans_TW(...)
> {
>     TCGv a = tcg_temp_new();
>     TCGv b = tcg_temp_new();
>
>     /* Note that consistent sign extensions work for unsigned 
> comparisons. */
>     tcg_gen_exts_i32_tl(a, ra);
>     tcg_gen_exts_i32_tl(b, rb);
>     return do_conditional_trap(ctx, to, a, b);
> }
>
> etc.
>
>
Thanks,
Chinmay
> r~



  reply	other threads:[~2024-04-19  9:29 UTC|newest]

Thread overview: 27+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-04-16  6:39 [PATCH 0/8] target/ppc: Move fixed-point insns to decodetree Chinmay Rath
2024-04-16  6:39 ` [PATCH 1/8] target/ppc: Move mul{li, lw, lwo, hw, hwu} instructions " Chinmay Rath
2024-04-16 17:56   ` Richard Henderson
2024-04-19  9:08     ` Chinmay Rath
2024-04-16  6:39 ` [PATCH 2/8] target/ppc: Make divw[u] handler method decodetree compatible Chinmay Rath
2024-04-16 17:57   ` Richard Henderson
2024-04-19  9:17     ` Chinmay Rath
2024-04-16  6:39 ` [PATCH 3/8] target/ppc: Move divw[u, e, eu] instructions to decodetree Chinmay Rath
2024-04-16 18:19   ` Richard Henderson
2024-04-19  9:18     ` Chinmay Rath
2024-04-16  6:39 ` [PATCH 4/8] target/ppc: Move neg, darn, mod{sw, uw} " Chinmay Rath
2024-04-16 18:25   ` Richard Henderson
2024-04-19  9:18     ` Chinmay Rath
2024-04-16  6:39 ` [PATCH 5/8] target/ppc: Move multiply fixed-point insns (64-bit operands) " Chinmay Rath
2024-04-16 18:36   ` Richard Henderson
2024-04-19  9:25     ` Chinmay Rath
2024-04-20 15:51       ` Richard Henderson
2024-04-22  6:32         ` Chinmay Rath
2024-04-16  6:39 ` [PATCH 6/8] target/ppc: Move div/mod fixed-point insns (64 bits " Chinmay Rath
2024-04-16 18:38   ` Richard Henderson
2024-04-19  9:26     ` Chinmay Rath
2024-04-16  6:39 ` [PATCH 7/8] target/ppc: Move cmp{rb, eqb}, tw[i], td[i], isel instructions " Chinmay Rath
2024-04-16 19:20   ` Richard Henderson
2024-04-19  9:28     ` Chinmay Rath [this message]
2024-04-16  6:39 ` [PATCH 8/8] target/ppc: Move logical fixed-point " Chinmay Rath
2024-04-16 19:35   ` Richard Henderson
2024-04-19  9:29     ` Chinmay Rath

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=0bc9e291-f072-42c2-b001-cf059f94291b@linux.vnet.ibm.com \
    --to=rathc@linux.vnet.ibm.com \
    --cc=danielhb413@gmail.com \
    --cc=harshpb@linux.ibm.com \
    --cc=npiggin@gmail.com \
    --cc=qemu-devel@nongnu.org \
    --cc=qemu-ppc@nongnu.org \
    --cc=rathc@linux.ibm.com \
    --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).