qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: zhiwei_liu@linux.alibaba.com, tangtiancheng.ttc@alibaba-inc.com,
	philmd@linaro.org
Subject: [PATCH v2 08/18] tcg/optimize: Optimize cmp_vec and cmpsel_vec
Date: Wed, 11 Sep 2024 09:50:37 -0700	[thread overview]
Message-ID: <20240911165047.1035764-9-richard.henderson@linaro.org> (raw)
In-Reply-To: <20240911165047.1035764-1-richard.henderson@linaro.org>

Place immediate values second in the comparison.
Place destination matches first in the true/false values.
All of this mirrors what we do for integer setcond and movcond.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 36 ++++++++++++++++++++++++++++++++++++
 1 file changed, 36 insertions(+)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index cf311790e0..f11f576fd4 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2422,6 +2422,36 @@ static bool fold_setcond2(OptContext *ctx, TCGOp *op)
     return tcg_opt_gen_movi(ctx, op, op->args[0], i);
 }
 
+static bool fold_cmp_vec(OptContext *ctx, TCGOp *op)
+{
+    /* Canonicalize the comparison to put immediate second. */
+    if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) {
+        op->args[3] = tcg_swap_cond(op->args[3]);
+    }
+    return false;
+}
+
+static bool fold_cmpsel_vec(OptContext *ctx, TCGOp *op)
+{
+    /* If true and false values are the same, eliminate the cmp. */
+    if (args_are_copies(op->args[3], op->args[4])) {
+        return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[3]);
+    }
+
+    /* Canonicalize the comparison to put immediate second. */
+    if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) {
+        op->args[5] = tcg_swap_cond(op->args[5]);
+    }
+    /*
+     * Canonicalize the "false" input reg to match the destination,
+     * so that the tcg backend can implement "move if true".
+     */
+    if (swap_commutative(op->args[0], &op->args[4], &op->args[3])) {
+        op->args[5] = tcg_invert_cond(op->args[5]);
+    }
+    return false;
+}
+
 static bool fold_sextract(OptContext *ctx, TCGOp *op)
 {
     uint64_t z_mask, s_mask, s_mask_old;
@@ -2928,6 +2958,12 @@ void tcg_optimize(TCGContext *s)
         case INDEX_op_setcond2_i32:
             done = fold_setcond2(&ctx, op);
             break;
+        case INDEX_op_cmp_vec:
+            done = fold_cmp_vec(&ctx, op);
+            break;
+        case INDEX_op_cmpsel_vec:
+            done = fold_cmpsel_vec(&ctx, op);
+            break;
         CASE_OP_32_64(sextract):
             done = fold_sextract(&ctx, op);
             break;
-- 
2.43.0



  parent reply	other threads:[~2024-09-11 16:52 UTC|newest]

Thread overview: 20+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-09-11 16:50 [PATCH v2 00/18] tcg: Improve support for cmpsel_vec Richard Henderson
2024-09-11 16:50 ` [PATCH v2 01/18] tcg: Export vec_gen_6 Richard Henderson
2024-09-11 16:50 ` [PATCH v2 02/18] tcg/i386: Split out tcg_out_vex_modrm_type Richard Henderson
2024-09-11 16:50 ` [PATCH v2 03/18] tcg/i386: Do not expand cmp_vec early Richard Henderson
2024-09-11 16:50 ` [PATCH v2 04/18] tcg/i386: Do not expand cmpsel_vec early Richard Henderson
2024-09-11 16:50 ` [PATCH v2 05/18] tcg/ppc: Do not expand cmp_vec early Richard Henderson
2024-09-11 16:50 ` [PATCH v2 06/18] tcg/s390x: " Richard Henderson
2024-09-11 16:50 ` [PATCH v2 07/18] tcg/optimize: Fold movcond with true and false values identical Richard Henderson
2024-09-11 16:50 ` Richard Henderson [this message]
2024-09-11 16:50 ` [PATCH v2 09/18] tcg/optimize: Optimize bitsel_vec Richard Henderson
2024-09-11 16:50 ` [PATCH v2 10/18] tcg/i386: Optimize cmpsel with constant 0 operand 3 Richard Henderson
2024-09-11 16:50 ` [PATCH v2 11/18] tcg/i386: Implement cmp_vec with avx512 insns Richard Henderson
2024-09-11 16:50 ` [PATCH v2 12/18] tcg/i386: Add predicate parameters to tcg_out_evex_opc Richard Henderson
2024-09-11 16:50 ` [PATCH v2 13/18] tcg/i386: Implement cmpsel_vec with avx512 insns Richard Henderson
2024-09-11 16:50 ` [PATCH v2 14/18] tcg/i386: Implement vector TST{EQ,NE} for avx512 Richard Henderson
2024-09-12  6:04   ` [PATCH v2 14/18] tcg/i386: Implement vector TST{EQ, NE} " Philippe Mathieu-Daudé
2024-09-11 16:50 ` [PATCH v2 15/18] tcg/ppc: Implement cmpsel_vec Richard Henderson
2024-09-11 16:50 ` [PATCH v2 16/18] tcg/ppc: Optimize cmpsel with constant 0/-1 arguments Richard Henderson
2024-09-11 16:50 ` [PATCH v2 17/18] tcg/s390x: Implement cmpsel_vec Richard Henderson
2024-09-11 16:50 ` [PATCH v2 18/18] tcg/s390x: Optimize cmpsel with constant 0/-1 arguments Richard Henderson

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=20240911165047.1035764-9-richard.henderson@linaro.org \
    --to=richard.henderson@linaro.org \
    --cc=philmd@linaro.org \
    --cc=qemu-devel@nongnu.org \
    --cc=tangtiancheng.ttc@alibaba-inc.com \
    --cc=zhiwei_liu@linux.alibaba.com \
    /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).