qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH 02/46] tcg/optimize: Copy mask writeback to fold_masks


From: Pierrick Bouvier
Subject: Re: [PATCH 02/46] tcg/optimize: Copy mask writeback to fold_masks
Date: Tue, 17 Dec 2024 12:00:28 -0800
User-agent: Mozilla Thunderbird

On 12/10/24 07:23, Richard Henderson wrote:
Use of fold_masks should be restricted to those opcodes that
can reliably make use of it -- those with a single output,
and from higher-level folders that set up the masks.
Prepare for conversion of each folder in turn.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
  tcg/optimize.c | 14 +++++++++++---
  1 file changed, 11 insertions(+), 3 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 20c918e83b..1a9e3258e3 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1050,6 +1050,11 @@ static bool fold_masks(OptContext *ctx, TCGOp *op)
      uint64_t a_mask = ctx->a_mask;
      uint64_t z_mask = ctx->z_mask;
      uint64_t s_mask = ctx->s_mask;
+    const TCGOpDef *def = &tcg_op_defs[op->opc];
+    TCGTemp *ts;
+
+    /* Only single-output opcodes are supported here. */
+    tcg_debug_assert(def->nb_oargs == 1);
/*
       * 32-bit ops generate 32-bit results, which for the purpose of
@@ -1062,8 +1067,6 @@ static bool fold_masks(OptContext *ctx, TCGOp *op)
          a_mask = (int32_t)a_mask;
          z_mask = (int32_t)z_mask;
          s_mask |= MAKE_64BIT_MASK(32, 32);
-        ctx->z_mask = z_mask;
-        ctx->s_mask = s_mask;
      }
if (z_mask == 0) {
@@ -1072,7 +1075,12 @@ static bool fold_masks(OptContext *ctx, TCGOp *op)
      if (a_mask == 0) {
          return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]);
      }
-    return false;
+
+    ts = arg_temp(op->args[0]);
+    reset_ts(ctx, ts);
+    ts_info(ts)->z_mask = z_mask;
+    ts_info(ts)->s_mask = s_mask;
+    return true;
  }
/*

Maybe it would be worth to add a comment stating that a_mask was not modified, but only clipped to a 32 bit value.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>




reply via email to

[Prev in Thread] Current Thread [Next in Thread]