[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PULL 07/72] tcg/optimize: Augment s_mask from z_mask in fold_masks_zs
From: |
Richard Henderson |
Subject: |
[PULL 07/72] tcg/optimize: Augment s_mask from z_mask in fold_masks_zs |
Date: |
Tue, 24 Dec 2024 12:04:16 -0800 |
Consider the passed s_mask to be a minimum deduced from
either existing s_mask or from a sign-extension operation.
We may be able to deduce more from the set of known zeros.
Remove identical logic from several opcode folders.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
tcg/optimize.c | 21 ++++++---------------
1 file changed, 6 insertions(+), 15 deletions(-)
diff --git a/tcg/optimize.c b/tcg/optimize.c
index d70127b88d..d8f6542c4f 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1048,6 +1048,7 @@ static bool fold_const2_commutative(OptContext *ctx,
TCGOp *op)
* Record "zero" and "sign" masks for the single output of @op.
* See TempOptInfo definition of z_mask and s_mask.
* If z_mask allows, fold the output to constant zero.
+ * The passed s_mask may be augmented by z_mask.
*/
static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
uint64_t z_mask, uint64_t s_mask)
@@ -1080,7 +1081,7 @@ static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
ti = ts_info(ts);
ti->z_mask = z_mask;
- ti->s_mask = s_mask;
+ ti->s_mask = s_mask | smask_from_zmask(z_mask);
return true;
}
@@ -1519,8 +1520,8 @@ static bool fold_bswap(OptContext *ctx, TCGOp *op)
default:
g_assert_not_reached();
}
- s_mask = smask_from_zmask(z_mask);
+ s_mask = 0;
switch (op->args[2] & (TCG_BSWAP_OZ | TCG_BSWAP_OS)) {
case TCG_BSWAP_OZ:
break;
@@ -1534,7 +1535,6 @@ static bool fold_bswap(OptContext *ctx, TCGOp *op)
default:
/* The high bits are undefined: force all bits above the sign to 1. */
z_mask |= sign << 1;
- s_mask = 0;
break;
}
ctx->z_mask = z_mask;
@@ -1605,7 +1605,6 @@ static bool fold_count_zeros(OptContext *ctx, TCGOp *op)
g_assert_not_reached();
}
ctx->z_mask = arg_info(op->args[2])->z_mask | z_mask;
- ctx->s_mask = smask_from_zmask(ctx->z_mask);
return false;
}
@@ -1625,7 +1624,6 @@ static bool fold_ctpop(OptContext *ctx, TCGOp *op)
default:
g_assert_not_reached();
}
- ctx->s_mask = smask_from_zmask(ctx->z_mask);
return false;
}
@@ -1746,7 +1744,6 @@ static bool fold_extract(OptContext *ctx, TCGOp *op)
return true;
}
ctx->z_mask = z_mask;
- ctx->s_mask = smask_from_zmask(z_mask);
return fold_masks(ctx, op);
}
@@ -1851,7 +1848,6 @@ static bool fold_extu(OptContext *ctx, TCGOp *op)
}
ctx->z_mask = z_mask;
- ctx->s_mask = smask_from_zmask(z_mask);
if (!type_change && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) {
return true;
}
@@ -2116,10 +2112,10 @@ static bool fold_qemu_ld(OptContext *ctx, TCGOp *op)
int width = 8 * memop_size(mop);
if (width < 64) {
- ctx->s_mask = MAKE_64BIT_MASK(width, 64 - width);
- if (!(mop & MO_SIGN)) {
+ if (mop & MO_SIGN) {
+ ctx->s_mask = MAKE_64BIT_MASK(width, 64 - width);
+ } else {
ctx->z_mask = MAKE_64BIT_MASK(0, width);
- ctx->s_mask <<= 1;
}
}
@@ -2354,7 +2350,6 @@ static bool fold_setcond(OptContext *ctx, TCGOp *op)
fold_setcond_tst_pow2(ctx, op, false);
ctx->z_mask = 1;
- ctx->s_mask = smask_from_zmask(1);
return false;
}
@@ -2455,7 +2450,6 @@ static bool fold_setcond2(OptContext *ctx, TCGOp *op)
}
ctx->z_mask = 1;
- ctx->s_mask = smask_from_zmask(1);
return false;
do_setcond_const:
@@ -2649,21 +2643,18 @@ static bool fold_tcg_ld(OptContext *ctx, TCGOp *op)
break;
CASE_OP_32_64(ld8u):
ctx->z_mask = MAKE_64BIT_MASK(0, 8);
- ctx->s_mask = MAKE_64BIT_MASK(9, 55);
break;
CASE_OP_32_64(ld16s):
ctx->s_mask = MAKE_64BIT_MASK(16, 48);
break;
CASE_OP_32_64(ld16u):
ctx->z_mask = MAKE_64BIT_MASK(0, 16);
- ctx->s_mask = MAKE_64BIT_MASK(17, 47);
break;
case INDEX_op_ld32s_i64:
ctx->s_mask = MAKE_64BIT_MASK(32, 32);
break;
case INDEX_op_ld32u_i64:
ctx->z_mask = MAKE_64BIT_MASK(0, 32);
- ctx->s_mask = MAKE_64BIT_MASK(33, 31);
break;
default:
g_assert_not_reached();
--
2.43.0
- [PULL 00/72] tcg patch queue, Richard Henderson, 2024/12/24
- [PULL 01/72] tests/tcg: Do not use inttypes.h in multiarch/system/memory.c, Richard Henderson, 2024/12/24
- [PULL 04/72] tcg/optimize: Split out fold_affected_mask, Richard Henderson, 2024/12/24
- [PULL 07/72] tcg/optimize: Augment s_mask from z_mask in fold_masks_zs,
Richard Henderson <=
- [PULL 09/72] tcg/optimize: Use finish_folding in fold_add, fold_add_vec, fold_addsub2, Richard Henderson, 2024/12/24
- [PULL 10/72] tcg/optimize: Introduce const value accessors for TempOptInfo, Richard Henderson, 2024/12/24
- [PULL 06/72] tcg/optimize: Split out fold_masks_zs, Richard Henderson, 2024/12/24
- [PULL 12/72] tcg/optimize: Use fold_masks_zs in fold_andc, Richard Henderson, 2024/12/24
- [PULL 14/72] tcg/optimize: Use fold_masks_zs in fold_count_zeros, Richard Henderson, 2024/12/24
- [PULL 16/72] tcg/optimize: Use fold_and and fold_masks_z in fold_deposit, Richard Henderson, 2024/12/24
- [PULL 19/72] tcg/optimize: Use finish_folding in fold_dup, fold_dup2, Richard Henderson, 2024/12/24
- [PULL 25/72] tcg/optimize: Use fold_masks_zs in fold_movcond, Richard Henderson, 2024/12/24
- [PULL 03/72] tcg/optimize: Split out finish_bb, finish_ebb, Richard Henderson, 2024/12/24
- [PULL 08/72] tcg/optimize: Change representation of s_mask, Richard Henderson, 2024/12/24