mirrors/qemu
0
0
Fork 0
mirror of https://github.com/qemu/qemu.git synced 2024-11-25 20:03:37 +08:00
Code Issues Packages Projects Releases Wiki Activity

tcg/optimize: Fix folding of vector ops

Browse Source 
Bitwise operations are easy to fold, because the operation is
identical regardless of element size.  But add and sub need
extra element size info that is not currently propagated.

Fixes: 2f9f08ba43
Cc: qemu-stable@nongnu.org
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/799
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This commit is contained in:
Richard Henderson 2021-12-16 06:07:25 -08:00
parent 67e41fe0cf
commit c578ff1858
1 changed files with 38 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
tcg/optimize.c
View File

@ -308,13 +308,13 @@ static uint64_t do_constant_folding_2(TCGOpcode op, uint64_t x, uint64_t y)
CASE_OP_32_64(mul):
return x * y;
CASE_OP_32_64(and):
CASE_OP_32_64_VEC(and):
return x & y;
CASE_OP_32_64(or):
CASE_OP_32_64_VEC(or):
return x | y;
CASE_OP_32_64(xor):
CASE_OP_32_64_VEC(xor):
return x ^ y;
case INDEX_op_shl_i32:
@ -347,16 +347,16 @@ static uint64_t do_constant_folding_2(TCGOpcode op, uint64_t x, uint64_t y)
case INDEX_op_rotl_i64:
return rol64(x, y & 63);
CASE_OP_32_64(not):
CASE_OP_32_64_VEC(not):
return ~x;
CASE_OP_32_64(neg):
return -x;
CASE_OP_32_64(andc):
CASE_OP_32_64_VEC(andc):
return x & ~y;
CASE_OP_32_64(orc):
CASE_OP_32_64_VEC(orc):
return x | ~y;
CASE_OP_32_64(eqv):
@ -751,6 +751,12 @@ static bool fold_const2(OptContext *ctx, TCGOp *op)
return false;
}
static bool fold_commutative(OptContext *ctx, TCGOp *op)
{
swap_commutative(op->args[0], &op->args[1], &op->args[2]);
return false;
}
static bool fold_const2_commutative(OptContext *ctx, TCGOp *op)
{
swap_commutative(op->args[0], &op->args[1], &op->args[2]);
@ -905,6 +911,16 @@ static bool fold_add(OptContext *ctx, TCGOp *op)
return false;
}
/* We cannot as yet do_constant_folding with vectors. */
static bool fold_add_vec(OptContext *ctx, TCGOp *op)
{
if (fold_commutative(ctx, op) ||
fold_xi_to_x(ctx, op, 0)) {
return true;
}
return false;
}
static bool fold_addsub2(OptContext *ctx, TCGOp *op, bool add)
{
if (arg_is_const(op->args[2]) && arg_is_const(op->args[3]) &&
@ -1938,10 +1954,10 @@ static bool fold_sub_to_neg(OptContext *ctx, TCGOp *op)
return false;
}
static bool fold_sub(OptContext *ctx, TCGOp *op)
/* We cannot as yet do_constant_folding with vectors. */
static bool fold_sub_vec(OptContext *ctx, TCGOp *op)
{
if (fold_const2(ctx, op) ||
fold_xx_to_i(ctx, op, 0) ||
if (fold_xx_to_i(ctx, op, 0) ||
fold_xi_to_x(ctx, op, 0) ||
fold_sub_to_neg(ctx, op)) {
return true;
@ -1949,6 +1965,11 @@ static bool fold_sub(OptContext *ctx, TCGOp *op)
return false;
}
static bool fold_sub(OptContext *ctx, TCGOp *op)
{
return fold_const2(ctx, op) || fold_sub_vec(ctx, op);
}
static bool fold_sub2(OptContext *ctx, TCGOp *op)
{
return fold_addsub2(ctx, op, false);
@ -2052,9 +2073,12 @@ void tcg_optimize(TCGContext *s)
* Sorted alphabetically by opcode as much as possible.
*/
switch (opc) {
CASE_OP_32_64_VEC(add):
CASE_OP_32_64(add):
done = fold_add(&ctx, op);
break;
case INDEX_op_add_vec:
done = fold_add_vec(&ctx, op);
break;
CASE_OP_32_64(add2):
done = fold_add2(&ctx, op);
break;
@ -2193,9 +2217,12 @@ void tcg_optimize(TCGContext *s)
CASE_OP_32_64(sextract):
done = fold_sextract(&ctx, op);
break;
CASE_OP_32_64_VEC(sub):
CASE_OP_32_64(sub):
done = fold_sub(&ctx, op);
break;
case INDEX_op_sub_vec:
done = fold_sub_vec(&ctx, op);
break;
CASE_OP_32_64(sub2):
done = fold_sub2(&ctx, op);
break;