[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PATCH v2 01/65] tcg: Improve expansion of deposit of constant
|
From: |
Richard Henderson |
|
Subject: |
[PATCH v2 01/65] tcg: Improve expansion of deposit of constant |
|
Date: |
Fri, 20 Oct 2023 13:42:27 -0700 |
The extract2 expansion is too difficult for the optimizer to
simplify. If we have an immediate input, use and+or instead,
skipping the and if the field becomes all 1's.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
tcg/tcg-op.c | 28 ++++++++++++++++++++++++++++
1 file changed, 28 insertions(+)
diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c
index 393dbcd01c..2ef4b866e2 100644
--- a/tcg/tcg-op.c
+++ b/tcg/tcg-op.c
@@ -602,6 +602,7 @@ void tcg_gen_deposit_i32(TCGv_i32 ret, TCGv_i32 arg1,
TCGv_i32 arg2,
{
uint32_t mask;
TCGv_i32 t1;
+ TCGTemp *ts;
tcg_debug_assert(ofs < 32);
tcg_debug_assert(len > 0);
@@ -617,6 +618,19 @@ void tcg_gen_deposit_i32(TCGv_i32 ret, TCGv_i32 arg1,
TCGv_i32 arg2,
return;
}
+ /* Deposit of a constant into a value. */
+ ts = tcgv_i32_temp(arg2);
+ if (ts->kind == TEMP_CONST) {
+ uint32_t mask0 = deposit32(-1, ofs, len, 0);
+ uint32_t maski = deposit32(0, ofs, len, ts->val);
+
+ if (mask0 != ~maski) {
+ tcg_gen_andi_i32(ret, arg1, mask0);
+ }
+ tcg_gen_ori_i32(ret, ret, maski);
+ return;
+ }
+
t1 = tcg_temp_ebb_new_i32();
if (TCG_TARGET_HAS_extract2_i32) {
@@ -2217,6 +2231,7 @@ void tcg_gen_deposit_i64(TCGv_i64 ret, TCGv_i64 arg1,
TCGv_i64 arg2,
{
uint64_t mask;
TCGv_i64 t1;
+ TCGTemp *ts;
tcg_debug_assert(ofs < 64);
tcg_debug_assert(len > 0);
@@ -2232,6 +2247,19 @@ void tcg_gen_deposit_i64(TCGv_i64 ret, TCGv_i64 arg1,
TCGv_i64 arg2,
return;
}
+ /* Deposit of a constant into a value. */
+ ts = tcgv_i64_temp(arg2);
+ if (ts->kind == TEMP_CONST) {
+ uint64_t mask0 = deposit64(-1, ofs, len, 0);
+ uint64_t maski = deposit64(0, ofs, len, ts->val);
+
+ if (mask0 != ~maski) {
+ tcg_gen_andi_i64(ret, arg1, mask0);
+ }
+ tcg_gen_ori_i64(ret, ret, maski);
+ return;
+ }
+
if (TCG_TARGET_REG_BITS == 32) {
if (ofs >= 32) {
tcg_gen_deposit_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1),
--
2.34.1
- [PATCH v2 00/65] target/hppa: Implement hppa64-cpu, Richard Henderson, 2023/10/20
- [PATCH v2 01/65] tcg: Improve expansion of deposit of constant,
Richard Henderson <=
- [PATCH v2 03/65] target/hppa: Remove get_temp, Richard Henderson, 2023/10/20
- [PATCH v2 02/65] tcg: Improve expansion of deposit into a constant, Richard Henderson, 2023/10/20
- [PATCH v2 04/65] target/hppa: Remove get_temp_tl, Richard Henderson, 2023/10/20
- [PATCH v2 10/65] target/hppa: Fix do_add, do_sub for hppa64, Richard Henderson, 2023/10/20
- [PATCH v2 17/65] target/hppa: Update cpu_hppa_get/put_psw for hppa64, Richard Henderson, 2023/10/20
- [PATCH v2 15/65] target/hppa: Implement cpu_list, Richard Henderson, 2023/10/20
- [PATCH v2 18/65] target/hppa: Handle absolute addresses for pa2.0, Richard Henderson, 2023/10/20
- [PATCH v2 07/65] target/hppa: Fix load in do_load_32, Richard Henderson, 2023/10/20
- [PATCH v2 08/65] target/hppa: Truncate rotate count in trans_shrpw_sar, Richard Henderson, 2023/10/20
- [PATCH v2 09/65] target/hppa: Fix trans_ds for hppa64, Richard Henderson, 2023/10/20