diff mbox series

[PULL,063/159] tcg: Do not require both rotr and rotl from the backend

Message ID 20250425215454.886111-64-richard.henderson@linaro.org
State New
Headers show
Series [PULL,001/159] tcg/loongarch64: Fix vec_val computation in tcg_target_const_match | expand

Commit Message

Richard Henderson April 25, 2025, 9:53 p.m. UTC 
Many host architectures do not implement both rotate right
and rotate left and require the compiler to negate the
shift count to rotate the opposite direction.  We have been
requiring the backend to perform this transformation.
Do this during opcode expansion so that the next patch
can drop support where possible in the backend.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg-op.c | 98 +++++++++++++++++++++++++++++-----------------------
 1 file changed, 54 insertions(+), 44 deletions(-)
diff mbox series

Patch

diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c
index 43848ebc4f..8c8b9d179b 100644
--- a/tcg/tcg-op.c
+++ b/tcg/tcg-op.c
@@ -829,15 +829,18 @@  void tcg_gen_ctpop_i32(TCGv_i32 ret, TCGv_i32 arg1)
 
 void tcg_gen_rotl_i32(TCGv_i32 ret, TCGv_i32 arg1, TCGv_i32 arg2)
 {
-    if (TCG_TARGET_HAS_rot_i32) {
+    if (tcg_op_supported(INDEX_op_rotl_i32, TCG_TYPE_I32, 0)) {
         tcg_gen_op3_i32(INDEX_op_rotl_i32, ret, arg1, arg2);
+    } else if (tcg_op_supported(INDEX_op_rotr_i32, TCG_TYPE_I32, 0)) {
+        TCGv_i32 t0 = tcg_temp_ebb_new_i32();
+        tcg_gen_neg_i32(t0, arg2);
+        tcg_gen_op3_i32(INDEX_op_rotr_i32, ret, arg1, t0);
+        tcg_temp_free_i32(t0);
     } else {
-        TCGv_i32 t0, t1;
-
-        t0 = tcg_temp_ebb_new_i32();
-        t1 = tcg_temp_ebb_new_i32();
+        TCGv_i32 t0 = tcg_temp_ebb_new_i32();
+        TCGv_i32 t1 = tcg_temp_ebb_new_i32();
         tcg_gen_shl_i32(t0, arg1, arg2);
-        tcg_gen_subfi_i32(t1, 32, arg2);
+        tcg_gen_neg_i32(t1, arg2);
         tcg_gen_shr_i32(t1, arg1, t1);
         tcg_gen_or_i32(ret, t0, t1);
         tcg_temp_free_i32(t0);
@@ -851,12 +854,15 @@  void tcg_gen_rotli_i32(TCGv_i32 ret, TCGv_i32 arg1, int32_t arg2)
     /* some cases can be optimized here */
     if (arg2 == 0) {
         tcg_gen_mov_i32(ret, arg1);
-    } else if (TCG_TARGET_HAS_rot_i32) {
-        tcg_gen_rotl_i32(ret, arg1, tcg_constant_i32(arg2));
+    } else if (tcg_op_supported(INDEX_op_rotl_i32, TCG_TYPE_I32, 0)) {
+        TCGv_i32 t0 = tcg_constant_i32(arg2);
+        tcg_gen_op3_i32(INDEX_op_rotl_i32, ret, arg1, t0);
+    } else if (tcg_op_supported(INDEX_op_rotr_i32, TCG_TYPE_I32, 0)) {
+        TCGv_i32 t0 = tcg_constant_i32(32 - arg2);
+        tcg_gen_op3_i32(INDEX_op_rotr_i32, ret, arg1, t0);
     } else {
-        TCGv_i32 t0, t1;
-        t0 = tcg_temp_ebb_new_i32();
-        t1 = tcg_temp_ebb_new_i32();
+        TCGv_i32 t0 = tcg_temp_ebb_new_i32();
+        TCGv_i32 t1 = tcg_temp_ebb_new_i32();
         tcg_gen_shli_i32(t0, arg1, arg2);
         tcg_gen_shri_i32(t1, arg1, 32 - arg2);
         tcg_gen_or_i32(ret, t0, t1);
@@ -867,15 +873,18 @@  void tcg_gen_rotli_i32(TCGv_i32 ret, TCGv_i32 arg1, int32_t arg2)
 
 void tcg_gen_rotr_i32(TCGv_i32 ret, TCGv_i32 arg1, TCGv_i32 arg2)
 {
-    if (TCG_TARGET_HAS_rot_i32) {
+    if (tcg_op_supported(INDEX_op_rotr_i32, TCG_TYPE_I32, 0)) {
         tcg_gen_op3_i32(INDEX_op_rotr_i32, ret, arg1, arg2);
+    } else if (tcg_op_supported(INDEX_op_rotl_i32, TCG_TYPE_I32, 0)) {
+        TCGv_i32 t0 = tcg_temp_ebb_new_i32();
+        tcg_gen_neg_i32(t0, arg2);
+        tcg_gen_op3_i32(INDEX_op_rotl_i32, ret, arg1, t0);
+        tcg_temp_free_i32(t0);
     } else {
-        TCGv_i32 t0, t1;
-
-        t0 = tcg_temp_ebb_new_i32();
-        t1 = tcg_temp_ebb_new_i32();
+        TCGv_i32 t0 = tcg_temp_ebb_new_i32();
+        TCGv_i32 t1 = tcg_temp_ebb_new_i32();
         tcg_gen_shr_i32(t0, arg1, arg2);
-        tcg_gen_subfi_i32(t1, 32, arg2);
+        tcg_gen_neg_i32(t1, arg2);
         tcg_gen_shl_i32(t1, arg1, t1);
         tcg_gen_or_i32(ret, t0, t1);
         tcg_temp_free_i32(t0);
@@ -886,12 +895,7 @@  void tcg_gen_rotr_i32(TCGv_i32 ret, TCGv_i32 arg1, TCGv_i32 arg2)
 void tcg_gen_rotri_i32(TCGv_i32 ret, TCGv_i32 arg1, int32_t arg2)
 {
     tcg_debug_assert(arg2 >= 0 && arg2 < 32);
-    /* some cases can be optimized here */
-    if (arg2 == 0) {
-        tcg_gen_mov_i32(ret, arg1);
-    } else {
-        tcg_gen_rotli_i32(ret, arg1, 32 - arg2);
-    }
+    tcg_gen_rotli_i32(ret, arg1, -arg2 & 31);
 }
 
 void tcg_gen_deposit_i32(TCGv_i32 ret, TCGv_i32 arg1, TCGv_i32 arg2,
@@ -2437,14 +2441,18 @@  void tcg_gen_ctpop_i64(TCGv_i64 ret, TCGv_i64 arg1)
 
 void tcg_gen_rotl_i64(TCGv_i64 ret, TCGv_i64 arg1, TCGv_i64 arg2)
 {
-    if (TCG_TARGET_HAS_rot_i64) {
+    if (tcg_op_supported(INDEX_op_rotl_i64, TCG_TYPE_I64, 0)) {
         tcg_gen_op3_i64(INDEX_op_rotl_i64, ret, arg1, arg2);
+    } else if (tcg_op_supported(INDEX_op_rotl_i64, TCG_TYPE_I64, 0)) {
+        TCGv_i64 t0 = tcg_temp_ebb_new_i64();
+        tcg_gen_neg_i64(t0, arg2);
+        tcg_gen_op3_i64(INDEX_op_rotr_i64, ret, arg1, t0);
+        tcg_temp_free_i64(t0);
     } else {
-        TCGv_i64 t0, t1;
-        t0 = tcg_temp_ebb_new_i64();
-        t1 = tcg_temp_ebb_new_i64();
+        TCGv_i64 t0 = tcg_temp_ebb_new_i64();
+        TCGv_i64 t1 = tcg_temp_ebb_new_i64();
         tcg_gen_shl_i64(t0, arg1, arg2);
-        tcg_gen_subfi_i64(t1, 64, arg2);
+        tcg_gen_neg_i64(t1, arg2);
         tcg_gen_shr_i64(t1, arg1, t1);
         tcg_gen_or_i64(ret, t0, t1);
         tcg_temp_free_i64(t0);
@@ -2458,12 +2466,15 @@  void tcg_gen_rotli_i64(TCGv_i64 ret, TCGv_i64 arg1, int64_t arg2)
     /* some cases can be optimized here */
     if (arg2 == 0) {
         tcg_gen_mov_i64(ret, arg1);
-    } else if (TCG_TARGET_HAS_rot_i64) {
-        tcg_gen_rotl_i64(ret, arg1, tcg_constant_i64(arg2));
+    } else if (tcg_op_supported(INDEX_op_rotl_i64, TCG_TYPE_I64, 0)) {
+        TCGv_i64 t0 = tcg_constant_i64(arg2);
+        tcg_gen_op3_i64(INDEX_op_rotl_i64, ret, arg1, t0);
+    } else if (tcg_op_supported(INDEX_op_rotr_i64, TCG_TYPE_I64, 0)) {
+        TCGv_i64 t0 = tcg_constant_i64(64 - arg2);
+        tcg_gen_op3_i64(INDEX_op_rotr_i64, ret, arg1, t0);
     } else {
-        TCGv_i64 t0, t1;
-        t0 = tcg_temp_ebb_new_i64();
-        t1 = tcg_temp_ebb_new_i64();
+        TCGv_i64 t0 = tcg_temp_ebb_new_i64();
+        TCGv_i64 t1 = tcg_temp_ebb_new_i64();
         tcg_gen_shli_i64(t0, arg1, arg2);
         tcg_gen_shri_i64(t1, arg1, 64 - arg2);
         tcg_gen_or_i64(ret, t0, t1);
@@ -2474,14 +2485,18 @@  void tcg_gen_rotli_i64(TCGv_i64 ret, TCGv_i64 arg1, int64_t arg2)
 
 void tcg_gen_rotr_i64(TCGv_i64 ret, TCGv_i64 arg1, TCGv_i64 arg2)
 {
-    if (TCG_TARGET_HAS_rot_i64) {
+    if (tcg_op_supported(INDEX_op_rotr_i64, TCG_TYPE_I64, 0)) {
         tcg_gen_op3_i64(INDEX_op_rotr_i64, ret, arg1, arg2);
+    } else if (tcg_op_supported(INDEX_op_rotl_i64, TCG_TYPE_I64, 0)) {
+        TCGv_i64 t0 = tcg_temp_ebb_new_i64();
+        tcg_gen_neg_i64(t0, arg2);
+        tcg_gen_op3_i64(INDEX_op_rotl_i64, ret, arg1, t0);
+        tcg_temp_free_i64(t0);
     } else {
-        TCGv_i64 t0, t1;
-        t0 = tcg_temp_ebb_new_i64();
-        t1 = tcg_temp_ebb_new_i64();
+        TCGv_i64 t0 = tcg_temp_ebb_new_i64();
+        TCGv_i64 t1 = tcg_temp_ebb_new_i64();
         tcg_gen_shr_i64(t0, arg1, arg2);
-        tcg_gen_subfi_i64(t1, 64, arg2);
+        tcg_gen_neg_i64(t1, arg2);
         tcg_gen_shl_i64(t1, arg1, t1);
         tcg_gen_or_i64(ret, t0, t1);
         tcg_temp_free_i64(t0);
@@ -2492,12 +2507,7 @@  void tcg_gen_rotr_i64(TCGv_i64 ret, TCGv_i64 arg1, TCGv_i64 arg2)
 void tcg_gen_rotri_i64(TCGv_i64 ret, TCGv_i64 arg1, int64_t arg2)
 {
     tcg_debug_assert(arg2 >= 0 && arg2 < 64);
-    /* some cases can be optimized here */
-    if (arg2 == 0) {
-        tcg_gen_mov_i64(ret, arg1);
-    } else {
-        tcg_gen_rotli_i64(ret, arg1, 64 - arg2);
-    }
+    tcg_gen_rotli_i64(ret, arg1, -arg2 & 63);
 }
 
 void tcg_gen_deposit_i64(TCGv_i64 ret, TCGv_i64 arg1, TCGv_i64 arg2,