tcg: Add tcg_gen_revbit{8,32,64}
Add generic expanders for reversing bits within a word. Reviewed-by: Philippe Mathieu-Daudé <philmd@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Richard Henderson committed
Jul 30, 2026 at 10:55 UTC
c2ca3737400ebf9432bf14fc59db69538868466d
3 files changed
+88
include/tcg/tcg-op-common.h
+5
@@ -164,6 +164,8 @@ void tcg_gen_smax_i32(TCGv_i32, TCGv_i32 arg1, TCGv_i32 arg2);
164
void tcg_gen_umin_i32(TCGv_i32, TCGv_i32 arg1, TCGv_i32 arg2);
165
void tcg_gen_umax_i32(TCGv_i32, TCGv_i32 arg1, TCGv_i32 arg2);
166
void tcg_gen_abs_i32(TCGv_i32, TCGv_i32);
167
+void tcg_gen_revbit8_i32(TCGv_i32 ret, TCGv_i32 arg);
168
+void tcg_gen_revbit32_i32(TCGv_i32 ret, TCGv_i32 arg);
169
170
/* Replicate a value of size @vece from @in to all the lanes in @out */
171
void tcg_gen_dup_i32(unsigned vece, TCGv_i32 out, TCGv_i32 in);
@@ -275,6 +277,9 @@ void tcg_gen_smax_i64(TCGv_i64, TCGv_i64 arg1, TCGv_i64 arg2);
277
void tcg_gen_umin_i64(TCGv_i64, TCGv_i64 arg1, TCGv_i64 arg2);
278
void tcg_gen_umax_i64(TCGv_i64, TCGv_i64 arg1, TCGv_i64 arg2);
279
void tcg_gen_abs_i64(TCGv_i64, TCGv_i64);
280
+void tcg_gen_revbit8_i64(TCGv_i64 ret, TCGv_i64 arg);
281
+void tcg_gen_revbit32_i64(TCGv_i64 ret, TCGv_i64 arg, int flags);
282
+void tcg_gen_revbit64_i64(TCGv_i64 ret, TCGv_i64 arg);
283
284
/* Replicate a value of size @vece from @in to all the lanes in @out */
285
void tcg_gen_dup_i64(unsigned vece, TCGv_i64 out, TCGv_i64 in);
include/tcg/tcg-op.h
+7
@@ -115,6 +115,10 @@ typedef TCGv_i64 TCGv;
115
#define tcg_gen_bswap_tl tcg_gen_bswap64_i64
116
#define tcg_gen_hswap_tl tcg_gen_hswap_i64
117
#define tcg_gen_wswap_tl tcg_gen_wswap_i64
118
+#define tcg_gen_revbit8_tl tcg_gen_revbit8_i64
119
+#define tcg_gen_revbit32_tl tcg_gen_revbit32_i64
120
+#define tcg_gen_revbit64_tl tcg_gen_revbit64_i64
121
+#define tcg_gen_revbit_tl tcg_gen_revbit64_i64
122
#define tcg_gen_concat_tl_i64 tcg_gen_concat32_i64
123
#define tcg_gen_extr_i64_tl tcg_gen_extr32_i64
124
#define tcg_gen_andc_tl tcg_gen_andc_i64
@@ -234,6 +238,9 @@ typedef TCGv_i64 TCGv;
238
#define tcg_gen_bswap32_tl(D, S, F) tcg_gen_bswap32_i32(D, S)
239
#define tcg_gen_bswap_tl tcg_gen_bswap32_i32
240
#define tcg_gen_hswap_tl tcg_gen_hswap_i32
241
+#define tcg_gen_revbit8_tl tcg_gen_revbit8_i32
242
+#define tcg_gen_revbit32_tl(D, S, F) tcg_gen_revbit32_i32(D, S)
243
+#define tcg_gen_revbit_tl tcg_gen_revbit32_i32
244
#define tcg_gen_concat_tl_i64 tcg_gen_concat_i32_i64
245
#define tcg_gen_extr_i64_tl tcg_gen_extr_i64_i32
246
#define tcg_gen_andc_tl tcg_gen_andc_i32
tcg/tcg-op.c
+76
@@ -1160,6 +1160,45 @@ void tcg_gen_ext16u_i32(TCGv_i32 ret, TCGv_i32 arg)
1160
tcg_gen_extract_i32(ret, arg, 0, 16);
1161
}
1162
1163
+/*
1164
+ * Internal helper for bit and byte reversal.
1165
+ * Given a repeating matched block of 1's and 0's, swap the bits within
1166
+ * those two blocks. E.g. mask=00ff00ff, shift the input bits left and
1167
+ * right 8 bits.
1168
+ */
1169
+static void gen_bitswap_i32(TCGv_i32 ret, TCGv_i32 arg, uint32_t mask)
1170
+{
1171
+ TCGv_i32 t0 = tcg_temp_ebb_new_i32();
1172
+ TCGv_i32 t1 = tcg_temp_ebb_new_i32();
1173
+ int sh = cto32(mask);
1174
+
1175
+ tcg_gen_andi_i32(t0, arg, mask);
1176
+ tcg_gen_shri_i32(t1, arg, sh);
1177
+ tcg_gen_shli_i32(t0, t0, sh);
1178
+ tcg_gen_andi_i32(t1, t1, mask);
1179
+ tcg_gen_or_i32(ret, t0, t1);
1180
+
1181
+ tcg_temp_free_i32(t0);
1182
+ tcg_temp_free_i32(t1);
1183
+}
1184
+
1185
+/* Similarly for 64-bit operands. */
1186
+static void gen_bitswap_i64(TCGv_i64 ret, TCGv_i64 arg, uint64_t mask)
1187
+{
1188
+ TCGv_i64 t0 = tcg_temp_ebb_new_i64();
1189
+ TCGv_i64 t1 = tcg_temp_ebb_new_i64();
1190
+ int sh = cto64(mask);
1191
+
1192
+ tcg_gen_andi_i64(t0, arg, mask);
1193
+ tcg_gen_shri_i64(t1, arg, sh);
1194
+ tcg_gen_shli_i64(t0, t0, sh);
1195
+ tcg_gen_andi_i64(t1, t1, mask);
1196
+ tcg_gen_or_i64(ret, t0, t1);
1197
+
1198
+ tcg_temp_free_i64(t0);
1199
+ tcg_temp_free_i64(t1);
1200
+}
1201
+
1202
/*
1203
* bswap16_i32: 16-bit byte swap on the low bits of a 32-bit value.
1204
*
@@ -1244,6 +1283,19 @@ void tcg_gen_hswap_i32(TCGv_i32 ret, TCGv_i32 arg)
1283
tcg_gen_rotli_i32(ret, arg, 16);
1284
}
1285
1286
+void tcg_gen_revbit8_i32(TCGv_i32 ret, TCGv_i32 arg)
1287
+{
1288
+ gen_bitswap_i32(ret, arg, 0x55555555u);
1289
+ gen_bitswap_i32(ret, ret, 0x33333333u);
1290
+ gen_bitswap_i32(ret, ret, 0x0f0f0f0fu);
1291
+}
1292
+
1293
+void tcg_gen_revbit32_i32(TCGv_i32 ret, TCGv_i32 arg)
1294
+{
1295
+ tcg_gen_revbit8_i32(ret, arg);
1296
+ tcg_gen_bswap32_i32(ret, ret);
1297
+}
1298
+
1299
void tcg_gen_smin_i32(TCGv_i32 ret, TCGv_i32 a, TCGv_i32 b)
1300
{
1301
tcg_gen_movcond_i32(TCG_COND_LT, ret, a, b, a, b);
@@ -1869,6 +1921,30 @@ void tcg_gen_wswap_i64(TCGv_i64 ret, TCGv_i64 arg)
1921
tcg_gen_rotli_i64(ret, arg, 32);
1922
}
1923
1924
+void tcg_gen_revbit32_i64(TCGv_i64 ret, TCGv_i64 arg, int flags)
1925
+{
1926
+ /* Only one extension flag may be present. */
1927
+ tcg_debug_assert(!(flags & TCG_BSWAP_OS) || !(flags & TCG_BSWAP_OZ));
1928
+
1929
+ gen_bitswap_i64(ret, arg, 0x55555555ull);
1930
+ gen_bitswap_i64(ret, ret, 0x33333333ull);
1931
+ gen_bitswap_i64(ret, ret, 0x0f0f0f0full);
1932
+ tcg_gen_bswap32_i64(ret, ret, flags | TCG_BSWAP_IZ);
1933
+}
1934
+
1935
+void tcg_gen_revbit8_i64(TCGv_i64 ret, TCGv_i64 arg)
1936
+{
1937
+ gen_bitswap_i64(ret, arg, 0x5555555555555555ull);
1938
+ gen_bitswap_i64(ret, ret, 0x3333333333333333ull);
1939
+ gen_bitswap_i64(ret, ret, 0x0f0f0f0f0f0f0f0full);
1940
+}
1941
+
1942
+void tcg_gen_revbit64_i64(TCGv_i64 ret, TCGv_i64 arg)
1943
+{
1944
+ tcg_gen_revbit8_i64(ret, arg);
1945
+ tcg_gen_bswap64_i64(ret, ret);
1946
+}
1947
+
1948
void tcg_gen_not_i64(TCGv_i64 ret, TCGv_i64 arg)
1949
{
1950
if (tcg_op_supported(INDEX_op_not, TCG_TYPE_I64, 0)) {