vpopcnt{d,q} supports ptr_b
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 4969921..0379e2a 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -348,20 +348,20 @@
{ 0xB5, "vpmadd52huq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
{ 0x70, "vpshldw", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, true },
- { 0x71, "vpshldd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z, true },
- { 0x71, "vpshldq", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, true },
+ { 0x71, "vpshldd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, true },
+ { 0x71, "vpshldq", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_B64, true },
{ 0x70, "vpshldvw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false },
- { 0x71, "vpshldvd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z, false },
- { 0x71, "vpshldvq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false },
+ { 0x71, "vpshldvd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
+ { 0x71, "vpshldvq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_B64, false },
{ 0x72, "vpshrdw", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, true },
- { 0x73, "vpshrdd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z, true },
- { 0x73, "vpshrdq", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, true },
+ { 0x73, "vpshrdd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, true },
+ { 0x73, "vpshrdq", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_B64, true },
{ 0x72, "vpshrdvw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false },
- { 0x73, "vpshrdvd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z, false },
- { 0x73, "vpshrdvq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false },
+ { 0x73, "vpshrdvd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
+ { 0x73, "vpshrdvq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_B64, false },
{ 0x50, "vpdpbusd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
{ 0x51, "vpdpbusds", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
@@ -651,8 +651,8 @@
{ 0x54, "vpopcntb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z, false },
{ 0x54, "vpopcntw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false },
- { 0x55, "vpopcntd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z, false },
- { 0x55, "vpopcntq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false },
+ { 0x55, "vpopcntd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
+ { 0x55, "vpopcntq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_B64, false },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
diff --git a/test/misc.cpp b/test/misc.cpp
index a786690..afa3cfd 100644
--- a/test/misc.cpp
+++ b/test/misc.cpp
@@ -325,6 +325,22 @@
vpshrdvq(xmm5|k3|T_z, xmm2, ptr [rax + 0x40]);
vpshrdvq(ymm5|k3|T_z, ymm2, ptr [rax + 0x40]);
vpshrdvq(zmm5|k3|T_z, zmm2, ptr [rax + 0x40]);
+
+ vpshrdd(xmm5|k3|T_z, xmm2, ptr_b [rax + 0x40], 5);
+ vpshrdd(ymm5|k3|T_z, ymm2, ptr_b [rax + 0x40], 5);
+ vpshrdd(zmm5|k3|T_z, zmm2, ptr_b [rax + 0x40], 5);
+
+ vpshrdq(xmm5|k3|T_z, xmm2, ptr_b [rax + 0x40], 5);
+ vpshrdq(ymm5|k3|T_z, ymm2, ptr_b [rax + 0x40], 5);
+ vpshrdq(zmm5|k3|T_z, zmm2, ptr_b [rax + 0x40], 5);
+
+ vpshrdvd(xmm5|k3|T_z, xmm2, ptr_b [rax + 0x40]);
+ vpshrdvd(ymm5|k3|T_z, ymm2, ptr_b [rax + 0x40]);
+ vpshrdvd(zmm5|k3|T_z, zmm2, ptr_b [rax + 0x40]);
+
+ vpshrdvq(xmm5|k3|T_z, xmm2, ptr_b [rax + 0x40]);
+ vpshrdvq(ymm5|k3|T_z, ymm2, ptr_b [rax + 0x40]);
+ vpshrdvq(zmm5|k3|T_z, zmm2, ptr_b [rax + 0x40]);
}
} c;
const uint8_t tbl[] = {
@@ -351,6 +367,22 @@
0x62, 0xf2, 0xed, 0x8b, 0x73, 0x68, 0x04,
0x62, 0xf2, 0xed, 0xab, 0x73, 0x68, 0x02,
0x62, 0xf2, 0xed, 0xcb, 0x73, 0x68, 0x01,
+
+ 0x62, 0xf3, 0x6d, 0x9b, 0x73, 0x68, 0x10, 0x05,
+ 0x62, 0xf3, 0x6d, 0xbb, 0x73, 0x68, 0x10, 0x05,
+ 0x62, 0xf3, 0x6d, 0xdb, 0x73, 0x68, 0x10, 0x05,
+
+ 0x62, 0xf3, 0xed, 0x9b, 0x73, 0x68, 0x08, 0x05,
+ 0x62, 0xf3, 0xed, 0xbb, 0x73, 0x68, 0x08, 0x05,
+ 0x62, 0xf3, 0xed, 0xdb, 0x73, 0x68, 0x08, 0x05,
+
+ 0x62, 0xf2, 0x6d, 0x9b, 0x73, 0x68, 0x10,
+ 0x62, 0xf2, 0x6d, 0xbb, 0x73, 0x68, 0x10,
+ 0x62, 0xf2, 0x6d, 0xdb, 0x73, 0x68, 0x10,
+
+ 0x62, 0xf2, 0xed, 0x9b, 0x73, 0x68, 0x08,
+ 0x62, 0xf2, 0xed, 0xbb, 0x73, 0x68, 0x08,
+ 0x62, 0xf2, 0xed, 0xdb, 0x73, 0x68, 0x08,
};
const size_t n = sizeof(tbl) / sizeof(tbl[0]);
CYBOZU_TEST_EQUAL(c.getSize(), n);
@@ -373,9 +405,17 @@
vpopcntd(ymm5|k3|T_z, ptr [rax + 0x40]);
vpopcntd(zmm5|k3|T_z, ptr [rax + 0x40]);
+ vpopcntd(xmm5|k3|T_z, ptr_b [rax + 0x40]);
+ vpopcntd(ymm5|k3|T_z, ptr_b [rax + 0x40]);
+ vpopcntd(zmm5|k3|T_z, ptr_b [rax + 0x40]);
+
vpopcntq(xmm5|k3|T_z, ptr [rax + 0x40]);
vpopcntq(ymm5|k3|T_z, ptr [rax + 0x40]);
vpopcntq(zmm5|k3|T_z, ptr [rax + 0x40]);
+
+ vpopcntq(xmm5|k3|T_z, ptr_b [rax + 0x40]);
+ vpopcntq(ymm5|k3|T_z, ptr_b [rax + 0x40]);
+ vpopcntq(zmm5|k3|T_z, ptr_b [rax + 0x40]);
}
} c;
const uint8_t tbl[] = {
@@ -391,9 +431,17 @@
0x62, 0xf2, 0x7d, 0xab, 0x55, 0x68, 0x02,
0x62, 0xf2, 0x7d, 0xcb, 0x55, 0x68, 0x01,
+ 0x62, 0xf2, 0x7d, 0x9b, 0x55, 0x68, 0x10,
+ 0x62, 0xf2, 0x7d, 0xbb, 0x55, 0x68, 0x10,
+ 0x62, 0xf2, 0x7d, 0xdb, 0x55, 0x68, 0x10,
+
0x62, 0xf2, 0xfd, 0x8b, 0x55, 0x68, 0x04,
0x62, 0xf2, 0xfd, 0xab, 0x55, 0x68, 0x02,
0x62, 0xf2, 0xfd, 0xcb, 0x55, 0x68, 0x01,
+
+ 0x62, 0xf2, 0xfd, 0x9b, 0x55, 0x68, 0x08,
+ 0x62, 0xf2, 0xfd, 0xbb, 0x55, 0x68, 0x08,
+ 0x62, 0xf2, 0xfd, 0xdb, 0x55, 0x68, 0x08,
};
const size_t n = sizeof(tbl) / sizeof(tbl[0]);
CYBOZU_TEST_EQUAL(c.getSize(), n);
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 861ba2b..5ccd7a1 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1848,8 +1848,8 @@
void vpmullq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x40); }
void vpmultishiftqb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x83); }
void vpopcntb(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x54); }
-void vpopcntd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x55); }
-void vpopcntq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x55); }
+void vpopcntd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x55); }
+void vpopcntq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x55); }
void vpopcntw(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x54); }
void vpord(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0xEB); }
void vporq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0xEB); }
@@ -1865,16 +1865,16 @@
void vpscatterdq(const Address& addr, const Xmm& x) { opGather2(x, addr, T_N8 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_M_K, 0xA0, 1); }
void vpscatterqd(const Address& addr, const Xmm& x) { opGather2(x, addr, T_N4 | T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0xA1, 2); }
void vpscatterqq(const Address& addr, const Xmm& x) { opGather2(x, addr, T_N8 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_M_K, 0xA1, 0); }
-void vpshldd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x71, imm); }
-void vpshldq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x71, imm); }
-void vpshldvd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x71); }
-void vpshldvq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x71); }
+void vpshldd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x71, imm); }
+void vpshldq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x71, imm); }
+void vpshldvd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x71); }
+void vpshldvq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x71); }
void vpshldvw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x70); }
void vpshldw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x70, imm); }
-void vpshrdd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x73, imm); }
-void vpshrdq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x73, imm); }
-void vpshrdvd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x73); }
-void vpshrdvq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x73); }
+void vpshrdd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x73, imm); }
+void vpshrdq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x73, imm); }
+void vpshrdvd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x73); }
+void vpshrdvq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x73); }
void vpshrdvw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x72); }
void vpshrdw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x72, imm); }
void vpsllvw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x12); }
