add vf[,n]m[add,sub][132,213,231]nebf16

4 files changed, 93 insertions, 4 deletions

diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index ff1ba30..b1bf0b1 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -959,6 +959,22 @@ void putAVX10_BF16()
 		{ "vmulnepbf16", T_66 | T_MAP5 | T_EW0 | T_YMM | T_B16, 0x59 },
 		{ "vscalefpbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0x2C },
 		{ "vsubnepbf16", T_66 | T_MAP5 | T_EW0 | T_YMM | T_B16, 0x5C },
+
+		{ "vfmadd132nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0x98 },
+		{ "vfmadd213nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0xA8 },
+		{ "vfmadd231nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0xB8 },
+
+		{ "vfnmadd132nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0x9C },
+		{ "vfnmadd213nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0xAC },
+		{ "vfnmadd231nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0xBC },
+
+		{ "vfmsub132nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0x9A },
+		{ "vfmsub213nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0xAA },
+		{ "vfmsub231nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0xBA },
+
+		{ "vfnmsub132nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0x9E },
+		{ "vfnmsub213nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0xAE },
+		{ "vfnmsub231nepbf16", T_MAP6 | T_EW0 | T_YMM | T_B16, 0xBE },
 	};
 	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 		const xxopTbl& p = tbl[i];
diff --git a/test/avx10/bf16.txt b/test/avx10/bf16.txt
index d8f4c5a..7dcdb25 100644
--- a/test/avx10/bf16.txt
+++ b/test/avx10/bf16.txt
@@ -32,3 +32,63 @@ vsubnepbf16(xm1, xm2, xm3);
 vsubnepbf16(ym1|k1, ym2, ptr[rax+128]);
 vsubnepbf16(ym1|k1, ym2, ptr_b[rax+128]);
 vsubnepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+// madd
+vfmadd132nepbf16(xm1, xm2, xm3);
+vfmadd132nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfmadd132nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfmadd132nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vfmadd213nepbf16(xm1, xm2, xm3);
+vfmadd213nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfmadd213nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfmadd213nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vfmadd231nepbf16(xm1, xm2, xm3);
+vfmadd231nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfmadd231nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfmadd231nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+// nmadd
+vfnmadd132nepbf16(xm1, xm2, xm3);
+vfnmadd132nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfnmadd132nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfnmadd132nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vfnmadd213nepbf16(xm1, xm2, xm3);
+vfnmadd213nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfnmadd213nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfnmadd213nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vfnmadd231nepbf16(xm1, xm2, xm3);
+vfnmadd231nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfnmadd231nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfnmadd231nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+// msub
+vfmsub132nepbf16(xm1, xm2, xm3);
+vfmsub132nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfmsub132nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfmsub132nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vfmsub213nepbf16(xm1, xm2, xm3);
+vfmsub213nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfmsub213nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfmsub213nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vfmsub231nepbf16(xm1, xm2, xm3);
+vfmsub231nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfmsub231nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfmsub231nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+// nmsub
+vfnmsub132nepbf16(xm1, xm2, xm3);
+vfnmsub132nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfnmsub132nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfnmsub132nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vfnmsub213nepbf16(xm1, xm2, xm3);
+vfnmsub213nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfnmsub213nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfnmsub213nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vfnmsub231nepbf16(xm1, xm2, xm3);
+vfnmsub231nepbf16(ym1|k1, ym2, ptr[rax+128]);
+vfnmsub231nepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vfnmsub231nepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
diff --git a/test/test_by_xed.py b/test/test_by_xed.py
index cd6b7bb..5b84995 100644
--- a/test/test_by_xed.py
+++ b/test/test_by_xed.py
@@ -287,17 +287,18 @@ def removeExtraInfo(s):
 def run(cppText, xedText):
   cpp = loadFile(cppText)
   xed = loadFile(xedText)
-  if len(cpp) != len(xed):
-    raise Exception(f'different line {len(cpp)} {len(xed)}')
+  n = len(cpp)
+  if n != len(xed):
+    raise Exception(f'different line {n} {len(xed)}')
 
-  for i in range(len(cpp)):
+  for i in range(n):
     line1 = cpp[i]
     line2 = removeExtraInfo(xed[i])
     m1 = parseNmemonic(line1)
     m2 = parseNmemonic(line2)
 
     assertEqual(m1, m2, f'{i+1}')
-  print('run ok')
+  print('run ok', n)
 
 def assertEqualStr(a, b, msg=None):
   if str(a) != str(b):
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index f6cafef..7ce61e0 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -2281,36 +2281,48 @@ void vfixupimmpd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) {
 void vfixupimmps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_66|T_0F3A|T_EW0|T_YMM|T_SAE_Z|T_MUST_EVEX|T_B32, 0x54, imm); }
 void vfixupimmsd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_N8|T_66|T_0F3A|T_EW1|T_SAE_Z|T_MUST_EVEX, 0x55, imm); }
 void vfixupimmss(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_N4|T_66|T_0F3A|T_EW0|T_SAE_Z|T_MUST_EVEX, 0x55, imm); }
+void vfmadd132nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x98); }
 void vfmadd132ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x98); }
 void vfmadd132sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0x99); }
+void vfmadd213nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0xA8); }
 void vfmadd213ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xA8); }
 void vfmadd213sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0xA9); }
+void vfmadd231nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0xB8); }
 void vfmadd231ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xB8); }
 void vfmadd231sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0xB9); }
 void vfmaddcph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B32, 0x56); }
 void vfmaddsub132ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x96); }
 void vfmaddsub213ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xA6); }
 void vfmaddsub231ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xB6); }
+void vfmsub132nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x9A); }
 void vfmsub132ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x9A); }
 void vfmsub132sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0x9B); }
+void vfmsub213nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0xAA); }
 void vfmsub213ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xAA); }
 void vfmsub213sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0xAB); }
+void vfmsub231nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0xBA); }
 void vfmsub231ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xBA); }
 void vfmsub231sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0xBB); }
 void vfmsubadd132ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x97); }
 void vfmsubadd213ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xA7); }
 void vfmsubadd231ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xB7); }
 void vfmulcph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B32, 0xD6); }
+void vfnmadd132nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x9C); }
 void vfnmadd132ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x9C); }
 void vfnmadd132sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0x9D); }
+void vfnmadd213nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0xAC); }
 void vfnmadd213ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xAC); }
 void vfnmadd213sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0xAD); }
+void vfnmadd231nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0xBC); }
 void vfnmadd231ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xBC); }
 void vfnmadd231sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0xBD); }
+void vfnmsub132nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x9E); }
 void vfnmsub132ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x9E); }
 void vfnmsub132sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0x9F); }
+void vfnmsub213nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0xAE); }
 void vfnmsub213ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xAE); }
 void vfnmsub213sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0xAF); }
+void vfnmsub231nepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0xBE); }
 void vfnmsub231ph(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP6|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0xBE); }
 void vfnmsub231sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_66|T_MAP6|T_EW0|T_ER_X|T_MUST_EVEX, 0xBF); }
 void vfpclasspd(const Opmask& k, const Operand& op, uint8_t imm) { if (!op.isBit(128|256|512)) XBYAK_THROW(ERR_BAD_MEM_SIZE) opVex(k.changeBit(op.getBit()), 0, op, T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, 0x66, imm); }