add type of w(x, x, op) in avx10 bf16

3 files changed, 46 insertions, 2 deletions

diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 23923b0..ff1ba30 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -946,18 +946,26 @@ void putFP16_2()
 
 void putAVX10_BF16()
 {
-	const struct Tbl {
+	// x, x, op
+	const struct xxopTbl {
 		const char *name;
 		uint64_t type;
 		uint8_t code;
 	} tbl[] = {
 		{ "vaddnepbf16", T_66 | T_MAP5 | T_EW0 | T_YMM | T_B16, 0x58 },
+		{ "vdivnepbf16", T_66 | T_MAP5 | T_EW0 | T_YMM | T_B16, 0x5E },
+		{ "vmaxpbf16", T_66 | T_MAP5 | T_EW0 | T_YMM | T_B16, 0x5F },
+		{ "vminpbf16", T_66 | T_MAP5 | T_EW0 | T_YMM | T_B16, 0x5D },
+		{ "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 },
 	};
 	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
-		const Tbl& p = tbl[i];
+		const xxopTbl& p = tbl[i];
 		std::string s = type2String(p.type | T_MUST_EVEX);
 		printf("void %s(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, %s, 0x%2X); }\n" , p.name, s.c_str(), p.code);
 	}
+//		{ "vrcppbf16", T_66 | T_MAP6 | T_EW0 | T_YMM | T_B16, 0x4C },
 }
 
 void putFP16()
diff --git a/test/avx10/bf16.txt b/test/avx10/bf16.txt
index 1c77f93..d8f4c5a 100644
--- a/test/avx10/bf16.txt
+++ b/test/avx10/bf16.txt
@@ -2,3 +2,33 @@ vaddnepbf16(xm1, xm2, xm3);
 vaddnepbf16(ym1|k1, ym2, ptr[rax+128]);
 vaddnepbf16(ym1|k1, ym2, ptr_b[rax+128]);
 vaddnepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vdivnepbf16(xm1, xm2, xm3);
+vdivnepbf16(ym1|k1, ym2, ptr[rax+128]);
+vdivnepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vdivnepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vmaxpbf16(xm1, xm2, xm3);
+vmaxpbf16(ym1|k1, ym2, ptr[rax+128]);
+vmaxpbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vmaxpbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vminpbf16(xm1, xm2, xm3);
+vminpbf16(ym1|k1, ym2, ptr[rax+128]);
+vminpbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vminpbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vmulnepbf16(xm1, xm2, xm3);
+vmulnepbf16(ym1|k1, ym2, ptr[rax+128]);
+vmulnepbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vmulnepbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+vscalefpbf16(xm1, xm2, xm3);
+vscalefpbf16(ym1|k1, ym2, ptr[rax+128]);
+vscalefpbf16(ym1|k1, ym2, ptr_b[rax+128]);
+vscalefpbf16(zm1|k2|T_z, zm2, ptr_b[rax+128]);
+
+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]);
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index f98e001..f6cafef 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -2259,6 +2259,7 @@ void vcvtusi2ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opCvt3(x1, x2
 void vcvtuw2ph(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F2|T_MAP5|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x7D); }
 void vcvtw2ph(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F3|T_MAP5|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x7D); }
 void vdbpsadbw(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_MUST_EVEX, 0x42, imm); }
+void vdivnepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP5|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x5E); }
 void vdivph(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_EW0 | T_YMM | T_MUST_EVEX | T_ER_Z | T_B16, 0x5E); }
 void vdivsh(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_F3 | T_EW0 | T_MUST_EVEX | T_ER_X | T_N2, 0x5E); }
 void vdpbf16ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3|T_0F38|T_EW0|T_YMM|T_SAE_Z|T_MUST_EVEX|T_B32, 0x52); }
@@ -2350,8 +2351,10 @@ void vinserti32x4(const Ymm& r1, const Ymm& r2, const Operand& op, uint8_t imm)
 void vinserti32x8(const Zmm& r1, const Zmm& r2, const Operand& op, uint8_t imm) {if (!op.is(Operand::MEM | Operand::YMM)) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(r1, &r2, op, T_N32|T_66|T_0F3A|T_EW0|T_YMM|T_MUST_EVEX, 0x3A, imm); }
 void vinserti64x2(const Ymm& r1, const Ymm& r2, const Operand& op, uint8_t imm) {if (!(r1.getKind() == r2.getKind() && op.is(Operand::MEM | Operand::XMM))) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(r1, &r2, op, T_N16|T_66|T_0F3A|T_EW1|T_YMM|T_MUST_EVEX, 0x38, imm); }
 void vinserti64x4(const Zmm& r1, const Zmm& r2, const Operand& op, uint8_t imm) {if (!op.is(Operand::MEM | Operand::YMM)) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(r1, &r2, op, T_N32|T_66|T_0F3A|T_EW1|T_YMM|T_MUST_EVEX, 0x3A, imm); }
+void vmaxpbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP5|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x5F); }
 void vmaxph(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_EW0 | T_YMM | T_MUST_EVEX | T_SAE_Z | T_B16, 0x5F); }
 void vmaxsh(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_F3 | T_EW0 | T_MUST_EVEX | T_SAE_X | T_N2, 0x5F); }
+void vminpbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP5|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x5D); }
 void vminph(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_EW0 | T_YMM | T_MUST_EVEX | T_SAE_Z | T_B16, 0x5D); }
 void vminsh(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_F3 | T_EW0 | T_MUST_EVEX | T_SAE_X | T_N2, 0x5D); }
 void vmovdqa32(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_66|T_0F|T_EW0|T_YMM|T_ER_X|T_ER_Y|T_ER_Z|T_MUST_EVEX|T_M_K, 0x7F); }
@@ -2372,6 +2375,7 @@ void vmovsh(const Xmm& x1, const Xmm& x2, const Xmm& x3) { opAVX_X_X_XM(x1, x2,
 void vmovw(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_N2|T_66|T_MAP5|T_MUST_EVEX, 0x7E); }
 void vmovw(const Reg32e& r, const Xmm& x) { opAVX_X_X_XM(x, xm0, r, T_N2|T_66|T_MAP5|T_MUST_EVEX, 0x7E); }
 void vmovw(const Xmm& x, const Operand& op) { if (!op.isREG(32|64) && !op.isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION) opAVX_X_X_XM(x, xm0, op, T_N2|T_66|T_MAP5|T_MUST_EVEX, 0x6E); }
+void vmulnepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP5|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x59); }
 void vmulph(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_EW0 | T_YMM | T_MUST_EVEX | T_ER_Z | T_B16, 0x59); }
 void vmulsh(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_F3 | T_EW0 | T_MUST_EVEX | T_ER_X | T_N2, 0x59); }
 void vp2intersectd(const Opmask& k, const Xmm& x, const Operand& op) { if (k.getOpmaskIdx() != 0) XBYAK_THROW(ERR_OPMASK_IS_ALREADY_SET) opAVX_K_X_XM(k, x, op, T_F2 | T_0F38 | T_YMM | T_EVEX | T_EW0 | T_B32, 0x68); }
@@ -2555,6 +2559,7 @@ void vrsqrt28sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(
 void vrsqrt28ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N4|T_66|T_0F38|T_EW0|T_SAE_X|T_MUST_EVEX, 0xCD); }
 void vrsqrtph(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66|T_MAP6|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x4E); }
 void vrsqrtsh(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_MUST_EVEX, 0x4F); }
+void vscalefpbf16(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, 0x2C); }
 void vscalefpd(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_ER_Z|T_MUST_EVEX|T_B64, 0x2C); }
 void vscalefph(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, 0x2C); }
 void vscalefps(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_ER_Z|T_MUST_EVEX|T_B32, 0x2C); }
@@ -2579,6 +2584,7 @@ void vshufi32x4(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) {
 void vshufi64x2(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, 0x43, imm); }
 void vsqrtph(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_MAP5|T_EW0|T_YMM|T_ER_Z|T_MUST_EVEX|T_B16, 0x51); }
 void vsqrtsh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2|T_F3|T_MAP5|T_EW0|T_ER_X|T_MUST_EVEX, 0x51); }
+void vsubnepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66|T_MAP5|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x5C); }
 void vsubph(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_EW0 | T_YMM | T_MUST_EVEX | T_ER_Z | T_B16, 0x5C); }
 void vsubsh(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_F3 | T_EW0 | T_MUST_EVEX | T_ER_X | T_N2, 0x5C); }
 void vucomish(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_MAP5 | T_MUST_EVEX | T_EW0 | T_SAE_X | T_N2, 0x2E); }