use opVex

3 files changed, 10 insertions, 14 deletions

diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index 9a0dbee..ff83afe 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1472,7 +1472,7 @@ void put()
 			printf("void %s(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, %s, 0x%02X); }\n", p.name, type.c_str(), p.code);
 		}
 
-		printf("void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, false, cvtIdx0(y), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x19, imm); }\n");
+		puts("void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { if (!((op.isMEM() || op.isXMM()) && y.isYMM())) throw Error(ERR_BAD_COMBINATION); opVex(y, 0, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x19, imm); }");
 		printf("void vextracti128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, false, cvtIdx0(y), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x39, imm); }\n");
 		printf("void vextractps(const Operand& op, const Xmm& x, uint8 imm) { if (!(op.isREG(32) || op.isMEM()) || x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, cvtIdx0(x), op, op.isREG(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_N4, 0x17, imm); }\n");
 		printf("void vinsertf128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, false, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x18, imm); }\n");
@@ -1614,10 +1614,10 @@ void put()
 		printf("#ifdef XBYAK64\n");
 		printf("void vmovq(const Xmm& x, const Reg64& r) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x6E); }\n");
 		printf("void vmovq(const Reg64& r, const Xmm& x) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x7E); }\n");
-		printf("void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x16, imm); }\n");
+		puts("void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(64) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, &xm0, op, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x16, imm); }");
 
-		printf("void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x22, imm); }\n");
-		printf("void vpinsrq(const Xmm& x, const Operand& op, uint8 imm) { vpinsrq(x, x, op, imm); }\n");
+		puts("void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(64) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x22, imm); }");
+		puts("void vpinsrq(const Xmm& x, const Operand& op, uint8 imm) { vpinsrq(x, x, op, imm); }");
 
 		printf("void vcvtss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1 | T_EVEX | T_EW1 | T_ER_X | T_N8, 0x2D); }\n");
 		printf("void vcvttss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1 | T_EVEX | T_EW1 | T_SAE_X | T_N8, 0x2C); }\n");
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index c56fed8..a60369e 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -1819,15 +1819,11 @@ private:
 	}
 	void opAVX_X_X_XM(const Xmm& x1, const Operand& op1, const Operand& op2, int type, int code0, int imm8 = NONE)
 	{
-		const Xmm *x2;
-		const Operand *op;
-		if (op2.isNone()) {
+		const Xmm *x2 = static_cast<const Xmm*>(&op1);
+		const Operand *op = &op2;
+		if (op2.isNone()) { // (x1, op1) -> (x1, x1, op1)
 			x2 = &x1;
 			op = &op1;
-		} else {
-			if (!(op1.isXMM() || ((type & T_YMM) && op1.is(Operand::YMM | Operand::ZMM)))) throw Error(ERR_BAD_COMBINATION);
-			x2 = static_cast<const Xmm*>(&op1);
-			op = &op2;
 		}
 		// (x1, x2, op)
 		if (!((x1.isXMM() && x2->isXMM()) || ((type & T_YMM) && ((x1.isYMM() && x2->isYMM()) || (x1.isZMM() && x2->isZMM()))))) throw Error(ERR_BAD_COMBINATION);
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 312d46b..70a9c2e 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1408,7 +1408,7 @@ void vpbroadcastb(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isME
 void vpbroadcastw(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 0x79); }
 void vpbroadcastd(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 0x58); }
 void vpbroadcastq(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_W0 | T_EW1 | T_YMM | T_EVEX | T_N8, 0x59); }
-void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, false, cvtIdx0(y), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x19, imm); }
+void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { if (!((op.isMEM() || op.isXMM()) && y.isYMM())) throw Error(ERR_BAD_COMBINATION); opVex(y, 0, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x19, imm); }
 void vextracti128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, false, cvtIdx0(y), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x39, imm); }
 void vextractps(const Operand& op, const Xmm& x, uint8 imm) { if (!(op.isREG(32) || op.isMEM()) || x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, cvtIdx0(x), op, op.isREG(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_N4, 0x17, imm); }
 void vinsertf128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, false, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x18, imm); }
@@ -1494,8 +1494,8 @@ void vcvtps2ph(const Operand& op, const Xmm& x, uint8 imm) { checkCvt1(x, op); o
 #ifdef XBYAK64
 void vmovq(const Xmm& x, const Reg64& r) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x6E); }
 void vmovq(const Reg64& r, const Xmm& x) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x7E); }
-void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x16, imm); }
-void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x22, imm); }
+void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(64) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, &xm0, op, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x16, imm); }
+void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(64) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x22, imm); }
 void vpinsrq(const Xmm& x, const Operand& op, uint8 imm) { vpinsrq(x, x, op, imm); }
 void vcvtss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1 | T_EVEX | T_EW1 | T_ER_X | T_N8, 0x2D); }
 void vcvttss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1 | T_EVEX | T_EW1 | T_SAE_X | T_N8, 0x2C); }