add aesdecwide{128,256}kl

3 files changed, 23 insertions, 9 deletions

diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index cdf015c..52baff9 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -2020,21 +2020,21 @@ void put64()
 			uint64_t type1;
 			uint64_t type2;
 			uint8_t code;
-			int n;
+			int idx;
 		} tbl[] = {
-			{ "aesdec128kl", T_F3|T_0F38, T_MUST_EVEX|T_F3, 0xDD, 2 },
-			{ "aesdec256kl", T_F3|T_0F38, T_MUST_EVEX|T_F3, 0xDF, 2 },
+			{ "aesdec128kl", T_F3|T_0F38, T_MUST_EVEX|T_F3, 0xDD, 8 },
+			{ "aesdec256kl", T_F3|T_0F38, T_MUST_EVEX|T_F3, 0xDF, 8 },
+			{ "aesdecwide128kl", T_F3|T_0F38, T_MUST_EVEX|T_F3, 0xD8, 1 },
+			{ "aesdecwide256kl", T_F3|T_0F38, T_MUST_EVEX|T_F3, 0xD8, 3 },
 		};
 		for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 			const Tbl *p = &tbl[i];
 			std::string s1 = type2String(p->type1);
 			std::string s2 = type2String(p->type2);
-			switch (p->n) {
-			case 1:
-				break;
-			case 2:
+			if (p->idx == 8) {
 				printf("void %s(const Xmm& x, const Address& addr) { opAESKL(&x, addr, %s, %s, 0x%02X); }\n", p->name, s1.c_str(), s2.c_str(), p->code);
-				break;
+			} else {
+				printf("void %s(const Address& addr) { opAESKL(&xmm%d, addr, %s, %s, 0x%02X); }\n", p->name, p->idx, s1.c_str(), s2.c_str(), p->code);
 			}
 		}
 	}
diff --git a/test/apx.cpp b/test/apx.cpp
index fc29653..ab3ca0a 100644
--- a/test/apx.cpp
+++ b/test/apx.cpp
@@ -1783,9 +1783,15 @@ CYBOZU_TEST_AUTO(aeskl)
 			aesdec128kl(xmm15, ptr[rax+rcx*4+0x12]);
 			aesdec128kl(xmm15, ptr[r30+r29*8+0x34]);
 
-
 			aesdec256kl(xmm15, ptr[rax+rcx*4+0x12]);
 			aesdec256kl(xmm15, ptr[r30+r29*8+0x34]);
+
+			aesdecwide128kl(ptr[rax+rcx*4+0x12]);
+			aesdecwide128kl(ptr[r30+r29*8+0x34]);
+
+			aesdecwide256kl(ptr[rax+rcx*4+0x12]);
+			aesdecwide256kl(ptr[r30+r29*8+0x34]);
+
 		}
 	} c;
 	const uint8_t tbl[] = {
@@ -1795,6 +1801,12 @@ CYBOZU_TEST_AUTO(aeskl)
 		// aesdec256kl
 		0xf3, 0x44, 0x0f, 0x38, 0xdf, 0x7c, 0x88, 0x12,
 		0x62, 0x1c, 0x7a, 0x08, 0xdf, 0x7c, 0xee, 0x34,
+		// aesdecwide128kl
+		0xf3, 0x0f, 0x38, 0xd8, 0x4c, 0x88, 0x12,
+		0x62, 0x9c, 0x7a, 0x08, 0xd8, 0x4c, 0xee, 0x34, 0xf3,
+		// aesdecwide256kl
+		0x0f, 0x38, 0xd8, 0x5c, 0x88, 0x12,
+		0x62, 0x9c, 0x7a, 0x08, 0xd8, 0x5c, 0xee, 0x34,
 	};
 	const size_t n = sizeof(tbl);
 	CYBOZU_TEST_EQUAL(c.getSize(), n);
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 2ea7699..915f938 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1928,6 +1928,8 @@ void cmpsxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opRRO(r
 void cmpzxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opRRO(r1, r2, addr, T_APX|T_66|T_0F38, 0xE4); }
 void aesdec128kl(const Xmm& x, const Address& addr) { opAESKL(&x, addr, T_F3|T_0F38, T_F3|T_MUST_EVEX, 0xDD); }
 void aesdec256kl(const Xmm& x, const Address& addr) { opAESKL(&x, addr, T_F3|T_0F38, T_F3|T_MUST_EVEX, 0xDF); }
+void aesdecwide128kl(const Address& addr) { opAESKL(&xmm1, addr, T_F3|T_0F38, T_F3|T_MUST_EVEX, 0xD8); }
+void aesdecwide256kl(const Address& addr) { opAESKL(&xmm3, addr, T_F3|T_0F38, T_F3|T_MUST_EVEX, 0xD8); }
 void ldtilecfg(const Address& addr) { if (opROO(Reg(), addr, tmm0, T_APX|T_0F38|T_W0, 0x49)) return; opVex(tmm0, &tmm0, addr, T_0F38|T_W0, 0x49); }
 void sttilecfg(const Address& addr) { if (opROO(Reg(), addr, tmm0, T_APX|T_66|T_0F38|T_W0, 0x49)) return; opVex(tmm0, &tmm0, addr, T_66|T_0F38 | T_W0, 0x49); }
 void tileloadd(const Tmm& tm, const Address& addr) { opAMX(tm, addr, T_F2|T_0F38|T_W0, 0x4B); }