amx supports apx

herumi · Dec 1, 2023 · bec145b · bec145b
1 parent 9444381
commit bec145b
Show file tree

Hide file tree

Showing 4 changed files with 33 additions and 10 deletions.
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
@@ -2017,12 +2017,12 @@ void put64()
 
 void putAMX_TILE()
 {
-	puts("void ldtilecfg(const Address& addr) { opVex(tmm0, &tmm0, addr, T_0F38 | T_W0, 0x49); }");
-	puts("void sttilecfg(const Address& addr) { opVex(tmm0, &tmm0, addr, T_66 | T_0F38 | T_W0, 0x49); }");
-	puts("void tileloadd(const Tmm& tm, const Address& addr) { opAMX(tm, addr, T_F2 | T_0F38 | T_W0, 0x4b); }");
-	puts("void tileloaddt1(const Tmm& tm, const Address& addr) { opAMX(tm, addr, T_66 | T_0F38 | T_W0, 0x4b); }");
+	puts("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); }");
+	puts("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); }");
+	puts("void tileloadd(const Tmm& tm, const Address& addr) { opAMX(tm, addr, T_F2|T_0F38|T_W0, 0x4B); }");
+	puts("void tileloaddt1(const Tmm& tm, const Address& addr) { opAMX(tm, addr, T_66|T_0F38|T_W0, 0x4B); }");
 	puts("void tilerelease() { db(0xc4); db(0xe2); db(0x78); db(0x49); db(0xc0); }");
-	puts("void tilestored(const Address& addr, const Tmm& tm) { opVex(tm, &tmm0, addr, T_F3 | T_0F38 | T_W0, 0x4b); }");
+	puts("void tilestored(const Address& addr, const Tmm& tm) { if (opROO(Reg(), addr, tm, T_APX|T_F3|T_0F38|T_W0, 0x4B)) return; opVex(tm, &tmm0, addr, T_F3|T_0F38|T_W0, 0x4B); }");
 	puts("void tilezero(const Tmm& Tmm) { opVex(Tmm, &tmm0, tmm0, T_F2 | T_0F38 | T_W0, 0x49); }");
 }
 void putAMX_INT8()

diff --git a/test/apx.cpp b/test/apx.cpp
@@ -1753,3 +1753,25 @@ CYBOZU_TEST_AUTO(kmov)
 	CYBOZU_TEST_EQUAL(c.getSize(), n);
 	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
 }
+
+CYBOZU_TEST_AUTO(amx)
+{
+	struct Code : Xbyak::CodeGenerator {
+		Code()
+		{
+			ldtilecfg(ptr [r30+r29*4+0x12]);
+			sttilecfg(ptr [r30+r29*4+0x12]);
+			tileloadd(tmm1, ptr [r30+r29*4+0x12]);
+			tileloaddt1(tmm3, ptr [r30+r29*4+0x12]);
+			tilestored(ptr [r30+r29*4+0x12], tmm5);
+		}
+	} c;
+	const uint8_t tbl[] = {
+		0x62, 0x9a, 0x78, 0x08, 0x49, 0x44, 0xae, 0x12, 0x62, 0x9a, 0x79, 0x08, 0x49, 0x44, 0xae, 0x12,
+		0x62, 0x9a, 0x7b, 0x08, 0x4b, 0x4c, 0xae, 0x12, 0x62, 0x9a, 0x79, 0x08, 0x4b, 0x5c, 0xae, 0x12,
+		0x62, 0x9a, 0x7a, 0x08, 0x4b, 0x6c, 0xae, 0x12,
+	};
+	const size_t n = sizeof(tbl);
+	CYBOZU_TEST_EQUAL(c.getSize(), n);
+	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
+}
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
@@ -2708,6 +2708,7 @@ class CodeGenerator : public CodeArray {
 		Address addr2 = addr.cloneNoOptimize();
 		const RegExp exp = addr2.getRegExp();
 		if (exp.getBase().getBit() == 0 || exp.getIndex().getBit() == 0) XBYAK_THROW(ERR_NOT_SUPPORTED)
+		if (opROO(Reg(), addr2, t1, T_APX|type, code)) return;
 		opVex(t1, &tmm0, addr2, type, code);
 	}
 #endif

diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
@@ -1926,12 +1926,12 @@ void cmpoxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opRRO(r
 void cmppxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opRRO(r1, r2, addr, T_APX|T_66|T_0F38, 0xEA); }
 void cmpsxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opRRO(r1, r2, addr, T_APX|T_66|T_0F38, 0xE8); }
 void cmpzxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opRRO(r1, r2, addr, T_APX|T_66|T_0F38, 0xE4); }
-void ldtilecfg(const Address& addr) { opVex(tmm0, &tmm0, addr, T_0F38 | T_W0, 0x49); }
-void sttilecfg(const Address& addr) { 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); }
-void tileloaddt1(const Tmm& tm, const Address& addr) { opAMX(tm, addr, T_66 | T_0F38 | T_W0, 0x4b); }
+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); }
+void tileloaddt1(const Tmm& tm, const Address& addr) { opAMX(tm, addr, T_66|T_0F38|T_W0, 0x4B); }
 void tilerelease() { db(0xc4); db(0xe2); db(0x78); db(0x49); db(0xc0); }
-void tilestored(const Address& addr, const Tmm& tm) { opVex(tm, &tmm0, addr, T_F3 | T_0F38 | T_W0, 0x4b); }
+void tilestored(const Address& addr, const Tmm& tm) { if (opROO(Reg(), addr, tm, T_APX|T_F3|T_0F38|T_W0, 0x4B)) return; opVex(tm, &tmm0, addr, T_F3|T_0F38|T_W0, 0x4B); }
 void tilezero(const Tmm& Tmm) { opVex(Tmm, &tmm0, tmm0, T_F2 | T_0F38 | T_W0, 0x49); }
 void tdpbssd(const Tmm& x1, const Tmm& x2, const Tmm& x3) { opVex(x1, &x3, x2, T_F2 | T_0F38 | T_W0, 0x5e); }
 void tdpbsud(const Tmm& x1, const Tmm& x2, const Tmm& x3) { opVex(x1, &x3, x2, T_F3 | T_0F38 | T_W0, 0x5e); }