diff options
author | MITSUNARI Shigeo <[email protected]> | 2024-10-07 10:59:08 +0900 |
---|---|---|
committer | MITSUNARI Shigeo <[email protected]> | 2024-10-07 10:59:08 +0900 |
commit | ee42b66543b7a416eff50540cc20221beb3e7739 (patch) | |
tree | bb54ca2e20d250d00918a1cb079a3c493afc5327 | |
parent | 4856dcca61daaea808a4e8050e4ecf678cabd93e (diff) | |
download | xbyak-ee42b66543b7a416eff50540cc20221beb3e7739.tar.gz xbyak-ee42b66543b7a416eff50540cc20221beb3e7739.zip |
add {rd,wr}{fs,gs}base
-rw-r--r-- | gen/gen_code.cpp | 9 | ||||
-rw-r--r-- | test/misc.cpp | 52 | ||||
-rw-r--r-- | xbyak/xbyak_mnemonic.h | 4 |
3 files changed, 64 insertions, 1 deletions
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp index e43c9b2..ee8494f 100644 --- a/gen/gen_code.cpp +++ b/gen/gen_code.cpp @@ -2057,6 +2057,15 @@ void put64() // encodekey puts("void encodekey128(const Reg32& r1, const Reg32& r2) { opEncodeKey(r1, r2, 0xFA, 0xDA); }"); puts("void encodekey256(const Reg32& r1, const Reg32& r2) { opEncodeKey(r1, r2, 0xFB, 0xDB); }"); + // read/write fs/gs + { + const char *tbl[] = { + "rdfs", "rdgs", "wrfs", "wrgs" + }; + for (int i = 0; i < 4; i++) { + printf("void %sbase(const Reg32e& r) { opRR(%s, r, T_F3|T_0F, 0xAE); }\n", tbl[i], Reg(i, Operand::REG, 32).toString()); + } + } } void putAMX_TILE() diff --git a/test/misc.cpp b/test/misc.cpp index 1457691..a15721f 100644 --- a/test/misc.cpp +++ b/test/misc.cpp @@ -2284,5 +2284,55 @@ CYBOZU_TEST_AUTO(avx_vnni_int) CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n); } - +CYBOZU_TEST_AUTO(rdfsbase) +{ + struct Code : Xbyak::CodeGenerator { + Code() + { + rdfsbase(rax); + rdfsbase(r15d); + rdfsbase(r30d); + rdfsbase(r31); + + rdgsbase(rax); + rdgsbase(r15d); + rdgsbase(r30d); + rdgsbase(r31); + + wrfsbase(rax); + wrfsbase(r15d); + wrfsbase(r30d); + wrfsbase(r31); + + wrgsbase(rax); + wrgsbase(r15d); + wrgsbase(r30d); + wrgsbase(r31); + } + } c; + const uint8_t tbl[] = { + 0xf3, 0x48, 0x0f, 0xae, 0xc0, + 0xf3, 0x41, 0x0f, 0xae, 0xc7, + 0xf3, 0xd5, 0x91, 0xae, 0xc6, + 0xf3, 0xd5, 0x99, 0xae, 0xc7, + + 0xf3, 0x48, 0x0f, 0xae, 0xc8, + 0xf3, 0x41, 0x0f, 0xae, 0xcf, + 0xf3, 0xd5, 0x91, 0xae, 0xce, + 0xf3, 0xd5, 0x99, 0xae, 0xcf, + + 0xf3, 0x48, 0x0f, 0xae, 0xd0, + 0xf3, 0x41, 0x0f, 0xae, 0xd7, + 0xf3, 0xd5, 0x91, 0xae, 0xd6, + 0xf3, 0xd5, 0x99, 0xae, 0xd7, + + 0xf3, 0x48, 0x0f, 0xae, 0xd8, + 0xf3, 0x41, 0x0f, 0xae, 0xdf, + 0xf3, 0xd5, 0x91, 0xae, 0xde, + 0xf3, 0xd5, 0x99, 0xae, 0xdf, + }; + const size_t n = sizeof(tbl) / sizeof(tbl[0]); + CYBOZU_TEST_EQUAL(c.getSize(), n); + CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n); +} #endif diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h index 6eb3d63..af6652f 100644 --- a/xbyak/xbyak_mnemonic.h +++ b/xbyak/xbyak_mnemonic.h @@ -1937,6 +1937,10 @@ void aesencwide128kl(const Address& addr) { opSSE_APX(xmm0, addr, T_F3|T_0F38, 0 void aesencwide256kl(const Address& addr) { opSSE_APX(xmm2, addr, T_F3|T_0F38, 0xD8, T_F3|T_MUST_EVEX, 0xD8); } void encodekey128(const Reg32& r1, const Reg32& r2) { opEncodeKey(r1, r2, 0xFA, 0xDA); } void encodekey256(const Reg32& r1, const Reg32& r2) { opEncodeKey(r1, r2, 0xFB, 0xDB); } +void rdfsbase(const Reg32e& r) { opRR(eax, r, T_F3|T_0F, 0xAE); } +void rdgsbase(const Reg32e& r) { opRR(ecx, r, T_F3|T_0F, 0xAE); } +void wrfsbase(const Reg32e& r) { opRR(edx, r, T_F3|T_0F, 0xAE); } +void wrgsbase(const Reg32e& r) { opRR(ebx, r, T_F3|T_0F, 0xAE); } 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); } |