diff options
| author | kvn <none@none> | 2014-02-25 15:11:18 -0800 |
|---|---|---|
| committer | kvn <none@none> | 2014-02-25 15:11:18 -0800 |
| commit | 9c0308b86746f885f1e6f31e9b474b0b7b550700 (patch) | |
| tree | 77e60af7c17331350a7114b4440b6a8e3303f52f /src | |
| parent | a2bcf5591c2afbec1a941b7b4de582de157aafed (diff) | |
| parent | 86005d48ffb5057779421bbe655f52eed21d974a (diff) | |
Merge
Diffstat (limited to 'src')
65 files changed, 1448 insertions, 417 deletions
diff --git a/src/cpu/sparc/vm/assembler_sparc.hpp b/src/cpu/sparc/vm/assembler_sparc.hpp index ccbc43e5a..11547cde9 100644 --- a/src/cpu/sparc/vm/assembler_sparc.hpp +++ b/src/cpu/sparc/vm/assembler_sparc.hpp @@ -88,6 +88,7 @@ class Assembler : public AbstractAssembler { orncc_op3 = 0x16, xnorcc_op3 = 0x17, addccc_op3 = 0x18, + aes4_op3 = 0x19, umulcc_op3 = 0x1a, smulcc_op3 = 0x1b, subccc_op3 = 0x1c, @@ -121,6 +122,8 @@ class Assembler : public AbstractAssembler { fpop1_op3 = 0x34, fpop2_op3 = 0x35, impdep1_op3 = 0x36, + aes3_op3 = 0x36, + flog3_op3 = 0x36, impdep2_op3 = 0x37, jmpl_op3 = 0x38, rett_op3 = 0x39, @@ -172,41 +175,56 @@ class Assembler : public AbstractAssembler { enum opfs { // selected opfs - fmovs_opf = 0x01, - fmovd_opf = 0x02, - - fnegs_opf = 0x05, - fnegd_opf = 0x06, - - fadds_opf = 0x41, - faddd_opf = 0x42, - fsubs_opf = 0x45, - fsubd_opf = 0x46, - - fmuls_opf = 0x49, - fmuld_opf = 0x4a, - fdivs_opf = 0x4d, - fdivd_opf = 0x4e, - - fcmps_opf = 0x51, - fcmpd_opf = 0x52, - - fstox_opf = 0x81, - fdtox_opf = 0x82, - fxtos_opf = 0x84, - fxtod_opf = 0x88, - fitos_opf = 0xc4, - fdtos_opf = 0xc6, - fitod_opf = 0xc8, - fstod_opf = 0xc9, - fstoi_opf = 0xd1, - fdtoi_opf = 0xd2, - - mdtox_opf = 0x110, - mstouw_opf = 0x111, - mstosw_opf = 0x113, - mxtod_opf = 0x118, - mwtos_opf = 0x119 + fmovs_opf = 0x01, + fmovd_opf = 0x02, + + fnegs_opf = 0x05, + fnegd_opf = 0x06, + + fadds_opf = 0x41, + faddd_opf = 0x42, + fsubs_opf = 0x45, + fsubd_opf = 0x46, + + fmuls_opf = 0x49, + fmuld_opf = 0x4a, + fdivs_opf = 0x4d, + fdivd_opf = 0x4e, + + fcmps_opf = 0x51, + fcmpd_opf = 0x52, + + fstox_opf = 0x81, + fdtox_opf = 0x82, + fxtos_opf = 0x84, + fxtod_opf = 0x88, + fitos_opf = 0xc4, + fdtos_opf = 0xc6, + fitod_opf = 0xc8, + fstod_opf = 0xc9, + fstoi_opf = 0xd1, + fdtoi_opf = 0xd2, + + mdtox_opf = 0x110, + mstouw_opf = 0x111, + mstosw_opf = 0x113, + mxtod_opf = 0x118, + mwtos_opf = 0x119, + + aes_kexpand0_opf = 0x130, + aes_kexpand2_opf = 0x131 + }; + + enum op5s { + aes_eround01_op5 = 0x00, + aes_eround23_op5 = 0x01, + aes_dround01_op5 = 0x02, + aes_dround23_op5 = 0x03, + aes_eround01_l_op5 = 0x04, + aes_eround23_l_op5 = 0x05, + aes_dround01_l_op5 = 0x06, + aes_dround23_l_op5 = 0x07, + aes_kexpand1_op5 = 0x08 }; enum RCondition { rc_z = 1, rc_lez = 2, rc_lz = 3, rc_nz = 5, rc_gz = 6, rc_gez = 7, rc_last = rc_gez }; @@ -427,6 +445,7 @@ class Assembler : public AbstractAssembler { static int immed( bool i) { return u_field(i ? 1 : 0, 13, 13); } static int opf_low6( int w) { return u_field(w, 10, 5); } static int opf_low5( int w) { return u_field(w, 9, 5); } + static int op5( int x) { return u_field(x, 8, 5); } static int trapcc( CC cc) { return u_field(cc, 12, 11); } static int sx( int i) { return u_field(i, 12, 12); } // shift x=1 means 64-bit static int opf( int x) { return u_field(x, 13, 5); } @@ -451,6 +470,7 @@ class Assembler : public AbstractAssembler { static int fd( FloatRegister r, FloatRegisterImpl::Width fwa) { return u_field(r->encoding(fwa), 29, 25); }; static int fs1(FloatRegister r, FloatRegisterImpl::Width fwa) { return u_field(r->encoding(fwa), 18, 14); }; static int fs2(FloatRegister r, FloatRegisterImpl::Width fwa) { return u_field(r->encoding(fwa), 4, 0); }; + static int fs3(FloatRegister r, FloatRegisterImpl::Width fwa) { return u_field(r->encoding(fwa), 13, 9); }; // some float instructions use this encoding on the op3 field static int alt_op3(int op, FloatRegisterImpl::Width w) { @@ -559,6 +579,12 @@ class Assembler : public AbstractAssembler { return x & ((1 << 10) - 1); } + // AES crypto instructions supported only on certain processors + static void aes_only() { assert( VM_Version::has_aes(), "This instruction only works on SPARC with AES instructions support"); } + + // instruction only in VIS1 + static void vis1_only() { assert( VM_Version::has_vis1(), "This instruction only works on SPARC with VIS1"); } + // instruction only in VIS3 static void vis3_only() { assert( VM_Version::has_vis3(), "This instruction only works on SPARC with VIS3"); } @@ -682,6 +708,24 @@ public: void addccc( Register s1, int simm13a, Register d ) { emit_int32( op(arith_op) | rd(d) | op3(addc_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); } + // 4-operand AES instructions + + void aes_eround01( FloatRegister s1, FloatRegister s2, FloatRegister s3, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | fs3(s3, FloatRegisterImpl::D) | op5(aes_eround01_op5) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_eround23( FloatRegister s1, FloatRegister s2, FloatRegister s3, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | fs3(s3, FloatRegisterImpl::D) | op5(aes_eround23_op5) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_dround01( FloatRegister s1, FloatRegister s2, FloatRegister s3, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | fs3(s3, FloatRegisterImpl::D) | op5(aes_dround01_op5) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_dround23( FloatRegister s1, FloatRegister s2, FloatRegister s3, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | fs3(s3, FloatRegisterImpl::D) | op5(aes_dround23_op5) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_eround01_l( FloatRegister s1, FloatRegister s2, FloatRegister s3, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | fs3(s3, FloatRegisterImpl::D) | op5(aes_eround01_l_op5) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_eround23_l( FloatRegister s1, FloatRegister s2, FloatRegister s3, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | fs3(s3, FloatRegisterImpl::D) | op5(aes_eround23_l_op5) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_dround01_l( FloatRegister s1, FloatRegister s2, FloatRegister s3, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | fs3(s3, FloatRegisterImpl::D) | op5(aes_dround01_l_op5) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_dround23_l( FloatRegister s1, FloatRegister s2, FloatRegister s3, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | fs3(s3, FloatRegisterImpl::D) | op5(aes_dround23_l_op5) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_kexpand1( FloatRegister s1, FloatRegister s2, int imm5a, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes4_op3) | fs1(s1, FloatRegisterImpl::D) | u_field(imm5a, 13, 9) | op5(aes_kexpand1_op5) | fs2(s2, FloatRegisterImpl::D) ); } + + + // 3-operand AES instructions + + void aes_kexpand0( FloatRegister s1, FloatRegister s2, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes3_op3) | fs1(s1, FloatRegisterImpl::D) | opf(aes_kexpand0_opf) | fs2(s2, FloatRegisterImpl::D) ); } + void aes_kexpand2( FloatRegister s1, FloatRegister s2, FloatRegister d ) { aes_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(aes3_op3) | fs1(s1, FloatRegisterImpl::D) | opf(aes_kexpand2_opf) | fs2(s2, FloatRegisterImpl::D) ); } + // pp 136 inline void bpr(RCondition c, bool a, Predict p, Register s1, address d, relocInfo::relocType rt = relocInfo::none); @@ -784,6 +828,10 @@ public: void fmul( FloatRegisterImpl::Width sw, FloatRegisterImpl::Width dw, FloatRegister s1, FloatRegister s2, FloatRegister d ) { emit_int32( op(arith_op) | fd(d, dw) | op3(fpop1_op3) | fs1(s1, sw) | opf(0x60 + sw + dw*4) | fs2(s2, sw)); } void fdiv( FloatRegisterImpl::Width w, FloatRegister s1, FloatRegister s2, FloatRegister d ) { emit_int32( op(arith_op) | fd(d, w) | op3(fpop1_op3) | fs1(s1, w) | opf(0x4c + w) | fs2(s2, w)); } + // FXORs/FXORd instructions + + void fxor( FloatRegisterImpl::Width w, FloatRegister s1, FloatRegister s2, FloatRegister d ) { vis1_only(); emit_int32( op(arith_op) | fd(d, w) | op3(flog3_op3) | fs1(s1, w) | opf(0x6E - w) | fs2(s2, w)); } + // pp 164 void fsqrt( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d ) { emit_int32( op(arith_op) | fd(d, w) | op3(fpop1_op3) | opf(0x28 + w) | fs2(s, w)); } diff --git a/src/cpu/sparc/vm/sparc.ad b/src/cpu/sparc/vm/sparc.ad index ef1ccf1c0..3a470b0f8 100644 --- a/src/cpu/sparc/vm/sparc.ad +++ b/src/cpu/sparc/vm/sparc.ad @@ -1853,6 +1853,12 @@ const bool Matcher::misaligned_vectors_ok() { return false; } +// Current (2013) SPARC platforms need to read original key +// to construct decryption expanded key +const bool Matcher::pass_original_key_for_aes() { + return true; +} + // USII supports fxtof through the whole range of number, USIII doesn't const bool Matcher::convL2FSupported(void) { return VM_Version::has_fast_fxtof(); diff --git a/src/cpu/sparc/vm/stubGenerator_sparc.cpp b/src/cpu/sparc/vm/stubGenerator_sparc.cpp index b99118281..25023404d 100644 --- a/src/cpu/sparc/vm/stubGenerator_sparc.cpp +++ b/src/cpu/sparc/vm/stubGenerator_sparc.cpp @@ -3304,6 +3304,775 @@ class StubGenerator: public StubCodeGenerator { } } + address generate_aescrypt_encryptBlock() { + __ align(CodeEntryAlignment); + StubCodeMark mark(this, "StubRoutines", "aesencryptBlock"); + Label L_doLast128bit, L_storeOutput; + address start = __ pc(); + Register from = O0; // source byte array + Register to = O1; // destination byte array + Register key = O2; // expanded key array + const Register keylen = O4; //reg for storing expanded key array length + + // read expanded key length + __ ldsw(Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)), keylen, 0); + + // load input into F54-F56; F30-F31 used as temp + __ ldf(FloatRegisterImpl::S, from, 0, F30); + __ ldf(FloatRegisterImpl::S, from, 4, F31); + __ fmov(FloatRegisterImpl::D, F30, F54); + __ ldf(FloatRegisterImpl::S, from, 8, F30); + __ ldf(FloatRegisterImpl::S, from, 12, F31); + __ fmov(FloatRegisterImpl::D, F30, F56); + + // load expanded key + for ( int i = 0; i <= 38; i += 2 ) { + __ ldf(FloatRegisterImpl::D, key, i*4, as_FloatRegister(i)); + } + + // perform cipher transformation + __ fxor(FloatRegisterImpl::D, F0, F54, F54); + __ fxor(FloatRegisterImpl::D, F2, F56, F56); + // rounds 1 through 8 + for ( int i = 4; i <= 28; i += 8 ) { + __ aes_eround01(as_FloatRegister(i), F54, F56, F58); + __ aes_eround23(as_FloatRegister(i+2), F54, F56, F60); + __ aes_eround01(as_FloatRegister(i+4), F58, F60, F54); + __ aes_eround23(as_FloatRegister(i+6), F58, F60, F56); + } + __ aes_eround01(F36, F54, F56, F58); //round 9 + __ aes_eround23(F38, F54, F56, F60); + + // 128-bit original key size + __ cmp_and_brx_short(keylen, 44, Assembler::equal, Assembler::pt, L_doLast128bit); + + for ( int i = 40; i <= 50; i += 2 ) { + __ ldf(FloatRegisterImpl::D, key, i*4, as_FloatRegister(i) ); + } + __ aes_eround01(F40, F58, F60, F54); //round 10 + __ aes_eround23(F42, F58, F60, F56); + __ aes_eround01(F44, F54, F56, F58); //round 11 + __ aes_eround23(F46, F54, F56, F60); + + // 192-bit original key size + __ cmp_and_brx_short(keylen, 52, Assembler::equal, Assembler::pt, L_storeOutput); + + __ ldf(FloatRegisterImpl::D, key, 208, F52); + __ aes_eround01(F48, F58, F60, F54); //round 12 + __ aes_eround23(F50, F58, F60, F56); + __ ldf(FloatRegisterImpl::D, key, 216, F46); + __ ldf(FloatRegisterImpl::D, key, 224, F48); + __ ldf(FloatRegisterImpl::D, key, 232, F50); + __ aes_eround01(F52, F54, F56, F58); //round 13 + __ aes_eround23(F46, F54, F56, F60); + __ br(Assembler::always, false, Assembler::pt, L_storeOutput); + __ delayed()->nop(); + + __ BIND(L_doLast128bit); + __ ldf(FloatRegisterImpl::D, key, 160, F48); + __ ldf(FloatRegisterImpl::D, key, 168, F50); + + __ BIND(L_storeOutput); + // perform last round of encryption common for all key sizes + __ aes_eround01_l(F48, F58, F60, F54); //last round + __ aes_eround23_l(F50, F58, F60, F56); + + // store output into the destination array, F0-F1 used as temp + __ fmov(FloatRegisterImpl::D, F54, F0); + __ stf(FloatRegisterImpl::S, F0, to, 0); + __ stf(FloatRegisterImpl::S, F1, to, 4); + __ fmov(FloatRegisterImpl::D, F56, F0); + __ stf(FloatRegisterImpl::S, F0, to, 8); + __ retl(); + __ delayed()->stf(FloatRegisterImpl::S, F1, to, 12); + + return start; + } + + address generate_aescrypt_decryptBlock() { + __ align(CodeEntryAlignment); + StubCodeMark mark(this, "StubRoutines", "aesdecryptBlock"); + address start = __ pc(); + Label L_expand192bit, L_expand256bit, L_common_transform; + Register from = O0; // source byte array + Register to = O1; // destination byte array + Register key = O2; // expanded key array + Register original_key = O3; // original key array only required during decryption + const Register keylen = O4; // reg for storing expanded key array length + + // read expanded key array length + __ ldsw(Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)), keylen, 0); + + // load input into F52-F54; F30,F31 used as temp + __ ldf(FloatRegisterImpl::S, from, 0, F30); + __ ldf(FloatRegisterImpl::S, from, 4, F31); + __ fmov(FloatRegisterImpl::D, F30, F52); + __ ldf(FloatRegisterImpl::S, from, 8, F30); + __ ldf(FloatRegisterImpl::S, from, 12, F31); + __ fmov(FloatRegisterImpl::D, F30, F54); + + // load original key from SunJCE expanded decryption key + for ( int i = 0; i <= 3; i++ ) { + __ ldf(FloatRegisterImpl::S, original_key, i*4, as_FloatRegister(i)); + } + + // 256-bit original key size + __ cmp_and_brx_short(keylen, 60, Assembler::equal, Assembler::pn, L_expand256bit); + + // 192-bit original key size + __ cmp_and_brx_short(keylen, 52, Assembler::equal, Assembler::pn, L_expand192bit); + + // 128-bit original key size + // perform key expansion since SunJCE decryption-key expansion is not compatible with SPARC crypto instructions + for ( int i = 0; i <= 36; i += 4 ) { + __ aes_kexpand1(as_FloatRegister(i), as_FloatRegister(i+2), i/4, as_FloatRegister(i+4)); + __ aes_kexpand2(as_FloatRegister(i+2), as_FloatRegister(i+4), as_FloatRegister(i+6)); + } + + // perform 128-bit key specific inverse cipher transformation + __ fxor(FloatRegisterImpl::D, F42, F54, F54); + __ fxor(FloatRegisterImpl::D, F40, F52, F52); + __ br(Assembler::always, false, Assembler::pt, L_common_transform); + __ delayed()->nop(); + + __ BIND(L_expand192bit); + + // start loading rest of the 192-bit key + __ ldf(FloatRegisterImpl::S, original_key, 16, F4); + __ ldf(FloatRegisterImpl::S, original_key, 20, F5); + + // perform key expansion since SunJCE decryption-key expansion is not compatible with SPARC crypto instructions + for ( int i = 0; i <= 36; i += 6 ) { + __ aes_kexpand1(as_FloatRegister(i), as_FloatRegister(i+4), i/6, as_FloatRegister(i+6)); + __ aes_kexpand2(as_FloatRegister(i+2), as_FloatRegister(i+6), as_FloatRegister(i+8)); + __ aes_kexpand2(as_FloatRegister(i+4), as_FloatRegister(i+8), as_FloatRegister(i+10)); + } + __ aes_kexpand1(F42, F46, 7, F48); + __ aes_kexpand2(F44, F48, F50); + + // perform 192-bit key specific inverse cipher transformation + __ fxor(FloatRegisterImpl::D, F50, F54, F54); + __ fxor(FloatRegisterImpl::D, F48, F52, F52); + __ aes_dround23(F46, F52, F54, F58); + __ aes_dround01(F44, F52, F54, F56); + __ aes_dround23(F42, F56, F58, F54); + __ aes_dround01(F40, F56, F58, F52); + __ br(Assembler::always, false, Assembler::pt, L_common_transform); + __ delayed()->nop(); + + __ BIND(L_expand256bit); + + // load rest of the 256-bit key + for ( int i = 4; i <= 7; i++ ) { + __ ldf(FloatRegisterImpl::S, original_key, i*4, as_FloatRegister(i)); + } + + // perform key expansion since SunJCE decryption-key expansion is not compatible with SPARC crypto instructions + for ( int i = 0; i <= 40; i += 8 ) { + __ aes_kexpand1(as_FloatRegister(i), as_FloatRegister(i+6), i/8, as_FloatRegister(i+8)); + __ aes_kexpand2(as_FloatRegister(i+2), as_FloatRegister(i+8), as_FloatRegister(i+10)); + __ aes_kexpand0(as_FloatRegister(i+4), as_FloatRegister(i+10), as_FloatRegister(i+12)); + __ aes_kexpand2(as_FloatRegister(i+6), as_FloatRegister(i+12), as_FloatRegister(i+14)); + } + __ aes_kexpand1(F48, F54, 6, F56); + __ aes_kexpand2(F50, F56, F58); + + for ( int i = 0; i <= 6; i += 2 ) { + __ fmov(FloatRegisterImpl::D, as_FloatRegister(58-i), as_FloatRegister(i)); + } + + // load input into F52-F54 + __ ldf(FloatRegisterImpl::D, from, 0, F52); + __ ldf(FloatRegisterImpl::D, from, 8, F54); + + // perform 256-bit key specific inverse cipher transformation + __ fxor(FloatRegisterImpl::D, F0, F54, F54); + __ fxor(FloatRegisterImpl::D, F2, F52, F52); + __ aes_dround23(F4, F52, F54, F58); + __ aes_dround01(F6, F52, F54, F56); + __ aes_dround23(F50, F56, F58, F54); + __ aes_dround01(F48, F56, F58, F52); + __ aes_dround23(F46, F52, F54, F58); + __ aes_dround01(F44, F52, F54, F56); + __ aes_dround23(F42, F56, F58, F54); + __ aes_dround01(F40, F56, F58, F52); + + for ( int i = 0; i <= 7; i++ ) { + __ ldf(FloatRegisterImpl::S, original_key, i*4, as_FloatRegister(i)); + } + + // perform inverse cipher transformations common for all key sizes + __ BIND(L_common_transform); + for ( int i = 38; i >= 6; i -= 8 ) { + __ aes_dround23(as_FloatRegister(i), F52, F54, F58); + __ aes_dround01(as_FloatRegister(i-2), F52, F54, F56); + if ( i != 6) { + __ aes_dround23(as_FloatRegister(i-4), F56, F58, F54); + __ aes_dround01(as_FloatRegister(i-6), F56, F58, F52); + } else { + __ aes_dround23_l(as_FloatRegister(i-4), F56, F58, F54); + __ aes_dround01_l(as_FloatRegister(i-6), F56, F58, F52); + } + } + + // store output to destination array, F0-F1 used as temp + __ fmov(FloatRegisterImpl::D, F52, F0); + __ stf(FloatRegisterImpl::S, F0, to, 0); + __ stf(FloatRegisterImpl::S, F1, to, 4); + __ fmov(FloatRegisterImpl::D, F54, F0); + __ stf(FloatRegisterImpl::S, F0, to, 8); + __ retl(); + __ delayed()->stf(FloatRegisterImpl::S, F1, to, 12); + + return start; + } + + address generate_cipherBlockChaining_encryptAESCrypt() { + __ align(CodeEntryAlignment); + StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt"); + Label L_cbcenc128, L_cbcenc192, L_cbcenc256; + address start = __ pc(); + Register from = O0; // source byte array + Register to = O1; // destination byte array + Register key = O2; // expanded key array + Register rvec = O3; // init vector + const Register len_reg = O4; // cipher length + const Register keylen = O5; // reg for storing expanded key array length + + // save cipher len to return in the end + __ mov(len_reg, L1); + + // read expanded key length + __ ldsw(Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)), keylen, 0); + + // load init vector + __ ldf(FloatRegisterImpl::D, rvec, 0, F60); + __ ldf(FloatRegisterImpl::D, rvec, 8, F62); + __ ldx(key,0,G1); + __ ldx(key,8,G2); + + // start loading expanded key + for ( int i = 0, j = 16; i <= 38; i += 2, j += 8 ) { + __ ldf(FloatRegisterImpl::D, key, j, as_FloatRegister(i)); + } + + // 128-bit original key size + __ cmp_and_brx_short(keylen, 44, Assembler::equal, Assembler::pt, L_cbcenc128); + + for ( int i = 40, j = 176; i <= 46; i += 2, j += 8 ) { + __ ldf(FloatRegisterImpl::D, key, j, as_FloatRegister(i)); + } + + // 192-bit original key size + __ cmp_and_brx_short(keylen, 52, Assembler::equal, Assembler::pt, L_cbcenc192); + + for ( int i = 48, j = 208; i <= 54; i += 2, j += 8 ) { + __ ldf(FloatRegisterImpl::D, key, j, as_FloatRegister(i)); + } + + // 256-bit original key size + __ br(Assembler::always, false, Assembler::pt, L_cbcenc256); + __ delayed()->nop(); + + __ align(OptoLoopAlignment); + __ BIND(L_cbcenc128); + __ ldx(from,0,G3); + __ ldx(from,8,G4); + __ xor3(G1,G3,G3); + __ xor3(G2,G4,G4); + __ movxtod(G3,F56); + __ movxtod(G4,F58); + __ fxor(FloatRegisterImpl::D, F60, F56, F60); + __ fxor(FloatRegisterImpl::D, F62, F58, F62); + + // TEN_EROUNDS + for ( int i = 0; i <= 32; i += 8 ) { + __ aes_eround01(as_FloatRegister(i), F60, F62, F56); + __ aes_eround23(as_FloatRegister(i+2), F60, F62, F58); + if (i != 32 ) { + __ aes_eround01(as_FloatRegister(i+4), F56, F58, F60); + __ aes_eround23(as_FloatRegister(i+6), F56, F58, F62); + } else { + __ aes_eround01_l(as_FloatRegister(i+4), F56, F58, F60); + __ aes_eround23_l(as_FloatRegister(i+6), F56, F58, F62); + } + } + + __ stf(FloatRegisterImpl::D, F60, to, 0); + __ stf(FloatRegisterImpl::D, F62, to, 8); + __ add(from, 16, from); + __ add(to, 16, to); + __ subcc(len_reg, 16, len_reg); + __ br(Assembler::notEqual, false, Assembler::pt, L_cbcenc128); + __ delayed()->nop(); + __ stf(FloatRegisterImpl::D, F60, rvec, 0); + __ stf(FloatRegisterImpl::D, F62, rvec, 8); + __ retl(); + __ delayed()->mov(L1, O0); + + __ align(OptoLoopAlignment); + __ BIND(L_cbcenc192); + __ ldx(from,0,G3); + __ ldx(from,8,G4); + __ xor3(G1,G3,G3); + __ xor3(G2,G4,G4); + __ movxtod(G3,F56); + __ movxtod(G4,F58); + __ fxor(FloatRegisterImpl::D, F60, F56, F60); + __ fxor(FloatRegisterImpl::D, F62, F58, F62); + + // TWELEVE_EROUNDS + for ( int i = 0; i <= 40; i += 8 ) { + __ aes_eround01(as_FloatRegister(i), F60, F62, F56); + __ aes_eround23(as_FloatRegister(i+2), F60, F62, F58); + if (i != 40 ) { + __ aes_eround01(as_FloatRegister(i+4), F56, F58, F60); + __ aes_eround23(as_FloatRegister(i+6), F56, F58, F62); + } else { + __ aes_eround01_l(as_FloatRegister(i+4), F56, F58, F60); + __ aes_eround23_l(as_FloatRegister(i+6), F56, F58, F62); + } + } + + __ stf(FloatRegisterImpl::D, F60, to, 0); + __ stf(FloatRegisterImpl::D, F62, to, 8); + __ add(from, 16, from); + __ subcc(len_reg, 16, len_reg); + __ add(to, 16, to); + __ br(Assembler::notEqual, false, Assembler::pt, L_cbcenc192); + __ delayed()->nop(); + __ stf(FloatRegisterImpl::D, F60, rvec, 0); + __ stf(FloatRegisterImpl::D, F62, rvec, 8); + __ retl(); + __ delayed()->mov(L1, O0); + + __ align(OptoLoopAlignment); + __ BIND(L_cbcenc256); + __ ldx(from,0,G3); + __ ldx(from,8,G4); + __ xor3(G1,G3,G3); + __ xor3(G2,G4,G4); + __ movxtod(G3,F56); + __ movxtod(G4,F58); + __ fxor(FloatRegisterImpl::D, F60, F56, F60); + __ fxor(FloatRegisterImpl::D, F62, F58, F62); + + // FOURTEEN_EROUNDS + for ( int i = 0; i <= 48; i += 8 ) { + __ aes_eround01(as_FloatRegister(i), F60, F62, F56); + __ aes_eround23(as_FloatRegister(i+2), F60, F62, F58); + if (i != 48 ) { + __ aes_eround01(as_FloatRegister(i+4), F56, F58, F60); + __ aes_eround23(as_FloatRegister(i+6), F56, F58, F62); + } else { + __ aes_eround01_l(as_FloatRegister(i+4), F56, F58, F60); + __ aes_eround23_l(as_FloatRegister(i+6), F56, F58, F62); + } + } + + __ stf(FloatRegisterImpl::D, F60, to, 0); + __ stf(FloatRegisterImpl::D, F62, to, 8); + __ add(from, 16, from); + __ subcc(len_reg, 16, len_reg); + __ add(to, 16, to); + __ br(Assembler::notEqual, false, Assembler::pt, L_cbcenc256); + __ delayed()->nop(); + __ stf(FloatRegisterImpl::D, F60, rvec, 0); + __ stf(FloatRegisterImpl::D, F62, rvec, 8); + __ retl(); + __ delayed()->mov(L1, O0); + + return start; + } + + address generate_cipherBlockChaining_decryptAESCrypt_Parallel() { + __ align(CodeEntryAlignment); + StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt"); + Label L_cbcdec_end, L_expand192bit, L_expand256bit, L_dec_first_block_start; + Label L_dec_first_block128, L_dec_first_block192, L_dec_next2_blocks128, L_dec_next2_blocks192, L_dec_next2_blocks256; + address start = __ pc(); + Register from = I0; // source byte array + Register to = I1; // destination byte array + Register key = I2; // expanded key array + Register rvec = I3; // init vector + const Register len_reg = I4; // cipher length + const Register original_key = I5; // original key array only required during decryption + const Register keylen = L6; // reg for storing expanded key array length + + // save cipher len before save_frame, to return in the end + __ mov(O4, L0); + __ save_frame(0); //args are read from I* registers since we save the frame in the beginning + + // load original key from SunJCE expanded decryption key + for ( int i = 0; i <= 3; i++ ) { + __ ldf(FloatRegisterImpl::S, original_key, i*4, as_FloatRegister(i)); + } + + // load initial vector + __ ldx(rvec,0,L0); + __ ldx(rvec,8,L1); + + // read expanded key array length + __ ldsw(Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)), keylen, 0); + + // 256-bit original key size + __ cmp_and_brx_short(keylen, 60, Assembler::equal, Assembler::pn, L_expand256bit); + + // 192-bit original key size + __ cmp_and_brx_short(keylen, 52, Assembler::equal, Assembler::pn, L_expand192bit); + + // 128-bit original key size + // perform key expansion since SunJCE decryption-key expansion is not compatible with SPARC crypto instructions + for ( int i = 0; i <= 36; i += 4 ) { + __ aes_kexpand1(as_FloatRegister(i), as_FloatRegister(i+2), i/4, as_FloatRegister(i+4)); + __ aes_kexpand2(as_FloatRegister(i+2), as_FloatRegister(i+4), as_FloatRegister(i+6)); + } + + // load expanded key[last-1] and key[last] elements + __ movdtox(F40,L2); + __ movdtox(F42,L3); + + __ and3(len_reg, 16, L4); + __ br_null(L4, false, Assembler::pt, L_dec_next2_blocks128); + __ delayed()->nop(); + + __ br(Assembler::always, false, Assembler::pt, L_dec_first_block_start); + __ delayed()->nop(); + + __ BIND(L_expand192bit); + // load rest of the 192-bit key + __ ldf(FloatRegisterImpl::S, original_key, 16, F4); + __ ldf(FloatRegisterImpl::S, original_key, 20, F5); + + // perform key expansion since SunJCE decryption-key expansion is not compatible with SPARC crypto instructions + for ( int i = 0; i <= 36; i += 6 ) { + __ aes_kexpand1(as_FloatRegister(i), as_FloatRegister(i+4), i/6, as_FloatRegister(i+6)); + __ aes_kexpand2(as_FloatRegister(i+2), as_FloatRegister(i+6), as_FloatRegister(i+8)); + __ aes_kexpand2(as_FloatRegister(i+4), as_FloatRegister(i+8), as_FloatRegister(i+10)); + } + __ aes_kexpand1(F42, F46, 7, F48); + __ aes_kexpand2(F44, F48, F50); + + // load expanded key[last-1] and key[last] elements + __ movdtox(F48,L2); + __ movdtox(F50,L3); + + __ and3(len_reg, 16, L4); + __ br_null(L4, false, Assembler::pt, L_dec_next2_blocks192); + __ delayed()->nop(); + + __ br(Assembler::always, false, Assembler::pt, L_dec_first_block_start); + __ delayed()->nop(); + + __ BIND(L_expand256bit); + // load rest of the 256-bit key + for ( int i = 4; i <= 7; i++ ) { + __ ldf(FloatRegisterImpl::S, original_key, i*4, as_FloatRegister(i)); + } + + // perform key expansion since SunJCE decryption-key expansion is not compatible with SPARC crypto instructions + for ( int i = 0; i <= 40; i += 8 ) { + __ aes_kexpand1(as_FloatRegister(i), as_FloatRegister(i+6), i/8, as_FloatRegister(i+8)); + __ aes_kexpand2(as_FloatRegister(i+2), as_FloatRegister(i+8), as_FloatRegister(i+10)); + __ aes_kexpand0(as_FloatRegister(i+4), as_FloatRegister(i+10), as_FloatRegister(i+12)); + __ aes_kexpand2(as_FloatRegister(i+6), as_FloatRegister(i+12), as_FloatRegister(i+14)); + } + __ aes_kexpand1(F48, F54, 6, F56); + __ aes_kexpand2(F50, F56, F58); + + // load expanded key[last-1] and key[last] elements + __ movdtox(F56,L2); + __ movdtox(F58,L3); + + __ and3(len_reg, 16, L4); + __ br_null(L4, false, Assembler::pt, L_dec_next2_blocks256); + __ delayed()->nop(); + + __ BIND(L_dec_first_block_start); + __ ldx(from,0,L4); + __ ldx(from,8,L5); + __ xor3(L2,L4,G1); + __ movxtod(G1,F60); + __ xor3(L3,L5,G1); + __ movxtod(G1,F62); + + // 128-bit original key size + __ cmp_and_brx_short(keylen, 44, Assembler::equal, Assembler::pn, L_dec_first_block128); + + // 192-bit original key size + __ cmp_and_brx_short(keylen, 52, Assembler::equal, Assembler::pn, L_dec_first_block192); + + __ aes_dround23(F54, F60, F62, F58); + __ aes_dround01(F52, F60, F62, F56); + __ aes_dround23(F50, F56, F58, F62); + __ aes_dround01(F48, F56, F58, F60); + + __ BIND(L_dec_first_block192); + __ aes_dround23(F46, F60, F62, F58); + __ aes_dround01(F44, F60, F62, F56); + __ aes_dround23(F42, F56, F58, F62); + __ aes_dround01(F40, F56, F58, F60); + + __ BIND(L_dec_first_block128); + for ( int i = 38; i >= 6; i -= 8 ) { + __ aes_dround23(as_FloatRegister(i), F60, F62, F58); + __ aes_dround01(as_FloatRegister(i-2), F60, F62, F56); + if ( i != 6) { + __ aes_dround23(as_FloatRegister(i-4), F56, F58, F62); + __ aes_dround01(as_FloatRegister(i-6), F56, F58, F60); + } else { + __ aes_dround23_l(as_FloatRegister(i-4), F56, F58, F62); + __ aes_dround01_l(as_FloatRegister(i-6), F56, F58, F60); + } + } + + __ movxtod(L0,F56); + __ movxtod(L1,F58); + __ mov(L4,L0); + __ mov(L5,L1); + __ fxor(FloatRegisterImpl::D, F56, F60, F60); + __ fxor(FloatRegisterImpl::D, F58, F62, F62); + + __ stf(FloatRegisterImpl::D, F60, to, 0); + __ stf(FloatRegisterImpl::D, F62, to, 8); + + __ add(from, 16, from); + __ add(to, 16, to); + __ subcc(len_reg, 16, len_reg); + __ br(Assembler::equal, false, Assembler::pt, L_cbcdec_end); + __ delayed()->nop(); + + // 256-bit original key size + __ cmp_and_brx_short(keylen, 60, Assembler::equal, Assembler::pn, L_dec_next2_blocks256); + + // 192-bit original key size + __ cmp_and_brx_short(keylen, 52, Assembler::equal, Assembler::pn, L_dec_next2_blocks192); + + __ align(OptoLoopAlignment); + __ BIND(L_dec_next2_blocks128); + __ nop(); + + // F40:F42 used for first 16-bytes + __ ldx(from,0,G4); + __ ldx(from,8,G5); + __ xor3(L2,G4,G1); + __ movxtod(G1,F40); + __ xor3(L3,G5,G1); + __ movxtod(G1,F42); + + // F60:F62 used for next 16-bytes + __ ldx(from,16,L4); + __ ldx(from,24,L5); + __ xor3(L2,L4,G1); + __ movxtod(G1,F60); + __ xor3(L3,L5,G1); + __ movxtod(G1,F62); + + for ( int i = 38; i >= 6; i -= 8 ) { + __ aes_dround23(as_FloatRegister(i), F40, F42, F44); + __ aes_dround01(as_FloatRegister(i-2), F40, F42, F46); + __ aes_dround23(as_FloatRegister(i), F60, F62, F58); + __ aes_dround01(as_FloatRegister(i-2), F60, F62, F56); + if (i != 6 ) { + __ aes_dround23(as_FloatRegister(i-4), F46, F44, F42); + __ aes_dround01(as_FloatRegister(i-6), F46, F44, F40); + __ aes_dround23(as_FloatRegister(i-4), F56, F58, F62); + __ aes_dround01(as_FloatRegister(i-6), F56, F58, F60); + } else { + __ aes_dround23_l(as_FloatRegister(i-4), F46, F44, F42); + __ aes_dround01_l(as_FloatRegister(i-6), F46, F44, F40); + __ aes_dround23_l(as_FloatRegister(i-4), F56, F58, F62); + __ aes_dround01_l(as_FloatRegister(i-6), F56, F58, F60); + } + } + + __ movxtod(L0,F46); + __ movxtod(L1,F44); + __ fxor(FloatRegisterImpl::D, F46, F40, F40); + __ fxor(FloatRegisterImpl::D, F44, F42, F42); + + __ stf(FloatRegisterImpl::D, F40, to, 0); + __ stf(FloatRegisterImpl::D, F42, to, 8); + + __ movxtod(G4,F56); + __ movxtod(G5,F58); + __ mov(L4,L0); + __ mov(L5,L1); + __ fxor(FloatRegisterImpl::D, F56, F60, F60); + __ fxor(FloatRegisterImpl::D, F58, F62, F62); + + __ stf(FloatRegisterImpl::D, F60, to, 16); + __ stf(FloatRegisterImpl::D, F62, to, 24); + + __ add(from, 32, from); + __ add(to, 32, to); + __ subcc(len_reg, 32, len_reg); + __ br(Assembler::notEqual, false, Assembler::pt, L_dec_next2_blocks128); + __ delayed()->nop(); + __ br(Assembler::always, false, Assembler::pt, L_cbcdec_end); + __ delayed()->nop(); + + __ align(OptoLoopAlignment); + __ BIND(L_dec_next2_blocks192); + __ nop(); + + // F48:F50 used for first 16-bytes + __ ldx(from,0,G4); + __ ldx(from,8,G5); + __ xor3(L2,G4,G1); + __ movxtod(G1,F48); + __ xor3(L3,G5,G1); + __ movxtod(G1,F50); + + // F60:F62 used for next 16-bytes + __ ldx(from,16,L4); + __ ldx(from,24,L5); + __ xor3(L2,L4,G1); + __ movxtod(G1,F60); + __ xor3(L3,L5,G1); + __ movxtod(G1,F62); + + for ( int i = 46; i >= 6; i -= 8 ) { + __ aes_dround23(as_FloatRegister(i), F48, F50, F52); + __ aes_dround01(as_FloatRegister(i-2), F48, F50, F54); + __ aes_dround23(as_FloatRegister(i), F60, F62, F58); + __ aes_dround01(as_FloatRegister(i-2), F60, F62, F56); + if (i != 6 ) { + __ aes_dround23(as_FloatRegister(i-4), F54, F52, F50); + __ aes_dround01(as_FloatRegister(i-6), F54, F52, F48); + __ aes_dround23(as_FloatRegister(i-4), F56, F58, F62); + __ aes_dround01(as_FloatRegister(i-6), F56, F58, F60); + } else { + __ aes_dround23_l(as_FloatRegister(i-4), F54, F52, F50); + __ aes_dround01_l(as_FloatRegister(i-6), F54, F52, F48); + __ aes_dround23_l(as_FloatRegister(i-4), F56, F58, F62); + __ aes_dround01_l(as_FloatRegister(i-6), F56, F58, F60); + } + } + + __ movxtod(L0,F54); + __ movxtod(L1,F52); + __ fxor(FloatRegisterImpl::D, F54, F48, F48); + __ fxor(FloatRegisterImpl::D, F52, F50, F50); + + __ stf(FloatRegisterImpl::D, F48, to, 0); + __ stf(FloatRegisterImpl::D, F50, to, 8); + + __ movxtod(G4,F56); + __ movxtod(G5,F58); + __ mov(L4,L0); + __ mov(L5,L1); + __ fxor(FloatRegisterImpl::D, F56, F60, F60); + __ fxor(FloatRegisterImpl::D, F58, F62, F62); + + __ stf(FloatRegisterImpl::D, F60, to, 16); + __ stf(FloatRegisterImpl::D, F62, to, 24); + + __ add(from, 32, from); + __ add(to, 32, to); + __ subcc(len_reg, 32, len_reg); + __ br(Assembler::notEqual, false, Assembler::pt, L_dec_next2_blocks192); + __ delayed()->nop(); + __ br(Assembler::always, false, Assembler::pt, L_cbcdec_end); + __ delayed()->nop(); + + __ align(OptoLoopAlignment); + __ BIND(L_dec_next2_blocks256); + __ nop(); + + // F0:F2 used for first 16-bytes + __ ldx(from,0,G4); + __ ldx(from,8,G5); + __ xor3(L2,G4,G1); + __ movxtod(G1,F0); + __ xor3(L3,G5,G1); + __ movxtod(G1,F2); + + // F60:F62 used for next 16-bytes + __ ldx(from,16,L4); + __ ldx(from,24,L5); + __ xor3(L2,L4,G1); + __ movxtod(G1,F60); + __ xor3(L3,L5,G1); + __ movxtod(G1,F62); + + __ aes_dround23(F54, F0, F2, F4); + __ aes_dround01(F52, F0, F2, F6); + __ aes_dround23(F54, F60, F62, F58); + __ aes_dround01(F52, F60, F62, F56); + __ aes_dround23(F50, F6, F4, F2); + __ aes_dround01(F48, F6, F4, F0); + __ aes_dround23(F50, F56, F58, F62); + __ aes_dround01(F48, F56, F58, F60); + // save F48:F54 in temp registers + __ movdtox(F54,G2); + __ movdtox(F52,G3); + __ movdtox(F50,G6); + __ movdtox(F48,G1); + for ( int i = 46; i >= 14; i -= 8 ) { + __ aes_dround23(as_FloatRegister(i), F0, F2, F4); + __ aes_dround01(as_FloatRegister(i-2), F0, F2, F6); + __ aes_dround23(as_FloatRegister(i), F60, F62, F58); + __ aes_dround01(as_FloatRegister(i-2), F60, F62, F56); + __ aes_dround23(as_FloatRegister(i-4), F6, F4, F2); + __ aes_dround01(as_FloatRegister(i-6), F6, F4, F0); + __ aes_dround23(as_FloatRegister(i-4), F56, F58, F62); + __ aes_dround01(as_FloatRegister(i-6), F56, F58, F60); + } + // init F48:F54 with F0:F6 values (original key) + __ ldf(FloatRegisterImpl::D, original_key, 0, F48); + __ ldf(FloatRegisterImpl::D, original_key, 8, F50); + __ ldf(FloatRegisterImpl::D, original_key, 16, F52); + __ ldf(FloatRegisterImpl::D, original_key, 24, F54); + __ aes_dround23(F54, F0, F2, F4); + __ aes_dround01(F52, F0, F2, F6); + __ aes_dround23(F54, F60, F62, F58); + __ aes_dround01(F52, F60, F62, F56); + __ aes_dround23_l(F50, F6, F4, F2); + __ aes_dround01_l(F48, F6, F4, F0); + __ aes_dround23_l(F50, F56, F58, F62); + __ aes_dround01_l(F48, F56, F58, F60); + // re-init F48:F54 with their original values + __ movxtod(G2,F54); + __ movxtod(G3,F52); + __ movxtod(G6,F50); + __ movxtod(G1,F48); + + __ movxtod(L0,F6); + __ movxtod(L1,F4); + __ fxor(FloatRegisterImpl::D, F6, F0, F0); + __ fxor(FloatRegisterImpl::D, F4, F2, F2); + + __ stf(FloatRegisterImpl::D, F0, to, 0); + __ stf(FloatRegisterImpl::D, F2, to, 8); + + __ movxtod(G4,F56); + __ movxtod(G5,F58); + __ mov(L4,L0); + __ mov(L5,L1); + __ fxor(FloatRegisterImpl::D, F56, F60, F60); + __ fxor(FloatRegisterImpl::D, F58, F62, F62); + + __ stf(FloatRegisterImpl::D, F60, to, 16); + __ stf(FloatRegisterImpl::D, F62, to, 24); + + __ add(from, 32, from); + __ add(to, 32, to); + __ subcc(len_reg, 32, len_reg); + __ br(Assembler::notEqual, false, Assembler::pt, L_dec_next2_blocks256); + __ delayed()->nop(); + + __ BIND(L_cbcdec_end); + __ stx(L0, rvec, 0); + __ stx(L1, rvec, 8); + __ restore(); + __ mov(L0, O0); + __ retl(); + __ delayed()->nop(); + + return start; + } + void generate_initial() { // Generates all stubs and initializes the entry points @@ -3368,6 +4137,14 @@ class StubGenerator: public StubCodeGenerator { generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry, &StubRoutines::_safefetchN_fault_pc, &StubRoutines::_safefetchN_continuation_pc); + + // generate AES intrinsics code + if (UseAESIntrinsics) { + StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock(); + StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock(); + StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt(); + StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel(); + } } diff --git a/src/cpu/sparc/vm/vm_version_sparc.cpp b/src/cpu/sparc/vm/vm_version_sparc.cpp index bae7b3510..b75d21f98 100644 --- a/src/cpu/sparc/vm/vm_version_sparc.cpp +++ b/src/cpu/sparc/vm/vm_version_sparc.cpp @@ -234,7 +234,7 @@ void VM_Version::initialize() { assert((OptoLoopAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size"); char buf[512]; - jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s%s%s%s%s%s", + jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s", (has_v9() ? ", v9" : (has_v8() ? ", v8" : "")), (has_hardware_popc() ? ", popc" : ""), (has_vis1() ? ", vis1" : ""), @@ -242,6 +242,7 @@ void VM_Version::initialize() { (has_vis3() ? ", vis3" : ""), (has_blk_init() ? ", blk_init" : ""), (has_cbcond() ? ", cbcond" : ""), + (has_aes() ? ", aes" : ""), (is_ultra3() ? ", ultra3" : ""), (is_sun4v() ? ", sun4v" : ""), (is_niagara_plus() ? ", niagara_plus" : (is_niagara() ? ", niagara" : "")), @@ -265,6 +266,41 @@ void VM_Version::initialize() { if (!has_vis1()) // Drop to 0 if no VIS1 support UseVIS = 0; + // T2 and above should have support for AES instructions + if (has_aes()) { + if (UseVIS > 0) { // AES intrinsics use FXOR instruction which is VIS1 + if (FLAG_IS_DEFAULT(UseAES)) { + FLAG_SET_DEFAULT(UseAES, true); + } + if (FLAG_IS_DEFAULT(UseAESIntrinsics)) { + FLAG_SET_DEFAULT(UseAESIntrinsics, true); + } + // we disable both the AES flags if either of them is disabled on the command line + if (!UseAES || !UseAESIntrinsics) { + FLAG_SET_DEFAULT(UseAES, false); + FLAG_SET_DEFAULT(UseAESIntrinsics, false); + } + } else { + if (UseAES || UseAESIntrinsics) { + warning("SPARC AES intrinsics require VIS1 instruction support. Intrinsics will be disabled."); + if (UseAES) { + FLAG_SET_DEFAULT(UseAES, false); + } + if (UseAESIntrinsics) { + FLAG_SET_DEFAULT(UseAESIntrinsics, false); + } + } + } + } else if (UseAES || UseAESIntrinsics) { + warning("AES instructions are not available on this CPU"); + if (UseAES) { + FLAG_SET_DEFAULT(UseAES, false); + } + if (UseAESIntrinsics) { + FLAG_SET_DEFAULT(UseAESIntrinsics, false); + } + } + if (FLAG_IS_DEFAULT(ContendedPaddingWidth) && (cache_line_size > ContendedPaddingWidth)) ContendedPaddingWidth = cache_line_size; diff --git a/src/cpu/sparc/vm/vm_version_sparc.hpp b/src/cpu/sparc/vm/vm_version_sparc.hpp index b8d63f4f8..6ddf57007 100644 --- a/src/cpu/sparc/vm/vm_version_sparc.hpp +++ b/src/cpu/sparc/vm/vm_version_sparc.hpp @@ -48,7 +48,8 @@ protected: sparc64_family = 14, M_family = 15, T_family = 16, - T1_model = 17 + T1_model = 17, + aes_instructions = 18 }; enum Feature_Flag_Set { @@ -73,6 +74,7 @@ protected: M_family_m = 1 << M_family, T_family_m = 1 << T_family, T1_model_m = 1 << T1_model, + aes_instructions_m = 1 << aes_instructions, generic_v8_m = v8_instructions_m | hardware_mul32_m | hardware_div32_m | hardware_fsmuld_m, generic_v9_m = generic_v8_m | v9_instructions_m, @@ -123,6 +125,7 @@ public: static bool has_vis3() { return (_features & vis3_instructions_m) != 0; } static bool has_blk_init() { return (_features & blk_init_instructions_m) != 0; } static bool has_cbcond() { return (_features & cbcond_instructions_m) != 0; } + static bool has_aes() { return (_features & aes_instructions_m) != 0; } static bool supports_compare_and_exchange() { return has_v9(); } diff --git a/src/cpu/x86/vm/stubGenerator_x86_32.cpp b/src/cpu/x86/vm/stubGenerator_x86_32.cpp index 12fccf9dd..1622fe5ff 100644 --- a/src/cpu/x86/vm/stubGenerator_x86_32.cpp +++ b/src/cpu/x86/vm/stubGenerator_x86_32.cpp @@ -2403,6 +2403,9 @@ class StubGenerator: public StubCodeGenerator { // c_rarg3 - r vector byte array address // c_rarg4 - input length // + // Output: + // rax - input length + // address generate_cipherBlockChaining_encryptAESCrypt() { assert(UseAES, "need AES instructions and misaligned SSE support"); __ align(CodeEntryAlignment); @@ -2483,7 +2486,7 @@ class StubGenerator: public StubCodeGenerator { __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object handleSOERegisters(false /*restoring*/); - __ movl(rax, 0); // return 0 (why?) + __ movptr(rax, len_param); // return length __ leave(); // required for proper stackwalking of RuntimeStub frame __ ret(0); @@ -2557,6 +2560,9 @@ class StubGenerator: public StubCodeGenerator { // c_rarg3 - r vector byte array address // c_rarg4 - input length // + // Output: + // rax - input length + // address generate_cipherBlockChaining_decryptAESCrypt() { assert(UseAES, "need AES instructions and misaligned SSE support"); @@ -2650,7 +2656,7 @@ class StubGenerator: public StubCodeGenerator { __ movptr(rvec , rvec_param); // restore this since used in loop __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object handleSOERegisters(false /*restoring*/); - __ movl(rax, 0); // return 0 (why?) + __ movptr(rax, len_param); // return length __ leave(); // required for proper stackwalking of RuntimeStub frame __ ret(0); diff --git a/src/cpu/x86/vm/stubGenerator_x86_64.cpp b/src/cpu/x86/vm/stubGenerator_x86_64.cpp index a16611280..0adb0d31e 100644 --- a/src/cpu/x86/vm/stubGenerator_x86_64.cpp +++ b/src/cpu/x86/vm/stubGenerator_x86_64.cpp @@ -3217,6 +3217,9 @@ class StubGenerator: public StubCodeGenerator { // c_rarg3 - r vector byte array address // c_rarg4 - input length // + // Output: + // rax - input length + // address generate_cipherBlockChaining_encryptAESCrypt() { assert(UseAES, "need AES instructions and misaligned SSE support"); __ align(CodeEntryAlignment); @@ -3232,7 +3235,7 @@ class StubGenerator: public StubCodeGenerator { #ifndef _WIN64 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16) #else - const Address len_mem(rsp, 6 * wordSize); // length is on stack on Win64 + const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64 const Register len_reg = r10; // pick the first volatile windows register #endif const Register pos = rax; @@ -3259,6 +3262,8 @@ class StubGenerator: public StubCodeGenerator { for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { __ movdqu(xmm_save(i), as_XMMRegister(i)); } +#else + __ push(len_reg); // Save #endif const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front @@ -3301,8 +3306,10 @@ class StubGenerator: public StubCodeGenerator { for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { __ movdqu(as_XMMRegister(i), xmm_save(i)); } + __ movl(rax, len_mem); +#else + __ pop(rax); // return length #endif - __ movl(rax, 0); // return 0 (why?) __ leave(); // required for proper stackwalking of RuntimeStub frame __ ret(0); @@ -3409,6 +3416,9 @@ class StubGenerator: public StubCodeGenerator { // c_rarg3 - r vector byte array address // c_rarg4 - input length // + // Output: + // rax - input length + // address generate_cipherBlockChaining_decryptAESCrypt_Parallel() { assert(UseAES, "need AES instructions and misaligned SSE support"); @@ -3427,7 +3437,7 @@ class StubGenerator: public StubCodeGenerator { #ifndef _WIN64 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16) #else - const Address len_mem(rsp, 6 * wordSize); // length is on stack on Win64 + const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64 const Register len_reg = r10; // pick the first volatile windows register #endif const Register pos = rax; @@ -3448,7 +3458,10 @@ class StubGenerator: public StubCodeGenerator { for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { __ movdqu(xmm_save(i), as_XMMRegister(i)); } +#else + __ push(len_reg); // Save #endif + // the java expanded key ordering is rotated one position from what we want // so we start from 0x10 here and hit 0x00 last const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front @@ -3554,8 +3567,10 @@ class StubGenerator: public StubCodeGenerator { for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) { __ movdqu(as_XMMRegister(i), xmm_save(i)); } + __ movl(rax, len_mem); +#else + __ pop(rax); // return length #endif - __ movl(rax, 0); // return 0 (why?) __ leave(); // required for proper stackwalking of RuntimeStub frame __ ret(0); diff --git a/src/cpu/x86/vm/x86.ad b/src/cpu/x86/vm/x86.ad index c49d0e6c3..b0077a5f5 100644 --- a/src/cpu/x86/vm/x86.ad +++ b/src/cpu/x86/vm/x86.ad @@ -581,6 +581,12 @@ const bool Matcher::misaligned_vectors_ok() { return !AlignVector; // can be changed by flag } +// x86 AES instructions are compatible with SunJCE expanded +// keys, hence we do not need to pass the original key to stubs +const bool Matcher::pass_original_key_for_aes() { + return false; +} + // Helper methods for MachSpillCopyNode::implementation(). static int vec_mov_helper(CodeBuffer *cbuf, bool do_size, int src_lo, int dst_lo, int src_hi, int dst_hi, uint ireg, outputStream* st) { diff --git a/src/os_cpu/solaris_sparc/vm/vm_version_solaris_sparc.cpp b/src/os_cpu/solaris_sparc/vm/vm_version_solaris_sparc.cpp index e46e21500..6dca05f91 100644 --- a/src/os_cpu/solaris_sparc/vm/vm_version_solaris_sparc.cpp +++ b/src/os_cpu/solaris_sparc/vm/vm_version_solaris_sparc.cpp @@ -119,6 +119,11 @@ int VM_Version::platform_features(int features) { #endif if (av & AV_SPARC_CBCOND) features |= cbcond_instructions_m; +#ifndef AV_SPARC_AES +#define AV_SPARC_AES 0x00020000 /* aes instrs supported */ +#endif + if (av & AV_SPARC_AES) features |= aes_instructions_m; + } else { // getisax(2) failed, use the old legacy code. #ifndef PRODUCT diff --git a/src/share/vm/classfile/classLoaderData.cpp b/src/share/vm/classfile/classLoaderData.cpp index 273e0184d..6b34e0150 100644 --- a/src/share/vm/classfile/classLoaderData.cpp +++ b/src/share/vm/classfile/classLoaderData.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2012, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -520,6 +520,13 @@ void ClassLoaderData::verify() { } } +bool ClassLoaderData::contains_klass(Klass* klass) { + for (Klass* k = _klasses; k != NULL; k = k->next_link()) { + if (k == klass) return true; + } + return false; +} + // GC root of class loader data created. ClassLoaderData* ClassLoaderDataGraph::_head = NULL; @@ -648,12 +655,12 @@ GrowableArray<ClassLoaderData*>* ClassLoaderDataGraph::new_clds() { return array; } -#ifndef PRODUCT -// for debugging and hsfind(x) -bool ClassLoaderDataGraph::contains(address x) { - // I think we need the _metaspace_lock taken here because the class loader - // data graph could be changing while we are walking it (new entries added, - // new entries being unloaded, etc). +// For profiling and hsfind() only. Otherwise, this is unsafe (and slow). This +// is done lock free to avoid lock inversion problems. It is safe because +// new ClassLoaderData are added to the end of the CLDG, and only removed at +// safepoint. The _unloading list can be deallocated concurrently with CMS so +// this doesn't look in metaspace for classes that have been unloaded. +bool ClassLoaderDataGraph::contains(const void* x) { if (DumpSharedSpaces) { // There are only two metaspaces to worry about. ClassLoaderData* ncld = ClassLoaderData::the_null_class_loader_data(); @@ -670,16 +677,11 @@ bool ClassLoaderDataGraph::contains(address x) { } } - // Could also be on an unloading list which is okay, ie. still allocated - // for a little while. - for (ClassLoaderData* ucld = _unloading; ucld != NULL; ucld = ucld->next()) { - if (ucld->metaspace_or_null() != NULL && ucld->metaspace_or_null()->contains(x)) { - return true; - } - } + // Do not check unloading list because deallocation can be concurrent. return false; } +#ifndef PRODUCT bool ClassLoaderDataGraph::contains_loader_data(ClassLoaderData* loader_data) { for (ClassLoaderData* data = _head; data != NULL; data = data->next()) { if (loader_data == data) { diff --git a/src/share/vm/classfile/classLoaderData.hpp b/src/share/vm/classfile/classLoaderData.hpp index 2cb21801d..3fcbb6b7d 100644 --- a/src/share/vm/classfile/classLoaderData.hpp +++ b/src/share/vm/classfile/classLoaderData.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2012, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2012, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -90,9 +90,9 @@ class ClassLoaderDataGraph : public AllStatic { static void dump() { dump_on(tty); } static void verify(); -#ifndef PRODUCT // expensive test for pointer in metaspace for debugging - static bool contains(address x); + static bool contains(const void* x); +#ifndef PRODUCT static bool contains_loader_data(ClassLoaderData* loader_data); #endif @@ -260,6 +260,7 @@ class ClassLoaderData : public CHeapObj<mtClass> { jobject add_handle(Handle h); void add_class(Klass* k); void remove_class(Klass* k); + bool contains_klass(Klass* k); void record_dependency(Klass* to, TRAPS); void init_dependencies(TRAPS); diff --git a/src/share/vm/classfile/dictionary.cpp b/src/share/vm/classfile/dictionary.cpp index e308791ca..9a2bb74f8 100644 --- a/src/share/vm/classfile/dictionary.cpp +++ b/src/share/vm/classfile/dictionary.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2003, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -707,7 +707,7 @@ void Dictionary::verify() { loader_data->class_loader() == NULL || loader_data->class_loader()->is_instance(), "checking type of class_loader"); - e->verify(/*check_dictionary*/false); + e->verify(); probe->verify_protection_domain_set(); element_count++; } diff --git a/src/share/vm/classfile/systemDictionary.cpp b/src/share/vm/classfile/systemDictionary.cpp index f5c5c017d..703443517 100644 --- a/src/share/vm/classfile/systemDictionary.cpp +++ b/src/share/vm/classfile/systemDictionary.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -2650,23 +2650,6 @@ void SystemDictionary::verify() { constraints()->verify(dictionary(), placeholders()); } - -void SystemDictionary::verify_obj_klass_present(Symbol* class_name, - ClassLoaderData* loader_data) { - GCMutexLocker mu(SystemDictionary_lock); - Symbol* name; - - Klass* probe = find_class(class_name, loader_data); - if (probe == NULL) { - probe = SystemDictionary::find_shared_class(class_name); - if (probe == NULL) { - name = find_placeholder(class_name, loader_data); - } - } - guarantee(probe != NULL || name != NULL, - "Loaded klasses should be in SystemDictionary"); -} - // utility function for class load event void SystemDictionary::post_class_load_event(const Ticks& start_time, instanceKlassHandle k, diff --git a/src/share/vm/classfile/systemDictionary.hpp b/src/share/vm/classfile/systemDictionary.hpp index b0e914ff9..88132f520 100644 --- a/src/share/vm/classfile/systemDictionary.hpp +++ b/src/share/vm/classfile/systemDictionary.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -375,10 +375,6 @@ public: static bool is_internal_format(Symbol* class_name); #endif - // Verify class is in dictionary - static void verify_obj_klass_present(Symbol* class_name, - ClassLoaderData* loader_data); - // Initialization static void initialize(TRAPS); diff --git a/src/share/vm/classfile/vmSymbols.hpp b/src/share/vm/classfile/vmSymbols.hpp index 58485959f..ed3c0dbcb 100644 --- a/src/share/vm/classfile/vmSymbols.hpp +++ b/src/share/vm/classfile/vmSymbols.hpp @@ -787,7 +787,7 @@ do_intrinsic(_cipherBlockChaining_decryptAESCrypt, com_sun_crypto_provider_cipherBlockChaining, decrypt_name, byteArray_int_int_byteArray_int_signature, F_R) \ do_name( encrypt_name, "encrypt") \ do_name( decrypt_name, "decrypt") \ - do_signature(byteArray_int_int_byteArray_int_signature, "([BII[BI)V") \ + do_signature(byteArray_int_int_byteArray_int_signature, "([BII[BI)I") \ \ /* support for java.util.zip */ \ do_class(java_util_zip_CRC32, "java/util/zip/CRC32") \ diff --git a/src/share/vm/code/dependencies.cpp b/src/share/vm/code/dependencies.cpp index 29aaf3250..778efe8fe 100644 --- a/src/share/vm/code/dependencies.cpp +++ b/src/share/vm/code/dependencies.cpp @@ -655,8 +655,6 @@ inline Metadata* Dependencies::DepStream::recorded_metadata_at(int i) { } else { o = _deps->oop_recorder()->metadata_at(i); } - assert(o == NULL || o->is_metaspace_object(), - err_msg("Should be metadata " PTR_FORMAT, o)); return o; } diff --git a/src/share/vm/code/vtableStubs.cpp b/src/share/vm/code/vtableStubs.cpp index 5af91a3a6..b3bfc258f 100644 --- a/src/share/vm/code/vtableStubs.cpp +++ b/src/share/vm/code/vtableStubs.cpp @@ -55,6 +55,9 @@ void* VtableStub::operator new(size_t size, int code_size) throw() { const int chunk_factor = 32; if (_chunk == NULL || _chunk + real_size > _chunk_end) { const int bytes = chunk_factor * real_size + pd_code_alignment(); + + // There is a dependency on the name of the blob in src/share/vm/prims/jvmtiCodeBlobEvents.cpp + // If changing the name, update the other file accordingly. BufferBlob* blob = BufferBlob::create("vtable chunks", bytes); if (blob == NULL) { return NULL; @@ -62,12 +65,6 @@ void* VtableStub::operator new(size_t size, int code_size) throw() { _chunk = blob->content_begin(); _chunk_end = _chunk + bytes; Forte::register_stub("vtable stub", _chunk, _chunk_end); - // Notify JVMTI about this stub. The event will be recorded by the enclosing - // JvmtiDynamicCodeEventCollector and posted when this thread has released - // all locks. - if (JvmtiExport::should_post_dynamic_code_generated()) { - JvmtiExport::post_dynamic_code_generated_while_holding_locks("vtable stub", _chunk, _chunk_end); - } align_chunk(); } assert(_chunk + real_size <= _chunk_end, "bad allocation"); @@ -130,6 +127,13 @@ address VtableStubs::find_stub(bool is_vtable_stub, int vtable_index) { is_vtable_stub? "vtbl": "itbl", vtable_index, VtableStub::receiver_location()); Disassembler::decode(s->code_begin(), s->code_end()); } + // Notify JVMTI about this stub. The event will be recorded by the enclosing + // JvmtiDynamicCodeEventCollector and posted when this thread has released + // all locks. + if (JvmtiExport::should_post_dynamic_code_generated()) { + JvmtiExport::post_dynamic_code_generated_while_holding_locks(is_vtable_stub? "vtable stub": "itable stub", + s->code_begin(), s->code_end()); + } } return s->entry_point(); } @@ -195,6 +199,14 @@ void vtableStubs_init() { VtableStubs::initialize(); } +void VtableStubs::vtable_stub_do(void f(VtableStub*)) { + for (int i = 0; i < N; i++) { + for (VtableStub* s = _table[i]; s != NULL; s = s->next()) { + f(s); + } + } +} + //----------------------------------------------------------------------------------------------------- // Non-product code diff --git a/src/share/vm/code/vtableStubs.hpp b/src/share/vm/code/vtableStubs.hpp index 06f2a67a8..b3d4f2d83 100644 --- a/src/share/vm/code/vtableStubs.hpp +++ b/src/share/vm/code/vtableStubs.hpp @@ -131,6 +131,7 @@ class VtableStubs : AllStatic { static VtableStub* stub_containing(address pc); // stub containing pc or NULL static int number_of_vtable_stubs() { return _number_of_vtable_stubs; } static void initialize(); + static void vtable_stub_do(void f(VtableStub*)); // iterates over all vtable stubs }; #endif // SHARE_VM_CODE_VTABLESTUBS_HPP diff --git a/src/share/vm/interpreter/linkResolver.cpp b/src/share/vm/interpreter/linkResolver.cpp index 52c88bd3a..aa11784b7 100644 --- a/src/share/vm/interpreter/linkResolver.cpp +++ b/src/share/vm/interpreter/linkResolver.cpp @@ -564,16 +564,7 @@ void LinkResolver::resolve_method(methodHandle& resolved_method, KlassHandle res } } - // 5. check if method is concrete - if (resolved_method->is_abstract() && !resolved_klass->is_abstract()) { - ResourceMark rm(THREAD); - THROW_MSG(vmSymbols::java_lang_AbstractMethodError(), - Method::name_and_sig_as_C_string(resolved_klass(), - method_name, - method_signature)); - } - - // 6. access checks, access checking may be turned off when calling from within the VM. + // 5. access checks, access checking may be turned off when calling from within the VM. if (check_access) { assert(current_klass.not_null() , "current_klass should not be null"); diff --git a/src/share/vm/interpreter/rewriter.cpp b/src/share/vm/interpreter/rewriter.cpp index 208f8617d..2474ae758 100644 --- a/src/share/vm/interpreter/rewriter.cpp +++ b/src/share/vm/interpreter/rewriter.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1998, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1998, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -250,8 +250,8 @@ void Rewriter::rewrite_invokedynamic(address bcp, int offset, bool reverse) { // We will reverse the bytecode rewriting _after_ adjusting them. // Adjust the cache index by offset to the invokedynamic entries in the // cpCache plus the delta if the invokedynamic bytecodes were adjusted. - cache_index = cp_cache_delta() + _first_iteration_cp_cache_limit; - int cp_index = invokedynamic_cp_cache_entry_pool_index(cache_index); + int adjustment = cp_cache_delta() + _first_iteration_cp_cache_limit; + int cp_index = invokedynamic_cp_cache_entry_pool_index(cache_index - adjustment); assert(_pool->tag_at(cp_index).is_invoke_dynamic(), "wrong index"); // zero out 4 bytes Bytes::put_Java_u4(p, 0); @@ -453,18 +453,7 @@ methodHandle Rewriter::rewrite_jsrs(methodHandle method, TRAPS) { return method; } -void Rewriter::rewrite(instanceKlassHandle klass, TRAPS) { - ResourceMark rm(THREAD); - Rewriter rw(klass, klass->constants(), klass->methods(), CHECK); - // (That's all, folks.) -} - - -Rewriter::Rewriter(instanceKlassHandle klass, constantPoolHandle cpool, Array<Method*>* methods, TRAPS) - : _klass(klass), - _pool(cpool), - _methods(methods) -{ +void Rewriter::rewrite_bytecodes(TRAPS) { assert(_pool->cache() == NULL, "constant pool cache must not be set yet"); // determine index maps for Method* rewriting @@ -508,6 +497,29 @@ Rewriter::Rewriter(instanceKlassHandle klass, constantPoolHandle cpool, Array<Me // May have to fix invokedynamic bytecodes if invokestatic/InterfaceMethodref // entries had to be added. patch_invokedynamic_bytecodes(); +} + +void Rewriter::rewrite(instanceKlassHandle klass, TRAPS) { + ResourceMark rm(THREAD); + Rewriter rw(klass, klass->constants(), klass->methods(), CHECK); + // (That's all, folks.) +} + + +Rewriter::Rewriter(instanceKlassHandle klass, constantPoolHandle cpool, Array<Method*>* methods, TRAPS) + : _klass(klass), + _pool(cpool), + _methods(methods) +{ + + // Rewrite bytecodes - exception here exits. + rewrite_bytecodes(CHECK); + + // Stress restoring bytecodes + if (StressRewriter) { + restore_bytecodes(); + rewrite_bytecodes(CHECK); + } // allocate constant pool cache, now that we've seen all the bytecodes make_constant_pool_cache(THREAD); @@ -523,6 +535,7 @@ Rewriter::Rewriter(instanceKlassHandle klass, constantPoolHandle cpool, Array<Me // so methods with jsrs in custom class lists in aren't attempted to be // rewritten in the RO section of the shared archive. // Relocated bytecodes don't have to be restored, only the cp cache entries + int len = _methods->length(); for (int i = len-1; i >= 0; i--) { methodHandle m(THREAD, _methods->at(i)); diff --git a/src/share/vm/interpreter/rewriter.hpp b/src/share/vm/interpreter/rewriter.hpp index 2c7990fa7..aa4b7cd52 100644 --- a/src/share/vm/interpreter/rewriter.hpp +++ b/src/share/vm/interpreter/rewriter.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1998, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1998, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -199,6 +199,9 @@ class Rewriter: public StackObj { void patch_invokedynamic_bytecodes(); + // Do all the work. + void rewrite_bytecodes(TRAPS); + // Revert bytecodes in case of an exception. void restore_bytecodes(); diff --git a/src/share/vm/memory/allocation.cpp b/src/share/vm/memory/allocation.cpp index 73d1d7858..8e382ba85 100644 --- a/src/share/vm/memory/allocation.cpp +++ b/src/share/vm/memory/allocation.cpp @@ -74,9 +74,8 @@ bool MetaspaceObj::is_shared() const { return MetaspaceShared::is_in_shared_space(this); } - bool MetaspaceObj::is_metaspace_object() const { - return Metaspace::contains((void*)this); + return ClassLoaderDataGraph::contains((void*)this); } void MetaspaceObj::print_address_on(outputStream* st) const { diff --git a/src/share/vm/memory/allocation.hpp b/src/share/vm/memory/allocation.hpp index feb482add..d635a0798 100644 --- a/src/share/vm/memory/allocation.hpp +++ b/src/share/vm/memory/allocation.hpp @@ -267,7 +267,7 @@ class ClassLoaderData; class MetaspaceObj { public: - bool is_metaspace_object() const; // more specific test but slower + bool is_metaspace_object() const; bool is_shared() const; void print_address_on(outputStream* st) const; // nonvirtual address printing diff --git a/src/share/vm/memory/metachunk.hpp b/src/share/vm/memory/metachunk.hpp index 3bd4d15bc..e873dc6a3 100644 --- a/src/share/vm/memory/metachunk.hpp +++ b/src/share/vm/memory/metachunk.hpp @@ -143,6 +143,8 @@ class Metachunk : public Metabase<Metachunk> { void set_is_tagged_free(bool v) { _is_tagged_free = v; } #endif + bool contains(const void* ptr) { return bottom() <= ptr && ptr < _top; } + NOT_PRODUCT(void mangle();) void print_on(outputStream* st) const; diff --git a/src/share/vm/memory/metaspace.cpp b/src/share/vm/memory/metaspace.cpp index 19c03661a..5a7e4ec2f 100644 --- a/src/share/vm/memory/metaspace.cpp +++ b/src/share/vm/memory/metaspace.cpp @@ -513,8 +513,6 @@ class VirtualSpaceList : public CHeapObj<mtClass> { // Unlink empty VirtualSpaceNodes and free it. void purge(ChunkManager* chunk_manager); - bool contains(const void *ptr); - void print_on(outputStream* st) const; class VirtualSpaceListIterator : public StackObj { @@ -558,7 +556,7 @@ class SpaceManager : public CHeapObj<mtClass> { private: - // protects allocations and contains. + // protects allocations Mutex* const _lock; // Type of metadata allocated. @@ -595,7 +593,11 @@ class SpaceManager : public CHeapObj<mtClass> { private: // Accessors Metachunk* chunks_in_use(ChunkIndex index) const { return _chunks_in_use[index]; } - void set_chunks_in_use(ChunkIndex index, Metachunk* v) { _chunks_in_use[index] = v; } + void set_chunks_in_use(ChunkIndex index, Metachunk* v) { + // ensure lock-free iteration sees fully initialized node + OrderAccess::storestore(); + _chunks_in_use[index] = v; + } BlockFreelist* block_freelists() const { return (BlockFreelist*) &_block_freelists; @@ -708,6 +710,8 @@ class SpaceManager : public CHeapObj<mtClass> { void print_on(outputStream* st) const; void locked_print_chunks_in_use_on(outputStream* st) const; + bool contains(const void *ptr); + void verify(); void verify_chunk_size(Metachunk* chunk); NOT_PRODUCT(void mangle_freed_chunks();) @@ -1159,8 +1163,6 @@ bool VirtualSpaceList::create_new_virtual_space(size_t vs_word_size) { } else { assert(new_entry->reserved_words() == vs_word_size, "Reserved memory size differs from requested memory size"); - // ensure lock-free iteration sees fully initialized node - OrderAccess::storestore(); link_vs(new_entry); return true; } @@ -1287,19 +1289,6 @@ void VirtualSpaceList::print_on(outputStream* st) const { } } -bool VirtualSpaceList::contains(const void *ptr) { - VirtualSpaceNode* list = virtual_space_list(); - VirtualSpaceListIterator iter(list); - while (iter.repeat()) { - VirtualSpaceNode* node = iter.get_next(); - if (node->reserved()->contains(ptr)) { - return true; - } - } - return false; -} - - // MetaspaceGC methods // VM_CollectForMetadataAllocation is the vm operation used to GC. @@ -2392,6 +2381,21 @@ MetaWord* SpaceManager::allocate_work(size_t word_size) { return result; } +// This function looks at the chunks in the metaspace without locking. +// The chunks are added with store ordering and not deleted except for at +// unloading time. +bool SpaceManager::contains(const void *ptr) { + for (ChunkIndex i = ZeroIndex; i < NumberOfInUseLists; i = next_chunk_index(i)) + { + Metachunk* curr = chunks_in_use(i); + while (curr != NULL) { + if (curr->contains(ptr)) return true; + curr = curr->next(); + } + } + return false; +} + void SpaceManager::verify() { // If there are blocks in the dictionary, then // verfication of chunks does not work since @@ -3463,17 +3467,12 @@ void Metaspace::print_on(outputStream* out) const { } } -bool Metaspace::contains(const void * ptr) { - if (MetaspaceShared::is_in_shared_space(ptr)) { - return true; +bool Metaspace::contains(const void* ptr) { + if (vsm()->contains(ptr)) return true; + if (using_class_space()) { + return class_vsm()->contains(ptr); } - // This is checked while unlocked. As long as the virtualspaces are added - // at the end, the pointer will be in one of them. The virtual spaces - // aren't deleted presently. When they are, some sort of locking might - // be needed. Note, locking this can cause inversion problems with the - // caller in MetaspaceObj::is_metadata() function. - return space_list()->contains(ptr) || - (using_class_space() && class_space_list()->contains(ptr)); + return false; } void Metaspace::verify() { diff --git a/src/share/vm/memory/metaspace.hpp b/src/share/vm/memory/metaspace.hpp index b98e69a41..f74cec355 100644 --- a/src/share/vm/memory/metaspace.hpp +++ b/src/share/vm/memory/metaspace.hpp @@ -226,7 +226,7 @@ class Metaspace : public CHeapObj<mtClass> { MetaWord* expand_and_allocate(size_t size, MetadataType mdtype); - static bool contains(const void *ptr); + bool contains(const void* ptr); void dump(outputStream* const out) const; // Free empty virtualspaces diff --git a/src/share/vm/oops/arrayKlass.cpp b/src/share/vm/oops/arrayKlass.cpp index 9e40206c2..fcb46b11f 100644 --- a/src/share/vm/oops/arrayKlass.cpp +++ b/src/share/vm/oops/arrayKlass.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -214,8 +214,8 @@ void ArrayKlass::oop_print_on(oop obj, outputStream* st) { // Verification -void ArrayKlass::verify_on(outputStream* st, bool check_dictionary) { - Klass::verify_on(st, check_dictionary); +void ArrayKlass::verify_on(outputStream* st) { + Klass::verify_on(st); if (component_mirror() != NULL) { guarantee(component_mirror()->klass() != NULL, "should have a class"); diff --git a/src/share/vm/oops/arrayKlass.hpp b/src/share/vm/oops/arrayKlass.hpp index 4b06f1c0e..7b4ad2e9a 100644 --- a/src/share/vm/oops/arrayKlass.hpp +++ b/src/share/vm/oops/arrayKlass.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -146,7 +146,7 @@ class ArrayKlass: public Klass { void oop_print_on(oop obj, outputStream* st); // Verification - void verify_on(outputStream* st, bool check_dictionary); + void verify_on(outputStream* st); void oop_verify_on(oop obj, outputStream* st); }; diff --git a/src/share/vm/oops/constantPool.cpp b/src/share/vm/oops/constantPool.cpp index 16b53a699..3f37810d2 100644 --- a/src/share/vm/oops/constantPool.cpp +++ b/src/share/vm/oops/constantPool.cpp @@ -82,6 +82,9 @@ ConstantPool::ConstantPool(Array<u1>* tags) { void ConstantPool::deallocate_contents(ClassLoaderData* loader_data) { MetadataFactory::free_metadata(loader_data, cache()); set_cache(NULL); + MetadataFactory::free_array<u2>(loader_data, reference_map()); + set_reference_map(NULL); + MetadataFactory::free_array<jushort>(loader_data, operands()); set_operands(NULL); diff --git a/src/share/vm/oops/instanceKlass.cpp b/src/share/vm/oops/instanceKlass.cpp index 4a27fc153..27ae4ff03 100644 --- a/src/share/vm/oops/instanceKlass.cpp +++ b/src/share/vm/oops/instanceKlass.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -3184,7 +3184,7 @@ class VerifyFieldClosure: public OopClosure { virtual void do_oop(narrowOop* p) { VerifyFieldClosure::do_oop_work(p); } }; -void InstanceKlass::verify_on(outputStream* st, bool check_dictionary) { +void InstanceKlass::verify_on(outputStream* st) { #ifndef PRODUCT // Avoid redundant verifies, this really should be in product. if (_verify_count == Universe::verify_count()) return; @@ -3192,14 +3192,11 @@ void InstanceKlass::verify_on(outputStream* st, bool check_dictionary) { #endif // Verify Klass - Klass::verify_on(st, check_dictionary); + Klass::verify_on(st); - // Verify that klass is present in SystemDictionary if not already - // verifying the SystemDictionary. - if (is_loaded() && !is_anonymous() && check_dictionary) { - Symbol* h_name = name(); - SystemDictionary::verify_obj_klass_present(h_name, class_loader_data()); - } + // Verify that klass is present in ClassLoaderData + guarantee(class_loader_data()->contains_klass(this), + "this class isn't found in class loader data"); // Verify vtables if (is_linked()) { diff --git a/src/share/vm/oops/instanceKlass.hpp b/src/share/vm/oops/instanceKlass.hpp index 2b7a73a31..ba2bce874 100644 --- a/src/share/vm/oops/instanceKlass.hpp +++ b/src/share/vm/oops/instanceKlass.hpp @@ -1086,7 +1086,7 @@ public: const char* internal_name() const; // Verification - void verify_on(outputStream* st, bool check_dictionary); + void verify_on(outputStream* st); void oop_verify_on(oop obj, outputStream* st); }; diff --git a/src/share/vm/oops/klass.cpp b/src/share/vm/oops/klass.cpp index 22b570bbf..adca246aa 100644 --- a/src/share/vm/oops/klass.cpp +++ b/src/share/vm/oops/klass.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -376,8 +376,6 @@ void Klass::append_to_sibling_list() { } bool Klass::is_loader_alive(BoolObjectClosure* is_alive) { - assert(ClassLoaderDataGraph::contains((address)this), "is in the metaspace"); - #ifdef ASSERT // The class is alive iff the class loader is alive. oop loader = class_loader(); @@ -640,7 +638,7 @@ void Klass::collect_statistics(KlassSizeStats *sz) const { // Verification -void Klass::verify_on(outputStream* st, bool check_dictionary) { +void Klass::verify_on(outputStream* st) { // This can be expensive, but it is worth checking that this klass is actually // in the CLD graph but not in production. diff --git a/src/share/vm/oops/klass.hpp b/src/share/vm/oops/klass.hpp index 9855fdd32..a31a23573 100644 --- a/src/share/vm/oops/klass.hpp +++ b/src/share/vm/oops/klass.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -695,8 +695,8 @@ class Klass : public Metadata { virtual const char* internal_name() const = 0; // Verification - virtual void verify_on(outputStream* st, bool check_dictionary); - void verify(bool check_dictionary = true) { verify_on(tty, check_dictionary); } + virtual void verify_on(outputStream* st); + void verify() { verify_on(tty); } #ifndef PRODUCT bool verify_vtable_index(int index); diff --git a/src/share/vm/oops/objArrayKlass.cpp b/src/share/vm/oops/objArrayKlass.cpp index 7294ebe9a..3316a8b27 100644 --- a/src/share/vm/oops/objArrayKlass.cpp +++ b/src/share/vm/oops/objArrayKlass.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -674,8 +674,8 @@ const char* ObjArrayKlass::internal_name() const { // Verification -void ObjArrayKlass::verify_on(outputStream* st, bool check_dictionary) { - ArrayKlass::verify_on(st, check_dictionary); +void ObjArrayKlass::verify_on(outputStream* st) { + ArrayKlass::verify_on(st); guarantee(element_klass()->is_klass(), "should be klass"); guarantee(bottom_klass()->is_klass(), "should be klass"); Klass* bk = bottom_klass(); diff --git a/src/share/vm/oops/objArrayKlass.hpp b/src/share/vm/oops/objArrayKlass.hpp index af06fd420..cfe31e86e 100644 --- a/src/share/vm/oops/objArrayKlass.hpp +++ b/src/share/vm/oops/objArrayKlass.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -151,7 +151,7 @@ class ObjArrayKlass : public ArrayKlass { const char* internal_name() const; // Verification - void verify_on(outputStream* st, bool check_dictionary); + void verify_on(outputStream* st); void oop_verify_on(oop obj, outputStream* st); }; diff --git a/src/share/vm/opto/bytecodeInfo.cpp b/src/share/vm/opto/bytecodeInfo.cpp index 4b3f39f08..101d5f733 100644 --- a/src/share/vm/opto/bytecodeInfo.cpp +++ b/src/share/vm/opto/bytecodeInfo.cpp @@ -63,34 +63,14 @@ InlineTree::InlineTree(Compile* c, assert(_caller_jvms->same_calls_as(caller_jvms), "consistent JVMS"); assert((caller_tree == NULL ? 0 : caller_tree->stack_depth() + 1) == stack_depth(), "correct (redundant) depth parameter"); assert(caller_bci == this->caller_bci(), "correct (redundant) bci parameter"); - if (UseOldInlining) { - // Update hierarchical counts, count_inline_bcs() and count_inlines() - InlineTree *caller = (InlineTree *)caller_tree; - for( ; caller != NULL; caller = ((InlineTree *)(caller->caller_tree())) ) { - caller->_count_inline_bcs += count_inline_bcs(); - NOT_PRODUCT(caller->_count_inlines++;) - } + // Update hierarchical counts, count_inline_bcs() and count_inlines() + InlineTree *caller = (InlineTree *)caller_tree; + for( ; caller != NULL; caller = ((InlineTree *)(caller->caller_tree())) ) { + caller->_count_inline_bcs += count_inline_bcs(); + NOT_PRODUCT(caller->_count_inlines++;) } } -InlineTree::InlineTree(Compile* c, ciMethod* callee_method, JVMState* caller_jvms, - float site_invoke_ratio, int max_inline_level) : - C(c), - _caller_jvms(caller_jvms), - _caller_tree(NULL), - _method(callee_method), - _site_invoke_ratio(site_invoke_ratio), - _max_inline_level(max_inline_level), - _count_inline_bcs(method()->code_size()), - _msg(NULL) -{ -#ifndef PRODUCT - _count_inlines = 0; - _forced_inline = false; -#endif - assert(!UseOldInlining, "do not use for old stuff"); -} - /** * Return true when EA is ON and a java constructor is called or * a super constructor is called from an inlined java constructor. @@ -161,11 +141,6 @@ bool InlineTree::should_inline(ciMethod* callee_method, ciMethod* caller_method, return true; } - if (!UseOldInlining) { - set_msg("!UseOldInlining"); - return true; // size and frequency are represented in a new way - } - int default_max_inline_size = C->max_inline_size(); int inline_small_code_size = InlineSmallCode / 4; int max_inline_size = default_max_inline_size; @@ -229,35 +204,6 @@ bool InlineTree::should_not_inline(ciMethod *callee_method, fail_msg = "don't inline by annotation"; } - if (!UseOldInlining) { - if (fail_msg != NULL) { - *wci_result = *(WarmCallInfo::always_cold()); - set_msg(fail_msg); - return true; - } - - if (callee_method->has_unloaded_classes_in_signature()) { - wci_result->set_profit(wci_result->profit() * 0.1); - } - - // don't inline exception code unless the top method belongs to an - // exception class - if (callee_method->holder()->is_subclass_of(C->env()->Throwable_klass())) { - ciMethod* top_method = jvms->caller() != NULL ? jvms->caller()->of_depth(1)->method() : method(); - if (!top_method->holder()->is_subclass_of(C->env()->Throwable_klass())) { - wci_result->set_profit(wci_result->profit() * 0.1); - } - } - - if (callee_method->has_compiled_code() && - callee_method->instructions_size() > InlineSmallCode) { - wci_result->set_profit(wci_result->profit() * 0.1); - // %%% adjust wci_result->size()? - } - - return false; - } - // one more inlining restriction if (fail_msg == NULL && callee_method->has_unloaded_classes_in_signature()) { fail_msg = "unloaded signature classes"; @@ -360,9 +306,7 @@ bool InlineTree::try_to_inline(ciMethod* callee_method, ciMethod* caller_method, int caller_bci, JVMState* jvms, ciCallProfile& profile, WarmCallInfo* wci_result, bool& should_delay) { - // Old algorithm had funny accumulating BC-size counters - if (UseOldInlining && ClipInlining - && (int)count_inline_bcs() >= DesiredMethodLimit) { + if (ClipInlining && (int)count_inline_bcs() >= DesiredMethodLimit) { if (!callee_method->force_inline() || !IncrementalInline) { set_msg("size > DesiredMethodLimit"); return false; @@ -465,8 +409,7 @@ bool InlineTree::try_to_inline(ciMethod* callee_method, ciMethod* caller_method, int size = callee_method->code_size_for_inlining(); - if (UseOldInlining && ClipInlining - && (int)count_inline_bcs() + size >= DesiredMethodLimit) { + if (ClipInlining && (int)count_inline_bcs() + size >= DesiredMethodLimit) { if (!callee_method->force_inline() || !IncrementalInline) { set_msg("size > DesiredMethodLimit"); return false; @@ -584,8 +527,7 @@ WarmCallInfo* InlineTree::ok_to_inline(ciMethod* callee_method, JVMState* jvms, jvms, profile, &wci, should_delay); #ifndef PRODUCT - if (UseOldInlining && InlineWarmCalls - && (PrintOpto || C->print_inlining())) { + if (InlineWarmCalls && (PrintOpto || C->print_inlining())) { bool cold = wci.is_cold(); bool hot = !cold && wci.is_hot(); bool old_cold = !success; @@ -599,13 +541,12 @@ WarmCallInfo* InlineTree::ok_to_inline(ciMethod* callee_method, JVMState* jvms, } } #endif - if (UseOldInlining) { - if (success) { - wci = *(WarmCallInfo::always_hot()); - } else { - wci = *(WarmCallInfo::always_cold()); - } + if (success) { + wci = *(WarmCallInfo::always_hot()); + } else { + wci = *(WarmCallInfo::always_cold()); } + if (!InlineWarmCalls) { if (!wci.is_cold() && !wci.is_hot()) { // Do not inline the warm calls. @@ -619,8 +560,7 @@ WarmCallInfo* InlineTree::ok_to_inline(ciMethod* callee_method, JVMState* jvms, set_msg("inline (hot)"); } print_inlining(callee_method, caller_bci, true /* success */); - if (UseOldInlining) - build_inline_tree_for_callee(callee_method, jvms, caller_bci); + build_inline_tree_for_callee(callee_method, jvms, caller_bci); if (InlineWarmCalls && !wci.is_hot()) return new (C) WarmCallInfo(wci); // copy to heap return WarmCallInfo::always_hot(); diff --git a/src/share/vm/opto/c2_globals.hpp b/src/share/vm/opto/c2_globals.hpp index da7ab96f7..d9829d3c6 100644 --- a/src/share/vm/opto/c2_globals.hpp +++ b/src/share/vm/opto/c2_globals.hpp @@ -357,9 +357,6 @@ "File to dump ideal graph to. If set overrides the " \ "use of the network") \ \ - product(bool, UseOldInlining, true, \ - "Enable the 1.3 inlining strategy") \ - \ product(bool, UseBimorphicInlining, true, \ "Profiling based inlining for two receivers") \ \ diff --git a/src/share/vm/opto/callGenerator.cpp b/src/share/vm/opto/callGenerator.cpp index 86e48e849..a655f27b5 100644 --- a/src/share/vm/opto/callGenerator.cpp +++ b/src/share/vm/opto/callGenerator.cpp @@ -722,7 +722,7 @@ JVMState* PredictedCallGenerator::generate(JVMState* jvms, Parse* parent_parser) Node* m = kit.map()->in(i); Node* n = slow_map->in(i); if (m != n) { - const Type* t = gvn.type(m)->meet(gvn.type(n)); + const Type* t = gvn.type(m)->meet_speculative(gvn.type(n)); Node* phi = PhiNode::make(region, m, t); phi->set_req(2, n); kit.map()->set_req(i, gvn.transform(phi)); @@ -975,7 +975,7 @@ JVMState* PredictedIntrinsicGenerator::generate(JVMState* jvms, Parse* parent_pa Node* m = kit.map()->in(i); Node* n = slow_map->in(i); if (m != n) { - const Type* t = gvn.type(m)->meet(gvn.type(n)); + const Type* t = gvn.type(m)->meet_speculative(gvn.type(n)); Node* phi = PhiNode::make(region, m, t); phi->set_req(2, n); kit.map()->set_req(i, gvn.transform(phi)); diff --git a/src/share/vm/opto/cfgnode.cpp b/src/share/vm/opto/cfgnode.cpp index 36818b75b..25223035f 100644 --- a/src/share/vm/opto/cfgnode.cpp +++ b/src/share/vm/opto/cfgnode.cpp @@ -951,7 +951,7 @@ const Type *PhiNode::Value( PhaseTransform *phase ) const { if (is_intf != ti_is_intf) { t = _type; break; } } - t = t->meet(ti); + t = t->meet_speculative(ti); } } @@ -968,11 +968,11 @@ const Type *PhiNode::Value( PhaseTransform *phase ) const { // // It is not possible to see Type::BOTTOM values as phi inputs, // because the ciTypeFlow pre-pass produces verifier-quality types. - const Type* ft = t->filter(_type); // Worst case type + const Type* ft = t->filter_speculative(_type); // Worst case type #ifdef ASSERT // The following logic has been moved into TypeOopPtr::filter. - const Type* jt = t->join(_type); + const Type* jt = t->join_speculative(_type); if( jt->empty() ) { // Emptied out??? // Check for evil case of 't' being a class and '_type' expecting an @@ -1757,7 +1757,7 @@ Node *PhiNode::Ideal(PhaseGVN *phase, bool can_reshape) { break; } // Accumulate type for resulting Phi - type = type->meet(in(i)->in(AddPNode::Base)->bottom_type()); + type = type->meet_speculative(in(i)->in(AddPNode::Base)->bottom_type()); } Node* base = NULL; if (doit) { diff --git a/src/share/vm/opto/compile.cpp b/src/share/vm/opto/compile.cpp index 1f4f4f683..d838a5b6a 100644 --- a/src/share/vm/opto/compile.cpp +++ b/src/share/vm/opto/compile.cpp @@ -705,10 +705,7 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr print_compile_messages(); - if (UseOldInlining || PrintCompilation NOT_PRODUCT( || PrintOpto) ) - _ilt = InlineTree::build_inline_tree_root(); - else - _ilt = NULL; + _ilt = InlineTree::build_inline_tree_root(); // Even if NO memory addresses are used, MergeMem nodes must have at least 1 slice assert(num_alias_types() >= AliasIdxRaw, ""); @@ -3948,16 +3945,18 @@ void Compile::remove_speculative_types(PhaseIterGVN &igvn) { // which may optimize it out. for (uint next = 0; next < worklist.size(); ++next) { Node *n = worklist.at(next); - if (n->is_Type() && n->as_Type()->type()->isa_oopptr() != NULL && - n->as_Type()->type()->is_oopptr()->speculative() != NULL) { + if (n->is_Type()) { TypeNode* tn = n->as_Type(); - const TypeOopPtr* t = tn->type()->is_oopptr(); - bool in_hash = igvn.hash_delete(n); - assert(in_hash, "node should be in igvn hash table"); - tn->set_type(t->remove_speculative()); - igvn.hash_insert(n); - igvn._worklist.push(n); // give it a chance to go away - modified++; + const Type* t = tn->type(); + const Type* t_no_spec = t->remove_speculative(); + if (t_no_spec != t) { + bool in_hash = igvn.hash_delete(n); + assert(in_hash, "node should be in igvn hash table"); + tn->set_type(t_no_spec); + igvn.hash_insert(n); + igvn._worklist.push(n); // give it a chance to go away + modified++; + } } uint max = n->len(); for( uint i = 0; i < max; ++i ) { @@ -3971,6 +3970,27 @@ void Compile::remove_speculative_types(PhaseIterGVN &igvn) { if (modified > 0) { igvn.optimize(); } +#ifdef ASSERT + // Verify that after the IGVN is over no speculative type has resurfaced + worklist.clear(); + worklist.push(root()); + for (uint next = 0; next < worklist.size(); ++next) { + Node *n = worklist.at(next); + const Type* t = igvn.type(n); + assert(t == t->remove_speculative(), "no more speculative types"); + if (n->is_Type()) { + t = n->as_Type()->type(); + assert(t == t->remove_speculative(), "no more speculative types"); + } + uint max = n->len(); + for( uint i = 0; i < max; ++i ) { + Node *m = n->in(i); + if (not_a_node(m)) continue; + worklist.push(m); + } + } + igvn.check_no_speculative_types(); +#endif } } diff --git a/src/share/vm/opto/connode.cpp b/src/share/vm/opto/connode.cpp index 948ff7b46..0a71bc598 100644 --- a/src/share/vm/opto/connode.cpp +++ b/src/share/vm/opto/connode.cpp @@ -188,7 +188,7 @@ Node *CMoveNode::Identity( PhaseTransform *phase ) { const Type *CMoveNode::Value( PhaseTransform *phase ) const { if( phase->type(in(Condition)) == Type::TOP ) return Type::TOP; - return phase->type(in(IfFalse))->meet(phase->type(in(IfTrue))); + return phase->type(in(IfFalse))->meet_speculative(phase->type(in(IfTrue))); } //------------------------------make------------------------------------------- @@ -392,14 +392,14 @@ Node *CMoveDNode::Ideal(PhaseGVN *phase, bool can_reshape) { //============================================================================= // If input is already higher or equal to cast type, then this is an identity. Node *ConstraintCastNode::Identity( PhaseTransform *phase ) { - return phase->type(in(1))->higher_equal(_type) ? in(1) : this; + return phase->type(in(1))->higher_equal_speculative(_type) ? in(1) : this; } //------------------------------Value------------------------------------------ // Take 'join' of input and cast-up type const Type *ConstraintCastNode::Value( PhaseTransform *phase ) const { if( in(0) && phase->type(in(0)) == Type::TOP ) return Type::TOP; - const Type* ft = phase->type(in(1))->filter(_type); +const Type* ft = phase->type(in(1))->filter_speculative(_type); #ifdef ASSERT // Previous versions of this function had some special case logic, @@ -409,7 +409,7 @@ const Type *ConstraintCastNode::Value( PhaseTransform *phase ) const { { const Type* t1 = phase->type(in(1)); if( t1 == Type::TOP ) assert(ft == Type::TOP, "special case #1"); - const Type* rt = t1->join(_type); + const Type* rt = t1->join_speculative(_type); if (rt->empty()) assert(ft == Type::TOP, "special case #2"); break; } diff --git a/src/share/vm/opto/connode.hpp b/src/share/vm/opto/connode.hpp index 53ff820d8..03a096b23 100644 --- a/src/share/vm/opto/connode.hpp +++ b/src/share/vm/opto/connode.hpp @@ -36,7 +36,7 @@ class MachNode; // Simple constants class ConNode : public TypeNode { public: - ConNode( const Type *t ) : TypeNode(t,1) { + ConNode( const Type *t ) : TypeNode(t->remove_speculative(),1) { init_req(0, (Node*)Compile::current()->root()); init_flags(Flag_is_Con); } diff --git a/src/share/vm/opto/doCall.cpp b/src/share/vm/opto/doCall.cpp index 223f7da36..e888b550f 100644 --- a/src/share/vm/opto/doCall.cpp +++ b/src/share/vm/opto/doCall.cpp @@ -161,19 +161,8 @@ CallGenerator* Compile::call_generator(ciMethod* callee, int vtable_index, bool // Try inlining a bytecoded method: if (!call_does_dispatch) { - InlineTree* ilt; - if (UseOldInlining) { - ilt = InlineTree::find_subtree_from_root(this->ilt(), jvms->caller(), jvms->method()); - } else { - // Make a disembodied, stateless ILT. - // TO DO: When UseOldInlining is removed, copy the ILT code elsewhere. - float site_invoke_ratio = prof_factor; - // Note: ilt is for the root of this parse, not the present call site. - ilt = new InlineTree(this, jvms->method(), jvms->caller(), site_invoke_ratio, MaxInlineLevel); - } + InlineTree* ilt = InlineTree::find_subtree_from_root(this->ilt(), jvms->caller(), jvms->method()); WarmCallInfo scratch_ci; - if (!UseOldInlining) - scratch_ci.init(jvms, callee, profile, prof_factor); bool should_delay = false; WarmCallInfo* ci = ilt->ok_to_inline(callee, jvms, profile, &scratch_ci, should_delay); assert(ci != &scratch_ci, "do not let this pointer escape"); diff --git a/src/share/vm/opto/graphKit.cpp b/src/share/vm/opto/graphKit.cpp index e2d7e385c..af4b2f452 100644 --- a/src/share/vm/opto/graphKit.cpp +++ b/src/share/vm/opto/graphKit.cpp @@ -420,7 +420,7 @@ void GraphKit::combine_exception_states(SafePointNode* ex_map, SafePointNode* ph } const Type* srctype = _gvn.type(src); if (phi->type() != srctype) { - const Type* dsttype = phi->type()->meet(srctype); + const Type* dsttype = phi->type()->meet_speculative(srctype); if (phi->type() != dsttype) { phi->set_type(dsttype); _gvn.set_type(phi, dsttype); @@ -1224,7 +1224,7 @@ Node* GraphKit::null_check_common(Node* value, BasicType type, // See if mixing in the NULL pointer changes type. // If so, then the NULL pointer was not allowed in the original // type. In other words, "value" was not-null. - if (t->meet(TypePtr::NULL_PTR) != t) { + if (t->meet(TypePtr::NULL_PTR) != t->remove_speculative()) { // same as: if (!TypePtr::NULL_PTR->higher_equal(t)) ... explicit_null_checks_elided++; return value; // Elided null check quickly! @@ -1357,7 +1357,7 @@ Node* GraphKit::null_check_common(Node* value, BasicType type, // Cast obj to not-null on this path Node* GraphKit::cast_not_null(Node* obj, bool do_replace_in_map) { const Type *t = _gvn.type(obj); - const Type *t_not_null = t->join(TypePtr::NOTNULL); + const Type *t_not_null = t->join_speculative(TypePtr::NOTNULL); // Object is already not-null? if( t == t_not_null ) return obj; @@ -3014,7 +3014,7 @@ Node* GraphKit::gen_checkcast(Node *obj, Node* superklass, if (failure_control != NULL) // failure is now impossible (*failure_control) = top(); // adjust the type of the phi to the exact klass: - phi->raise_bottom_type(_gvn.type(cast_obj)->meet(TypePtr::NULL_PTR)); + phi->raise_bottom_type(_gvn.type(cast_obj)->meet_speculative(TypePtr::NULL_PTR)); } } diff --git a/src/share/vm/opto/library_call.cpp b/src/share/vm/opto/library_call.cpp index 66141a1f9..3a2279ee5 100644 --- a/src/share/vm/opto/library_call.cpp +++ b/src/share/vm/opto/library_call.cpp @@ -304,6 +304,7 @@ class LibraryCallKit : public GraphKit { bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id); Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting); Node* get_key_start_from_aescrypt_object(Node* aescrypt_object); + Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object); bool inline_encodeISOArray(); bool inline_updateCRC32(); bool inline_updateBytesCRC32(); @@ -5945,10 +5946,22 @@ bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) { Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object); if (k_start == NULL) return false; - // Call the stub. - make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), - stubAddr, stubName, TypePtr::BOTTOM, - src_start, dest_start, k_start); + if (Matcher::pass_original_key_for_aes()) { + // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to + // compatibility issues between Java key expansion and SPARC crypto instructions + Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); + if (original_k_start == NULL) return false; + + // Call the stub. + make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), + stubAddr, stubName, TypePtr::BOTTOM, + src_start, dest_start, k_start, original_k_start); + } else { + // Call the stub. + make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(), + stubAddr, stubName, TypePtr::BOTTOM, + src_start, dest_start, k_start); + } return true; } @@ -6026,14 +6039,29 @@ bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) { if (objRvec == NULL) return false; Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE); - // Call the stub, passing src_start, dest_start, k_start, r_start and src_len - make_runtime_call(RC_LEAF|RC_NO_FP, - OptoRuntime::cipherBlockChaining_aescrypt_Type(), - stubAddr, stubName, TypePtr::BOTTOM, - src_start, dest_start, k_start, r_start, len); + Node* cbcCrypt; + if (Matcher::pass_original_key_for_aes()) { + // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to + // compatibility issues between Java key expansion and SPARC crypto instructions + Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object); + if (original_k_start == NULL) return false; - // return is void so no result needs to be pushed + // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start + cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, + OptoRuntime::cipherBlockChaining_aescrypt_Type(), + stubAddr, stubName, TypePtr::BOTTOM, + src_start, dest_start, k_start, r_start, len, original_k_start); + } else { + // Call the stub, passing src_start, dest_start, k_start, r_start and src_len + cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP, + OptoRuntime::cipherBlockChaining_aescrypt_Type(), + stubAddr, stubName, TypePtr::BOTTOM, + src_start, dest_start, k_start, r_start, len); + } + // return cipher length (int) + Node* retvalue = _gvn.transform(new (C) ProjNode(cbcCrypt, TypeFunc::Parms)); + set_result(retvalue); return true; } @@ -6048,6 +6076,17 @@ Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) return k_start; } +//------------------------------get_original_key_start_from_aescrypt_object----------------------- +Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) { + Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false); + assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt"); + if (objAESCryptKey == NULL) return (Node *) NULL; + + // now have the array, need to get the start address of the lastKey array + Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE); + return original_k_start; +} + //----------------------------inline_cipherBlockChaining_AESCrypt_predicate---------------------------- // Return node representing slow path of predicate check. // the pseudo code we want to emulate with this predicate is: diff --git a/src/share/vm/opto/loopopts.cpp b/src/share/vm/opto/loopopts.cpp index ba19f7493..ac97d3ede 100644 --- a/src/share/vm/opto/loopopts.cpp +++ b/src/share/vm/opto/loopopts.cpp @@ -1115,8 +1115,8 @@ BoolNode *PhaseIdealLoop::clone_iff( PhiNode *phi, IdealLoopTree *loop ) { Node *n2 = phi->in(i)->in(1)->in(2); phi1->set_req( i, n1 ); phi2->set_req( i, n2 ); - phi1->set_type( phi1->type()->meet(n1->bottom_type()) ); - phi2->set_type( phi2->type()->meet(n2->bottom_type()) ); + phi1->set_type( phi1->type()->meet_speculative(n1->bottom_type())); + phi2->set_type( phi2->type()->meet_speculative(n2->bottom_type())); } // See if these Phis have been made before. // Register with optimizer @@ -1189,8 +1189,8 @@ CmpNode *PhaseIdealLoop::clone_bool( PhiNode *phi, IdealLoopTree *loop ) { } phi1->set_req( j, n1 ); phi2->set_req( j, n2 ); - phi1->set_type( phi1->type()->meet(n1->bottom_type()) ); - phi2->set_type( phi2->type()->meet(n2->bottom_type()) ); + phi1->set_type(phi1->type()->meet_speculative(n1->bottom_type())); + phi2->set_type(phi2->type()->meet_speculative(n2->bottom_type())); } // See if these Phis have been made before. diff --git a/src/share/vm/opto/matcher.hpp b/src/share/vm/opto/matcher.hpp index df641495d..6d0c8e3b0 100644 --- a/src/share/vm/opto/matcher.hpp +++ b/src/share/vm/opto/matcher.hpp @@ -286,6 +286,9 @@ public: // CPU supports misaligned vectors store/load. static const bool misaligned_vectors_ok(); + // Should original key array reference be passed to AES stubs + static const bool pass_original_key_for_aes(); + // Used to determine a "low complexity" 64-bit constant. (Zero is simple.) // The standard of comparison is one (StoreL ConL) vs. two (StoreI ConI). // Depends on the details of 64-bit constant generation on the CPU. diff --git a/src/share/vm/opto/memnode.cpp b/src/share/vm/opto/memnode.cpp index 4cfe229fd..8097eade8 100644 --- a/src/share/vm/opto/memnode.cpp +++ b/src/share/vm/opto/memnode.cpp @@ -657,7 +657,7 @@ const TypePtr* MemNode::calculate_adr_type(const Type* t, const TypePtr* cross_c // disregarding "null"-ness. // (We make an exception for TypeRawPtr::BOTTOM, which is a bit bucket.) const TypePtr* tp_notnull = tp->join(TypePtr::NOTNULL)->is_ptr(); - assert(cross_check->meet(tp_notnull) == cross_check, + assert(cross_check->meet(tp_notnull) == cross_check->remove_speculative(), "real address must not escape from expected memory type"); } #endif @@ -1685,7 +1685,7 @@ const Type *LoadNode::Value( PhaseTransform *phase ) const { // t might actually be lower than _type, if _type is a unique // concrete subclass of abstract class t. if (off_beyond_header) { // is the offset beyond the header? - const Type* jt = t->join(_type); + const Type* jt = t->join_speculative(_type); // In any case, do not allow the join, per se, to empty out the type. if (jt->empty() && !t->empty()) { // This can happen if a interface-typed array narrows to a class type. diff --git a/src/share/vm/opto/multnode.cpp b/src/share/vm/opto/multnode.cpp index 106a0086c..d9e04a274 100644 --- a/src/share/vm/opto/multnode.cpp +++ b/src/share/vm/opto/multnode.cpp @@ -94,7 +94,7 @@ const Type* ProjNode::proj_type(const Type* t) const { if ((_con == TypeFunc::Parms) && n->is_CallStaticJava() && n->as_CallStaticJava()->is_boxing_method()) { // The result of autoboxing is always non-null on normal path. - t = t->join(TypePtr::NOTNULL); + t = t->join_speculative(TypePtr::NOTNULL); } return t; } diff --git a/src/share/vm/opto/node.cpp b/src/share/vm/opto/node.cpp index 7de4f8881..2ba71f6bc 100644 --- a/src/share/vm/opto/node.cpp +++ b/src/share/vm/opto/node.cpp @@ -995,13 +995,13 @@ void Node::raise_bottom_type(const Type* new_type) { if (is_Type()) { TypeNode *n = this->as_Type(); if (VerifyAliases) { - assert(new_type->higher_equal(n->type()), "new type must refine old type"); + assert(new_type->higher_equal_speculative(n->type()), "new type must refine old type"); } n->set_type(new_type); } else if (is_Load()) { LoadNode *n = this->as_Load(); if (VerifyAliases) { - assert(new_type->higher_equal(n->type()), "new type must refine old type"); + assert(new_type->higher_equal_speculative(n->type()), "new type must refine old type"); } n->set_type(new_type); } diff --git a/src/share/vm/opto/parse1.cpp b/src/share/vm/opto/parse1.cpp index a873e75ba..74b1f1a4a 100644 --- a/src/share/vm/opto/parse1.cpp +++ b/src/share/vm/opto/parse1.cpp @@ -1656,7 +1656,7 @@ void Parse::merge_common(Parse::Block* target, int pnum) { assert(bt1 != Type::BOTTOM, "should not be building conflict phis"); map()->set_req(j, _gvn.transform_no_reclaim(phi)); debug_only(const Type* bt2 = phi->bottom_type()); - assert(bt2->higher_equal(bt1), "must be consistent with type-flow"); + assert(bt2->higher_equal_speculative(bt1), "must be consistent with type-flow"); record_for_igvn(phi); } } @@ -2029,7 +2029,7 @@ void Parse::return_current(Node* value) { !tp->klass()->is_interface()) { // sharpen the type eagerly; this eases certain assert checking if (tp->higher_equal(TypeInstPtr::NOTNULL)) - tr = tr->join(TypeInstPtr::NOTNULL)->is_instptr(); + tr = tr->join_speculative(TypeInstPtr::NOTNULL)->is_instptr(); value = _gvn.transform(new (C) CheckCastPPNode(0,value,tr)); } } diff --git a/src/share/vm/opto/parse2.cpp b/src/share/vm/opto/parse2.cpp index a341780e0..f82d246c3 100644 --- a/src/share/vm/opto/parse2.cpp +++ b/src/share/vm/opto/parse2.cpp @@ -88,7 +88,7 @@ Node* Parse::array_addressing(BasicType type, int vals, const Type* *result2) { if (toop->klass()->as_instance_klass()->unique_concrete_subklass()) { // If we load from "AbstractClass[]" we must see "ConcreteSubClass". const Type* subklass = Type::get_const_type(toop->klass()); - elemtype = subklass->join(el); + elemtype = subklass->join_speculative(el); } } } @@ -1278,7 +1278,7 @@ void Parse::sharpen_type_after_if(BoolTest::mask btest, // Bool(CmpP(LoadKlass(obj._klass), ConP(Foo.klass)), [eq]) // or the narrowOop equivalent. const Type* obj_type = _gvn.type(obj); - const TypeOopPtr* tboth = obj_type->join(con_type)->isa_oopptr(); + const TypeOopPtr* tboth = obj_type->join_speculative(con_type)->isa_oopptr(); if (tboth != NULL && tboth->klass_is_exact() && tboth != obj_type && tboth->higher_equal(obj_type)) { // obj has to be of the exact type Foo if the CmpP succeeds. @@ -1288,7 +1288,7 @@ void Parse::sharpen_type_after_if(BoolTest::mask btest, (jvms->is_loc(obj_in_map) || jvms->is_stk(obj_in_map))) { TypeNode* ccast = new (C) CheckCastPPNode(control(), obj, tboth); const Type* tcc = ccast->as_Type()->type(); - assert(tcc != obj_type && tcc->higher_equal(obj_type), "must improve"); + assert(tcc != obj_type && tcc->higher_equal_speculative(obj_type), "must improve"); // Delay transform() call to allow recovery of pre-cast value // at the control merge. _gvn.set_type_bottom(ccast); @@ -1318,7 +1318,7 @@ void Parse::sharpen_type_after_if(BoolTest::mask btest, switch (btest) { case BoolTest::eq: // Constant test? { - const Type* tboth = tcon->join(tval); + const Type* tboth = tcon->join_speculative(tval); if (tboth == tval) break; // Nothing to gain. if (tcon->isa_int()) { ccast = new (C) CastIINode(val, tboth); @@ -1352,7 +1352,7 @@ void Parse::sharpen_type_after_if(BoolTest::mask btest, if (ccast != NULL) { const Type* tcc = ccast->as_Type()->type(); - assert(tcc != tval && tcc->higher_equal(tval), "must improve"); + assert(tcc != tval && tcc->higher_equal_speculative(tval), "must improve"); // Delay transform() call to allow recovery of pre-cast value // at the control merge. ccast->set_req(0, control()); diff --git a/src/share/vm/opto/parse3.cpp b/src/share/vm/opto/parse3.cpp index 3deb9f426..466e1ca3b 100644 --- a/src/share/vm/opto/parse3.cpp +++ b/src/share/vm/opto/parse3.cpp @@ -361,7 +361,7 @@ bool Parse::push_constant(ciConstant constant, bool require_constant, bool is_au // should_be_constant = (oop not scavengable || ScavengeRootsInCode >= 2) // An oop is not scavengable if it is in the perm gen. if (stable_type != NULL && con_type != NULL && con_type->isa_oopptr()) - con_type = con_type->join(stable_type); + con_type = con_type->join_speculative(stable_type); break; case T_ILLEGAL: diff --git a/src/share/vm/opto/phaseX.cpp b/src/share/vm/opto/phaseX.cpp index 3e6ab4eaf..fd08b065c 100644 --- a/src/share/vm/opto/phaseX.cpp +++ b/src/share/vm/opto/phaseX.cpp @@ -323,6 +323,23 @@ void NodeHash::remove_useless_nodes(VectorSet &useful) { } } + +void NodeHash::check_no_speculative_types() { +#ifdef ASSERT + uint max = size(); + Node *sentinel_node = sentinel(); + for (uint i = 0; i < max; ++i) { + Node *n = at(i); + if(n != NULL && n != sentinel_node && n->is_Type()) { + TypeNode* tn = n->as_Type(); + const Type* t = tn->type(); + const Type* t_no_spec = t->remove_speculative(); + assert(t == t_no_spec, "dead node in hash table or missed node during speculative cleanup"); + } + } +#endif +} + #ifndef PRODUCT //------------------------------dump------------------------------------------- // Dump statistics for the hash table @@ -1392,11 +1409,11 @@ void PhaseIterGVN::remove_speculative_types() { assert(UseTypeSpeculation, "speculation is off"); for (uint i = 0; i < _types.Size(); i++) { const Type* t = _types.fast_lookup(i); - if (t != NULL && t->isa_oopptr()) { - const TypeOopPtr* to = t->is_oopptr(); - _types.map(i, to->remove_speculative()); + if (t != NULL) { + _types.map(i, t->remove_speculative()); } } + _table.check_no_speculative_types(); } //============================================================================= diff --git a/src/share/vm/opto/phaseX.hpp b/src/share/vm/opto/phaseX.hpp index d03d47d95..fcdd47ee8 100644 --- a/src/share/vm/opto/phaseX.hpp +++ b/src/share/vm/opto/phaseX.hpp @@ -92,7 +92,8 @@ public: } void remove_useless_nodes(VectorSet &useful); // replace with sentinel - void replace_with(NodeHash* nh); + void replace_with(NodeHash* nh); + void check_no_speculative_types(); // Check no speculative part for type nodes in table Node *sentinel() { return _sentinel; } @@ -501,6 +502,9 @@ public: Deoptimization::DeoptReason reason); void remove_speculative_types(); + void check_no_speculative_types() { + _table.check_no_speculative_types(); + } #ifndef PRODUCT protected: diff --git a/src/share/vm/opto/runtime.cpp b/src/share/vm/opto/runtime.cpp index 0b439c45c..da775791a 100644 --- a/src/share/vm/opto/runtime.cpp +++ b/src/share/vm/opto/runtime.cpp @@ -826,12 +826,18 @@ const TypeFunc* OptoRuntime::array_fill_Type() { const TypeFunc* OptoRuntime::aescrypt_block_Type() { // create input type (domain) int num_args = 3; + if (Matcher::pass_original_key_for_aes()) { + num_args = 4; + } int argcnt = num_args; const Type** fields = TypeTuple::fields(argcnt); int argp = TypeFunc::Parms; fields[argp++] = TypePtr::NOTNULL; // src fields[argp++] = TypePtr::NOTNULL; // dest fields[argp++] = TypePtr::NOTNULL; // k array + if (Matcher::pass_original_key_for_aes()) { + fields[argp++] = TypePtr::NOTNULL; // original k array + } assert(argp == TypeFunc::Parms+argcnt, "correct decoding"); const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields); @@ -868,6 +874,9 @@ const TypeFunc* OptoRuntime::updateBytesCRC32_Type() { const TypeFunc* OptoRuntime::cipherBlockChaining_aescrypt_Type() { // create input type (domain) int num_args = 5; + if (Matcher::pass_original_key_for_aes()) { + num_args = 6; + } int argcnt = num_args; const Type** fields = TypeTuple::fields(argcnt); int argp = TypeFunc::Parms; @@ -876,13 +885,16 @@ const TypeFunc* OptoRuntime::cipherBlockChaining_aescrypt_Type() { fields[argp++] = TypePtr::NOTNULL; // k array fields[argp++] = TypePtr::NOTNULL; // r array fields[argp++] = TypeInt::INT; // src len + if (Matcher::pass_original_key_for_aes()) { + fields[argp++] = TypePtr::NOTNULL; // original k array + } assert(argp == TypeFunc::Parms+argcnt, "correct decoding"); const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields); - // no result type needed + // returning cipher len (int) fields = TypeTuple::fields(1); - fields[TypeFunc::Parms+0] = NULL; // void - const TypeTuple* range = TypeTuple::make(TypeFunc::Parms, fields); + fields[TypeFunc::Parms+0] = TypeInt::INT; + const TypeTuple* range = TypeTuple::make(TypeFunc::Parms+1, fields); return TypeFunc::make(domain, range); } diff --git a/src/share/vm/opto/type.cpp b/src/share/vm/opto/type.cpp index 0e1c4e216..c907fc712 100644 --- a/src/share/vm/opto/type.cpp +++ b/src/share/vm/opto/type.cpp @@ -241,6 +241,13 @@ int Type::cmp( const Type *const t1, const Type *const t2 ) { return !t1->eq(t2); // Return ZERO if equal } +const Type* Type::maybe_remove_speculative(bool include_speculative) const { + if (!include_speculative) { + return remove_speculative(); + } + return this; +} + //------------------------------hash------------------------------------------- int Type::uhash( const Type *const t ) { return t->hash(); @@ -633,41 +640,44 @@ bool Type::interface_vs_oop(const Type *t) const { //------------------------------meet------------------------------------------- // Compute the MEET of two types. NOT virtual. It enforces that meet is // commutative and the lattice is symmetric. -const Type *Type::meet( const Type *t ) const { +const Type *Type::meet_helper(const Type *t, bool include_speculative) const { if (isa_narrowoop() && t->isa_narrowoop()) { - const Type* result = make_ptr()->meet(t->make_ptr()); + const Type* result = make_ptr()->meet_helper(t->make_ptr(), include_speculative); return result->make_narrowoop(); } if (isa_narrowklass() && t->isa_narrowklass()) { - const Type* result = make_ptr()->meet(t->make_ptr()); + const Type* result = make_ptr()->meet_helper(t->make_ptr(), include_speculative); return result->make_narrowklass(); } - const Type *mt = xmeet(t); + const Type *this_t = maybe_remove_speculative(include_speculative); + t = t->maybe_remove_speculative(include_speculative); + + const Type *mt = this_t->xmeet(t); if (isa_narrowoop() || t->isa_narrowoop()) return mt; if (isa_narrowklass() || t->isa_narrowklass()) return mt; #ifdef ASSERT - assert( mt == t->xmeet(this), "meet not commutative" ); + assert(mt == t->xmeet(this_t), "meet not commutative"); const Type* dual_join = mt->_dual; const Type *t2t = dual_join->xmeet(t->_dual); - const Type *t2this = dual_join->xmeet( _dual); + const Type *t2this = dual_join->xmeet(this_t->_dual); // Interface meet Oop is Not Symmetric: // Interface:AnyNull meet Oop:AnyNull == Interface:AnyNull // Interface:NotNull meet Oop:NotNull == java/lang/Object:NotNull - if( !interface_vs_oop(t) && (t2t != t->_dual || t2this != _dual) ) { + if( !interface_vs_oop(t) && (t2t != t->_dual || t2this != this_t->_dual) ) { tty->print_cr("=== Meet Not Symmetric ==="); - tty->print("t = "); t->dump(); tty->cr(); - tty->print("this= "); dump(); tty->cr(); - tty->print("mt=(t meet this)= "); mt->dump(); tty->cr(); + tty->print("t = "); t->dump(); tty->cr(); + tty->print("this= "); this_t->dump(); tty->cr(); + tty->print("mt=(t meet this)= "); mt->dump(); tty->cr(); - tty->print("t_dual= "); t->_dual->dump(); tty->cr(); - tty->print("this_dual= "); _dual->dump(); tty->cr(); - tty->print("mt_dual= "); mt->_dual->dump(); tty->cr(); + tty->print("t_dual= "); t->_dual->dump(); tty->cr(); + tty->print("this_dual= "); this_t->_dual->dump(); tty->cr(); + tty->print("mt_dual= "); mt->_dual->dump(); tty->cr(); - tty->print("mt_dual meet t_dual= "); t2t ->dump(); tty->cr(); - tty->print("mt_dual meet this_dual= "); t2this ->dump(); tty->cr(); + tty->print("mt_dual meet t_dual= "); t2t ->dump(); tty->cr(); + tty->print("mt_dual meet this_dual= "); t2this ->dump(); tty->cr(); fatal("meet not symmetric" ); } @@ -759,8 +769,8 @@ const Type *Type::xmeet( const Type *t ) const { } //-----------------------------filter------------------------------------------ -const Type *Type::filter( const Type *kills ) const { - const Type* ft = join(kills); +const Type *Type::filter_helper(const Type *kills, bool include_speculative) const { + const Type* ft = join_helper(kills, include_speculative); if (ft->empty()) return Type::TOP; // Canonical empty value return ft; @@ -1314,8 +1324,8 @@ const Type *TypeInt::narrow( const Type *old ) const { } //-----------------------------filter------------------------------------------ -const Type *TypeInt::filter( const Type *kills ) const { - const TypeInt* ft = join(kills)->isa_int(); +const Type *TypeInt::filter_helper(const Type *kills, bool include_speculative) const { + const TypeInt* ft = join_helper(kills, include_speculative)->isa_int(); if (ft == NULL || ft->empty()) return Type::TOP; // Canonical empty value if (ft->_widen < this->_widen) { @@ -1575,8 +1585,8 @@ const Type *TypeLong::narrow( const Type *old ) const { } //-----------------------------filter------------------------------------------ -const Type *TypeLong::filter( const Type *kills ) const { - const TypeLong* ft = join(kills)->isa_long(); +const Type *TypeLong::filter_helper(const Type *kills, bool include_speculative) const { + const TypeLong* ft = join_helper(kills, include_speculative)->isa_long(); if (ft == NULL || ft->empty()) return Type::TOP; // Canonical empty value if (ft->_widen < this->_widen) { @@ -1731,7 +1741,7 @@ const TypeTuple *TypeTuple::make_domain(ciInstanceKlass* recv, ciSignature* sig) total_fields++; field_array = fields(total_fields); // Use get_const_type here because it respects UseUniqueSubclasses: - field_array[pos++] = get_const_type(recv)->join(TypePtr::NOTNULL); + field_array[pos++] = get_const_type(recv)->join_speculative(TypePtr::NOTNULL); } else { field_array = fields(total_fields); } @@ -1921,7 +1931,7 @@ const Type *TypeAry::xmeet( const Type *t ) const { case Array: { // Meeting 2 arrays? const TypeAry *a = t->is_ary(); - return TypeAry::make(_elem->meet(a->_elem), + return TypeAry::make(_elem->meet_speculative(a->_elem), _size->xmeet(a->_size)->is_int(), _stable & a->_stable); } @@ -1954,6 +1964,13 @@ int TypeAry::hash(void) const { return (intptr_t)_elem + (intptr_t)_size + (_stable ? 43 : 0); } +/** + * Return same type without a speculative part in the element + */ +const Type* TypeAry::remove_speculative() const { + return make(_elem->remove_speculative(), _size, _stable); +} + //----------------------interface_vs_oop--------------------------------------- #ifdef ASSERT bool TypeAry::interface_vs_oop(const Type *t) const { @@ -2566,14 +2583,14 @@ const Type *TypeOopPtr::xmeet(const Type *t) const { return res; } - if (res->isa_oopptr() != NULL) { + const TypeOopPtr* res_oopptr = res->is_oopptr(); + if (res_oopptr->speculative() != NULL) { // type->speculative() == NULL means that speculation is no better // than type, i.e. type->speculative() == type. So there are 2 // ways to represent the fact that we have no useful speculative // data and we should use a single one to be able to test for // equality between types. Check whether type->speculative() == // type and set speculative to NULL if it is the case. - const TypeOopPtr* res_oopptr = res->is_oopptr(); if (res_oopptr->remove_speculative() == res_oopptr->speculative()) { return res_oopptr->remove_speculative(); } @@ -2639,7 +2656,7 @@ const Type *TypeOopPtr::xmeet_helper(const Type *t) const { case OopPtr: { // Meeting to other OopPtrs const TypeOopPtr *tp = t->is_oopptr(); int instance_id = meet_instance_id(tp->instance_id()); - const TypeOopPtr* speculative = meet_speculative(tp); + const TypeOopPtr* speculative = xmeet_speculative(tp); return make(meet_ptr(tp->ptr()), meet_offset(tp->offset()), instance_id, speculative); } @@ -2793,9 +2810,9 @@ intptr_t TypeOopPtr::get_con() const { //-----------------------------filter------------------------------------------ // Do not allow interface-vs.-noninterface joins to collapse to top. -const Type *TypeOopPtr::filter(const Type *kills) const { +const Type *TypeOopPtr::filter_helper(const Type *kills, bool include_speculative) const { - const Type* ft = join(kills); + const Type* ft = join_helper(kills, include_speculative); const TypeInstPtr* ftip = ft->isa_instptr(); const TypeInstPtr* ktip = kills->isa_instptr(); @@ -2907,7 +2924,10 @@ const TypePtr *TypeOopPtr::add_offset(intptr_t offset) const { /** * Return same type without a speculative part */ -const TypeOopPtr* TypeOopPtr::remove_speculative() const { +const Type* TypeOopPtr::remove_speculative() const { + if (_speculative == NULL) { + return this; + } return make(_ptr, _offset, _instance_id, NULL); } @@ -2933,7 +2953,7 @@ int TypeOopPtr::dual_instance_id( ) const { * * @param other type to meet with */ -const TypeOopPtr* TypeOopPtr::meet_speculative(const TypeOopPtr* other) const { +const TypeOopPtr* TypeOopPtr::xmeet_speculative(const TypeOopPtr* other) const { bool this_has_spec = (_speculative != NULL); bool other_has_spec = (other->speculative() != NULL); @@ -2958,7 +2978,7 @@ const TypeOopPtr* TypeOopPtr::meet_speculative(const TypeOopPtr* other) const { other_spec = other; } - return this_spec->meet(other_spec)->is_oopptr(); + return this_spec->meet_speculative(other_spec)->is_oopptr(); } /** @@ -3117,7 +3137,7 @@ const TypeInstPtr *TypeInstPtr::xmeet_unloaded(const TypeInstPtr *tinst) const { int off = meet_offset(tinst->offset()); PTR ptr = meet_ptr(tinst->ptr()); int instance_id = meet_instance_id(tinst->instance_id()); - const TypeOopPtr* speculative = meet_speculative(tinst); + const TypeOopPtr* speculative = xmeet_speculative(tinst); const TypeInstPtr *loaded = is_loaded() ? this : tinst; const TypeInstPtr *unloaded = is_loaded() ? tinst : this; @@ -3194,7 +3214,7 @@ const Type *TypeInstPtr::xmeet_helper(const Type *t) const { int offset = meet_offset(tp->offset()); PTR ptr = meet_ptr(tp->ptr()); int instance_id = meet_instance_id(tp->instance_id()); - const TypeOopPtr* speculative = meet_speculative(tp); + const TypeOopPtr* speculative = xmeet_speculative(tp); switch (ptr) { case TopPTR: case AnyNull: // Fall 'down' to dual of object klass @@ -3244,14 +3264,14 @@ const Type *TypeInstPtr::xmeet_helper(const Type *t) const { case TopPTR: case AnyNull: { int instance_id = meet_instance_id(InstanceTop); - const TypeOopPtr* speculative = meet_speculative(tp); + const TypeOopPtr* speculative = xmeet_speculative(tp); return make(ptr, klass(), klass_is_exact(), (ptr == Constant ? const_oop() : NULL), offset, instance_id, speculative); } case NotNull: case BotPTR: { int instance_id = meet_instance_id(tp->instance_id()); - const TypeOopPtr* speculative = meet_speculative(tp); + const TypeOopPtr* speculative = xmeet_speculative(tp); return TypeOopPtr::make(ptr, offset, instance_id, speculative); } default: typerr(t); @@ -3303,7 +3323,7 @@ const Type *TypeInstPtr::xmeet_helper(const Type *t) const { int off = meet_offset( tinst->offset() ); PTR ptr = meet_ptr( tinst->ptr() ); int instance_id = meet_instance_id(tinst->instance_id()); - const TypeOopPtr* speculative = meet_speculative(tinst); + const TypeOopPtr* speculative = xmeet_speculative(tinst); // Check for easy case; klasses are equal (and perhaps not loaded!) // If we have constants, then we created oops so classes are loaded @@ -3552,7 +3572,10 @@ const TypePtr *TypeInstPtr::add_offset(intptr_t offset) const { return make(_ptr, klass(), klass_is_exact(), const_oop(), xadd_offset(offset), _instance_id, add_offset_speculative(offset)); } -const TypeOopPtr *TypeInstPtr::remove_speculative() const { +const Type *TypeInstPtr::remove_speculative() const { + if (_speculative == NULL) { + return this; + } return make(_ptr, klass(), klass_is_exact(), const_oop(), _offset, _instance_id, NULL); } @@ -3754,14 +3777,14 @@ const Type *TypeAryPtr::xmeet_helper(const Type *t) const { case TopPTR: case AnyNull: { int instance_id = meet_instance_id(InstanceTop); - const TypeOopPtr* speculative = meet_speculative(tp); + const TypeOopPtr* speculative = xmeet_speculative(tp); return make(ptr, (ptr == Constant ? const_oop() : NULL), _ary, _klass, _klass_is_exact, offset, instance_id, speculative); } case BotPTR: case NotNull: { int instance_id = meet_instance_id(tp->instance_id()); - const TypeOopPtr* speculative = meet_speculative(tp); + const TypeOopPtr* speculative = xmeet_speculative(tp); return TypeOopPtr::make(ptr, offset, instance_id, speculative); } default: ShouldNotReachHere(); @@ -3799,10 +3822,10 @@ const Type *TypeAryPtr::xmeet_helper(const Type *t) const { case AryPtr: { // Meeting 2 references? const TypeAryPtr *tap = t->is_aryptr(); int off = meet_offset(tap->offset()); - const TypeAry *tary = _ary->meet(tap->_ary)->is_ary(); + const TypeAry *tary = _ary->meet_speculative(tap->_ary)->is_ary(); PTR ptr = meet_ptr(tap->ptr()); int instance_id = meet_instance_id(tap->instance_id()); - const TypeOopPtr* speculative = meet_speculative(tap); + const TypeOopPtr* speculative = xmeet_speculative(tap); ciKlass* lazy_klass = NULL; if (tary->_elem->isa_int()) { // Integral array element types have irrelevant lattice relations. @@ -3882,7 +3905,7 @@ const Type *TypeAryPtr::xmeet_helper(const Type *t) const { int offset = meet_offset(tp->offset()); PTR ptr = meet_ptr(tp->ptr()); int instance_id = meet_instance_id(tp->instance_id()); - const TypeOopPtr* speculative = meet_speculative(tp); + const TypeOopPtr* speculative = xmeet_speculative(tp); switch (ptr) { case TopPTR: case AnyNull: // Fall 'down' to dual of object klass @@ -3996,8 +4019,8 @@ const TypePtr *TypeAryPtr::add_offset(intptr_t offset) const { return make(_ptr, _const_oop, _ary, _klass, _klass_is_exact, xadd_offset(offset), _instance_id, add_offset_speculative(offset)); } -const TypeOopPtr *TypeAryPtr::remove_speculative() const { - return make(_ptr, _const_oop, _ary, _klass, _klass_is_exact, _offset, _instance_id, NULL); +const Type *TypeAryPtr::remove_speculative() const { + return make(_ptr, _const_oop, _ary->remove_speculative()->is_ary(), _klass, _klass_is_exact, _offset, _instance_id, NULL); } //============================================================================= @@ -4037,9 +4060,9 @@ const Type *TypeNarrowPtr::xdual() const { // Compute dual right now. } -const Type *TypeNarrowPtr::filter( const Type *kills ) const { +const Type *TypeNarrowPtr::filter_helper(const Type *kills, bool include_speculative) const { if (isa_same_narrowptr(kills)) { - const Type* ft =_ptrtype->filter(is_same_narrowptr(kills)->_ptrtype); + const Type* ft =_ptrtype->filter_helper(is_same_narrowptr(kills)->_ptrtype, include_speculative); if (ft->empty()) return Type::TOP; // Canonical empty value if (ft->isa_ptr()) { @@ -4047,7 +4070,7 @@ const Type *TypeNarrowPtr::filter( const Type *kills ) const { } return ft; } else if (kills->isa_ptr()) { - const Type* ft = _ptrtype->join(kills); + const Type* ft = _ptrtype->join_helper(kills, include_speculative); if (ft->empty()) return Type::TOP; // Canonical empty value return ft; @@ -4177,8 +4200,8 @@ const TypePtr *TypeMetadataPtr::add_offset( intptr_t offset ) const { //-----------------------------filter------------------------------------------ // Do not allow interface-vs.-noninterface joins to collapse to top. -const Type *TypeMetadataPtr::filter( const Type *kills ) const { - const TypeMetadataPtr* ft = join(kills)->isa_metadataptr(); +const Type *TypeMetadataPtr::filter_helper(const Type *kills, bool include_speculative) const { + const TypeMetadataPtr* ft = join_helper(kills, include_speculative)->isa_metadataptr(); if (ft == NULL || ft->empty()) return Type::TOP; // Canonical empty value return ft; @@ -4380,10 +4403,10 @@ bool TypeKlassPtr::singleton(void) const { } // Do not allow interface-vs.-noninterface joins to collapse to top. -const Type *TypeKlassPtr::filter(const Type *kills) const { +const Type *TypeKlassPtr::filter_helper(const Type *kills, bool include_speculative) const { // logic here mirrors the one from TypeOopPtr::filter. See comments // there. - const Type* ft = join(kills); + const Type* ft = join_helper(kills, include_speculative); const TypeKlassPtr* ftkp = ft->isa_klassptr(); const TypeKlassPtr* ktkp = kills->isa_klassptr(); diff --git a/src/share/vm/opto/type.hpp b/src/share/vm/opto/type.hpp index 9aca89886..d02792f7b 100644 --- a/src/share/vm/opto/type.hpp +++ b/src/share/vm/opto/type.hpp @@ -164,6 +164,8 @@ private: virtual bool interface_vs_oop_helper(const Type *t) const; #endif + const Type *meet_helper(const Type *t, bool include_speculative) const; + protected: // Each class of type is also identified by its base. const TYPES _base; // Enum of Types type @@ -171,6 +173,10 @@ protected: Type( TYPES t ) : _dual(NULL), _base(t) {} // Simple types // ~Type(); // Use fast deallocation const Type *hashcons(); // Hash-cons the type + virtual const Type *filter_helper(const Type *kills, bool include_speculative) const; + const Type *join_helper(const Type *t, bool include_speculative) const { + return dual()->meet_helper(t->dual(), include_speculative)->dual(); + } public: @@ -202,10 +208,24 @@ public: // Test for equivalence of types static int cmp( const Type *const t1, const Type *const t2 ); // Test for higher or equal in lattice - int higher_equal( const Type *t ) const { return !cmp(meet(t),t); } + // Variant that drops the speculative part of the types + int higher_equal(const Type *t) const { + return !cmp(meet(t),t->remove_speculative()); + } + // Variant that keeps the speculative part of the types + int higher_equal_speculative(const Type *t) const { + return !cmp(meet_speculative(t),t); + } // MEET operation; lower in lattice. - const Type *meet( const Type *t ) const; + // Variant that drops the speculative part of the types + const Type *meet(const Type *t) const { + return meet_helper(t, false); + } + // Variant that keeps the speculative part of the types + const Type *meet_speculative(const Type *t) const { + return meet_helper(t, true); + } // WIDEN: 'widens' for Ints and other range types virtual const Type *widen( const Type *old, const Type* limit ) const { return this; } // NARROW: complement for widen, used by pessimistic phases @@ -221,13 +241,26 @@ public: // JOIN operation; higher in lattice. Done by finding the dual of the // meet of the dual of the 2 inputs. - const Type *join( const Type *t ) const { - return dual()->meet(t->dual())->dual(); } + // Variant that drops the speculative part of the types + const Type *join(const Type *t) const { + return join_helper(t, false); + } + // Variant that keeps the speculative part of the types + const Type *join_speculative(const Type *t) const { + return join_helper(t, true); + } // Modified version of JOIN adapted to the needs Node::Value. // Normalizes all empty values to TOP. Does not kill _widen bits. // Currently, it also works around limitations involving interface types. - virtual const Type *filter( const Type *kills ) const; + // Variant that drops the speculative part of the types + const Type *filter(const Type *kills) const { + return filter_helper(kills, false); + } + // Variant that keeps the speculative part of the types + const Type *filter_speculative(const Type *kills) const { + return filter_helper(kills, true); + } #ifdef ASSERT // One type is interface, the other is oop @@ -383,6 +416,8 @@ public: // Speculative type. See TypeInstPtr virtual ciKlass* speculative_type() const { return NULL; } + const Type* maybe_remove_speculative(bool include_speculative) const; + virtual const Type* remove_speculative() const { return this; } private: // support arrays @@ -450,12 +485,14 @@ public: // upper bound, inclusive. class TypeInt : public Type { TypeInt( jint lo, jint hi, int w ); +protected: + virtual const Type *filter_helper(const Type *kills, bool include_speculative) const; + public: virtual bool eq( const Type *t ) const; virtual int hash() const; // Type specific hashing virtual bool singleton(void) const; // TRUE if type is a singleton virtual bool empty(void) const; // TRUE if type is vacuous -public: const jint _lo, _hi; // Lower bound, upper bound const short _widen; // Limit on times we widen this sucker @@ -475,7 +512,6 @@ public: virtual const Type *widen( const Type *t, const Type* limit_type ) const; virtual const Type *narrow( const Type *t ) const; // Do not kill _widen bits. - virtual const Type *filter( const Type *kills ) const; // Convenience common pre-built types. static const TypeInt *MINUS_1; static const TypeInt *ZERO; @@ -506,6 +542,9 @@ public: // an upper bound, inclusive. class TypeLong : public Type { TypeLong( jlong lo, jlong hi, int w ); +protected: + // Do not kill _widen bits. + virtual const Type *filter_helper(const Type *kills, bool include_speculative) const; public: virtual bool eq( const Type *t ) const; virtual int hash() const; // Type specific hashing @@ -533,8 +572,6 @@ public: virtual const Type *xdual() const; // Compute dual right now. virtual const Type *widen( const Type *t, const Type* limit_type ) const; virtual const Type *narrow( const Type *t ) const; - // Do not kill _widen bits. - virtual const Type *filter( const Type *kills ) const; // Convenience common pre-built types. static const TypeLong *MINUS_1; static const TypeLong *ZERO; @@ -625,6 +662,7 @@ public: virtual const Type *xmeet( const Type *t ) const; virtual const Type *xdual() const; // Compute dual right now. bool ary_must_be_exact() const; // true if arrays of such are never generic + virtual const Type* remove_speculative() const; #ifdef ASSERT // One type is interface, the other is oop virtual bool interface_vs_oop(const Type *t) const; @@ -835,7 +873,7 @@ protected: // utility methods to work on the speculative part of the type const TypeOopPtr* dual_speculative() const; - const TypeOopPtr* meet_speculative(const TypeOopPtr* other) const; + const TypeOopPtr* xmeet_speculative(const TypeOopPtr* other) const; bool eq_speculative(const TypeOopPtr* other) const; int hash_speculative() const; const TypeOopPtr* add_offset_speculative(intptr_t offset) const; @@ -843,6 +881,9 @@ protected: void dump_speculative(outputStream *st) const; #endif + // Do not allow interface-vs.-noninterface joins to collapse to top. + virtual const Type *filter_helper(const Type *kills, bool include_speculative) const; + public: // Creates a type given a klass. Correctly handles multi-dimensional arrays // Respects UseUniqueSubclasses. @@ -898,16 +939,13 @@ public: virtual const TypePtr *add_offset( intptr_t offset ) const; // Return same type without a speculative part - virtual const TypeOopPtr* remove_speculative() const; + virtual const Type* remove_speculative() const; virtual const Type *xmeet(const Type *t) const; virtual const Type *xdual() const; // Compute dual right now. // the core of the computation of the meet for TypeOopPtr and for its subclasses virtual const Type *xmeet_helper(const Type *t) const; - // Do not allow interface-vs.-noninterface joins to collapse to top. - virtual const Type *filter( const Type *kills ) const; - // Convenience common pre-built type. static const TypeOopPtr *BOTTOM; #ifndef PRODUCT @@ -984,7 +1022,7 @@ class TypeInstPtr : public TypeOopPtr { virtual const TypePtr *add_offset( intptr_t offset ) const; // Return same type without a speculative part - virtual const TypeOopPtr* remove_speculative() const; + virtual const Type* remove_speculative() const; // the core of the computation of the meet of 2 types virtual const Type *xmeet_helper(const Type *t) const; @@ -1062,7 +1100,7 @@ public: virtual bool empty(void) const; // TRUE if type is vacuous virtual const TypePtr *add_offset( intptr_t offset ) const; // Return same type without a speculative part - virtual const TypeOopPtr* remove_speculative() const; + virtual const Type* remove_speculative() const; // the core of the computation of the meet of 2 types virtual const Type *xmeet_helper(const Type *t) const; @@ -1103,6 +1141,8 @@ public: class TypeMetadataPtr : public TypePtr { protected: TypeMetadataPtr(PTR ptr, ciMetadata* metadata, int offset); + // Do not allow interface-vs.-noninterface joins to collapse to top. + virtual const Type *filter_helper(const Type *kills, bool include_speculative) const; public: virtual bool eq( const Type *t ) const; virtual int hash() const; // Type specific hashing @@ -1128,9 +1168,6 @@ public: virtual intptr_t get_con() const; - // Do not allow interface-vs.-noninterface joins to collapse to top. - virtual const Type *filter( const Type *kills ) const; - // Convenience common pre-built types. static const TypeMetadataPtr *BOTTOM; @@ -1144,6 +1181,8 @@ public: class TypeKlassPtr : public TypePtr { TypeKlassPtr( PTR ptr, ciKlass* klass, int offset ); +protected: + virtual const Type *filter_helper(const Type *kills, bool include_speculative) const; public: virtual bool eq( const Type *t ) const; virtual int hash() const; // Type specific hashing @@ -1205,9 +1244,6 @@ public: virtual intptr_t get_con() const; - // Do not allow interface-vs.-noninterface joins to collapse to top. - virtual const Type *filter( const Type *kills ) const; - // Convenience common pre-built types. static const TypeKlassPtr* OBJECT; // Not-null object klass or below static const TypeKlassPtr* OBJECT_OR_NULL; // Maybe-null version of same @@ -1231,6 +1267,8 @@ protected: virtual const TypeNarrowPtr *is_same_narrowptr(const Type *t) const = 0; virtual const TypeNarrowPtr *make_same_narrowptr(const TypePtr *t) const = 0; virtual const TypeNarrowPtr *make_hash_same_narrowptr(const TypePtr *t) const = 0; + // Do not allow interface-vs.-noninterface joins to collapse to top. + virtual const Type *filter_helper(const Type *kills, bool include_speculative) const; public: virtual bool eq( const Type *t ) const; virtual int hash() const; // Type specific hashing @@ -1241,9 +1279,6 @@ public: virtual intptr_t get_con() const; - // Do not allow interface-vs.-noninterface joins to collapse to top. - virtual const Type *filter( const Type *kills ) const; - virtual bool empty(void) const; // TRUE if type is vacuous // returns the equivalent ptr type for this compressed pointer @@ -1294,6 +1329,10 @@ public: static const TypeNarrowOop *BOTTOM; static const TypeNarrowOop *NULL_PTR; + virtual const Type* remove_speculative() const { + return make(_ptrtype->remove_speculative()->is_ptr()); + } + #ifndef PRODUCT virtual void dump2( Dict &d, uint depth, outputStream *st ) const; #endif diff --git a/src/share/vm/prims/jvmtiCodeBlobEvents.cpp b/src/share/vm/prims/jvmtiCodeBlobEvents.cpp index c46afdc10..18d8786b8 100644 --- a/src/share/vm/prims/jvmtiCodeBlobEvents.cpp +++ b/src/share/vm/prims/jvmtiCodeBlobEvents.cpp @@ -26,6 +26,7 @@ #include "code/codeBlob.hpp" #include "code/codeCache.hpp" #include "code/scopeDesc.hpp" +#include "code/vtableStubs.hpp" #include "memory/resourceArea.hpp" #include "oops/oop.inline.hpp" #include "prims/jvmtiCodeBlobEvents.hpp" @@ -63,6 +64,7 @@ class CodeBlobCollector : StackObj { // used during a collection static GrowableArray<JvmtiCodeBlobDesc*>* _global_code_blobs; static void do_blob(CodeBlob* cb); + static void do_vtable_stub(VtableStub* vs); public: CodeBlobCollector() { _code_blobs = NULL; @@ -119,6 +121,10 @@ void CodeBlobCollector::do_blob(CodeBlob* cb) { if (cb->is_nmethod()) { return; } + // exclude VtableStubs, which are processed separately + if (cb->is_buffer_blob() && strcmp(cb->name(), "vtable chunks") == 0) { + return; + } // check if this starting address has been seen already - the // assumption is that stubs are inserted into the list before the @@ -136,6 +142,13 @@ void CodeBlobCollector::do_blob(CodeBlob* cb) { _global_code_blobs->append(scb); } +// called for each VtableStub in VtableStubs + +void CodeBlobCollector::do_vtable_stub(VtableStub* vs) { + JvmtiCodeBlobDesc* scb = new JvmtiCodeBlobDesc(vs->is_vtable_stub() ? "vtable stub" : "itable stub", + vs->code_begin(), vs->code_end()); + _global_code_blobs->append(scb); +} // collects a list of CodeBlobs in the CodeCache. // @@ -166,6 +179,10 @@ void CodeBlobCollector::collect() { _global_code_blobs->append(new JvmtiCodeBlobDesc(desc->name(), desc->begin(), desc->end())); } + // Vtable stubs are not described with StubCodeDesc, + // process them separately + VtableStubs::vtable_stub_do(do_vtable_stub); + // next iterate over all the non-nmethod code blobs and add them to // the list - as noted above this will filter out duplicates and // enclosing blobs. diff --git a/src/share/vm/runtime/arguments.cpp b/src/share/vm/runtime/arguments.cpp index 60a7f904d..0dcb9e756 100644 --- a/src/share/vm/runtime/arguments.cpp +++ b/src/share/vm/runtime/arguments.cpp @@ -293,6 +293,7 @@ static ObsoleteFlag obsolete_jvm_flags[] = { { "UsePermISM", JDK_Version::jdk(8), JDK_Version::jdk(9) }, { "UseMPSS", JDK_Version::jdk(8), JDK_Version::jdk(9) }, { "UseStringCache", JDK_Version::jdk(8), JDK_Version::jdk(9) }, + { "UseOldInlining", JDK_Version::jdk(9), JDK_Version::jdk(10) }, #ifdef PRODUCT { "DesiredMethodLimit", JDK_Version::jdk_update(7, 2), JDK_Version::jdk(8) }, @@ -881,7 +882,7 @@ bool Arguments::process_argument(const char* arg, arg_len = equal_sign - argname; } - Flag* found_flag = Flag::find_flag((const char*)argname, arg_len, true); + Flag* found_flag = Flag::find_flag((const char*)argname, arg_len, true, true); if (found_flag != NULL) { char locked_message_buf[BUFLEN]; found_flag->get_locked_message(locked_message_buf, BUFLEN); diff --git a/src/share/vm/runtime/globals.cpp b/src/share/vm/runtime/globals.cpp index de6fbf708..fbfc08ed0 100644 --- a/src/share/vm/runtime/globals.cpp +++ b/src/share/vm/runtime/globals.cpp @@ -62,6 +62,14 @@ ARCH_FLAGS(MATERIALIZE_DEVELOPER_FLAG, MATERIALIZE_PRODUCT_FLAG, \ MATERIALIZE_FLAGS_EXT +static bool is_product_build() { +#ifdef PRODUCT + return true; +#else + return false; +#endif +} + void Flag::check_writable() { if (is_constant_in_binary()) { fatal(err_msg("flag is constant: %s", _name)); @@ -235,6 +243,27 @@ bool Flag::is_unlocked() const { // Get custom message for this locked flag, or return NULL if // none is available. void Flag::get_locked_message(char* buf, int buflen) const { + buf[0] = '\0'; + if (is_diagnostic() && !is_unlocked()) { + jio_snprintf(buf, buflen, "Error: VM option '%s' is diagnostic and must be enabled via -XX:+UnlockDiagnosticVMOptions.\n", + _name); + return; + } + if (is_experimental() && !is_unlocked()) { + jio_snprintf(buf, buflen, "Error: VM option '%s' is experimental and must be enabled via -XX:+UnlockExperimentalVMOptions.\n", + _name); + return; + } + if (is_develop() && is_product_build()) { + jio_snprintf(buf, buflen, "Error: VM option '%s' is develop and is available only in debug version of VM.\n", + _name); + return; + } + if (is_notproduct() && is_product_build()) { + jio_snprintf(buf, buflen, "Error: VM option '%s' is notproduct and is available only in debug version of VM.\n", + _name); + return; + } get_locked_message_ext(buf, buflen); } @@ -464,13 +493,13 @@ inline bool str_equal(const char* s, const char* q, size_t len) { } // Search the flag table for a named flag -Flag* Flag::find_flag(const char* name, size_t length, bool allow_locked) { +Flag* Flag::find_flag(const char* name, size_t length, bool allow_locked, bool return_flag) { for (Flag* current = &flagTable[0]; current->_name != NULL; current++) { if (str_equal(current->_name, name, length)) { // Found a matching entry. // Don't report notproduct and develop flags in product builds. if (current->is_constant_in_binary()) { - return NULL; + return (return_flag == true ? current : NULL); } // Report locked flags only if allowed. if (!(current->is_unlocked() || current->is_unlocker())) { diff --git a/src/share/vm/runtime/globals.hpp b/src/share/vm/runtime/globals.hpp index b828de5ab..6b6fe0251 100644 --- a/src/share/vm/runtime/globals.hpp +++ b/src/share/vm/runtime/globals.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -255,7 +255,7 @@ struct Flag { // number of flags static size_t numFlags; - static Flag* find_flag(const char* name, size_t length, bool allow_locked = false); + static Flag* find_flag(const char* name, size_t length, bool allow_locked = false, bool return_flag = false); static Flag* fuzzy_match(const char* name, size_t length, bool allow_locked = false); void check_writable(); @@ -1272,6 +1272,9 @@ class CommandLineFlags { develop(bool, TraceJNICalls, false, \ "Trace JNI calls") \ \ + develop(bool, StressRewriter, false, \ + "Stress linktime bytecode rewriting") \ + \ notproduct(bool, TraceJVMCalls, false, \ "Trace JVM calls") \ \ diff --git a/src/share/vm/runtime/os.cpp b/src/share/vm/runtime/os.cpp index 5e7ec4c9c..04598e985 100644 --- a/src/share/vm/runtime/os.cpp +++ b/src/share/vm/runtime/os.cpp @@ -1081,7 +1081,6 @@ void os::print_location(outputStream* st, intptr_t x, bool verbose) { } -#ifndef PRODUCT // Check if in metaspace. if (ClassLoaderDataGraph::contains((address)addr)) { // Use addr->print() from the debugger instead (not here) @@ -1089,7 +1088,6 @@ void os::print_location(outputStream* st, intptr_t x, bool verbose) { " is pointing into metadata", addr); return; } -#endif // Try an OS specific find if (os::find(addr, st)) { diff --git a/src/share/vm/utilities/array.hpp b/src/share/vm/utilities/array.hpp index 9f8e45f32..0fbcd94d2 100644 --- a/src/share/vm/utilities/array.hpp +++ b/src/share/vm/utilities/array.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved. + * Copyright (c) 2000, 2014, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -58,7 +58,7 @@ class ResourceArray: public ResourceObj { void initialize(size_t esize, int length) { assert(length >= 0, "illegal length"); - assert(_data == NULL, "must be new object"); + assert(StressRewriter || _data == NULL, "must be new object"); _length = length; _data = resource_allocate_bytes(esize * length); DEBUG_ONLY(init_nesting();) diff --git a/src/share/vm/utilities/bitMap.cpp b/src/share/vm/utilities/bitMap.cpp index a0e5d6c89..b2a2ab7b2 100644 --- a/src/share/vm/utilities/bitMap.cpp +++ b/src/share/vm/utilities/bitMap.cpp @@ -110,7 +110,7 @@ void BitMap::par_put_range_within_word(idx_t beg, idx_t end, bool value) { while (true) { intptr_t res = Atomic::cmpxchg_ptr(nw, pw, w); if (res == w) break; - w = *pw; + w = res; nw = value ? (w | ~mr) : (w & mr); } } |