1.1 --- a/src/cpu/ppc/vm/vm_version_ppc.cpp Tue Dec 10 14:29:43 2013 +0100
1.2 +++ b/src/cpu/ppc/vm/vm_version_ppc.cpp Wed Dec 11 00:06:11 2013 +0100
1.3 @@ -89,16 +89,17 @@
1.4 }
1.5
1.6 // On Power6 test for section size.
1.7 - if (PowerArchitecturePPC64 == 6)
1.8 + if (PowerArchitecturePPC64 == 6) {
1.9 determine_section_size();
1.10 - // TODO: PPC port else
1.11 + // TODO: PPC port } else {
1.12 // TODO: PPC port PdScheduling::power6SectorSize = 0x20;
1.13 + }
1.14
1.15 MaxVectorSize = 8;
1.16 #endif
1.17
1.18 // Create and print feature-string.
1.19 - char buf[(num_features+1) * 16]; // max 16 chars per feature
1.20 + char buf[(num_features+1) * 16]; // Max 16 chars per feature.
1.21 jio_snprintf(buf, sizeof(buf),
1.22 "ppc64%s%s%s%s%s%s%s%s",
1.23 (has_fsqrt() ? " fsqrt" : ""),
1.24 @@ -127,21 +128,21 @@
1.25 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1;
1.26
1.27 if (AllocatePrefetchStyle == 4) {
1.28 - AllocatePrefetchStepSize = cache_line_size; // need exact value
1.29 - if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // use larger blocks by default
1.30 - if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // default is not defined ?
1.31 + AllocatePrefetchStepSize = cache_line_size; // Need exact value.
1.32 + if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default.
1.33 + if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined?
1.34 } else {
1.35 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size;
1.36 - if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value
1.37 - if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // default is not defined ?
1.38 + if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value.
1.39 + if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined?
1.40 }
1.41
1.42 assert(AllocatePrefetchLines > 0, "invalid value");
1.43 if (AllocatePrefetchLines < 1) // Set valid value in product VM.
1.44 - AllocatePrefetchLines = 1; // Conservative value
1.45 + AllocatePrefetchLines = 1; // Conservative value.
1.46
1.47 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size)
1.48 - AllocatePrefetchStyle = 1; // fall back if inappropriate
1.49 + AllocatePrefetchStyle = 1; // Fall back if inappropriate.
1.50
1.51 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
1.52 }
1.53 @@ -160,13 +161,13 @@
1.54
1.55 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord;
1.56
1.57 - // Allocate space for the code
1.58 + // Allocate space for the code.
1.59 ResourceMark rm;
1.60 CodeBuffer cb("detect_section_size", code_size, 0);
1.61 MacroAssembler* a = new MacroAssembler(&cb);
1.62
1.63 uint32_t *code = (uint32_t *)a->pc();
1.64 - // emit code.
1.65 + // Emit code.
1.66 void (*test1)() = (void(*)())(void *)a->emit_fd();
1.67
1.68 Label l1;
1.69 @@ -189,58 +190,58 @@
1.70
1.71 // ;; 1
1.72 a->nop(); // 5
1.73 - a->fmr(F6, F6); // 6
1.74 - a->fmr(F7, F7); // 7
1.75 + a->fmr(F6, F6); // 6
1.76 + a->fmr(F7, F7); // 7
1.77 a->endgroup(); // 8
1.78 // ------- sector 8 ------------
1.79
1.80 // ;; 2
1.81 a->nop(); // 9
1.82 a->nop(); // 10
1.83 - a->fmr(F8, F8); // 11
1.84 - a->fmr(F9, F9); // 12
1.85 + a->fmr(F8, F8); // 11
1.86 + a->fmr(F9, F9); // 12
1.87
1.88 // ;; 3
1.89 a->nop(); // 13
1.90 - a->fmr(F10, F10); // 14
1.91 - a->fmr(F11, F11); // 15
1.92 + a->fmr(F10, F10); // 14
1.93 + a->fmr(F11, F11); // 15
1.94 a->endgroup(); // 16
1.95 // -------- sector 16 -------------
1.96
1.97 // ;; 4
1.98 a->nop(); // 17
1.99 a->nop(); // 18
1.100 - a->fmr(F15, F15); // 19
1.101 - a->fmr(F16, F16); // 20
1.102 + a->fmr(F15, F15); // 19
1.103 + a->fmr(F16, F16); // 20
1.104
1.105 // ;; 5
1.106 a->nop(); // 21
1.107 - a->fmr(F17, F17); // 22
1.108 - a->fmr(F18, F18); // 23
1.109 + a->fmr(F17, F17); // 22
1.110 + a->fmr(F18, F18); // 23
1.111 a->endgroup(); // 24
1.112 // ------- sector 24 ------------
1.113
1.114 // ;; 6
1.115 a->nop(); // 25
1.116 a->nop(); // 26
1.117 - a->fmr(F19, F19); // 27
1.118 - a->fmr(F20, F20); // 28
1.119 + a->fmr(F19, F19); // 27
1.120 + a->fmr(F20, F20); // 28
1.121
1.122 // ;; 7
1.123 a->nop(); // 29
1.124 - a->fmr(F21, F21); // 30
1.125 - a->fmr(F22, F22); // 31
1.126 + a->fmr(F21, F21); // 30
1.127 + a->fmr(F22, F22); // 31
1.128 a->brnop0(); // 32
1.129
1.130 // ------- sector 32 ------------
1.131 }
1.132
1.133 // ;; 8
1.134 - a->cmpdi(CCR0, R4, unroll);// 33
1.135 - a->bge(CCR0, l1); // 34
1.136 + a->cmpdi(CCR0, R4, unroll); // 33
1.137 + a->bge(CCR0, l1); // 34
1.138 a->blr();
1.139
1.140 - // emit code.
1.141 + // Emit code.
1.142 void (*test2)() = (void(*)())(void *)a->emit_fd();
1.143 // uint32_t *code = (uint32_t *)a->pc();
1.144
1.145 @@ -382,39 +383,40 @@
1.146 #endif // COMPILER2
1.147
1.148 void VM_Version::determine_features() {
1.149 - const int code_size = (num_features+1+2*7)*BytesPerInstWord; // 7 InstWords for each call (function descriptor + blr instruction)
1.150 + // 7 InstWords for each call (function descriptor + blr instruction).
1.151 + const int code_size = (num_features+1+2*7)*BytesPerInstWord;
1.152 int features = 0;
1.153
1.154 // create test area
1.155 - enum { BUFFER_SIZE = 2*4*K }; // needs to be >=2* max cache line size (cache line size can't exceed min page size)
1.156 + enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size).
1.157 char test_area[BUFFER_SIZE];
1.158 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1];
1.159
1.160 - // Allocate space for the code
1.161 + // Allocate space for the code.
1.162 ResourceMark rm;
1.163 CodeBuffer cb("detect_cpu_features", code_size, 0);
1.164 MacroAssembler* a = new MacroAssembler(&cb);
1.165
1.166 - // emit code.
1.167 + // Emit code.
1.168 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->emit_fd();
1.169 uint32_t *code = (uint32_t *)a->pc();
1.170 // Don't use R0 in ldarx.
1.171 - // keep R3_ARG1 = R3 unmodified, it contains &field (see below)
1.172 - // keep R4_ARG2 = R4 unmodified, it contains offset = 0 (see below)
1.173 - a->fsqrt(F3, F4); // code[0] -> fsqrt_m
1.174 - a->isel(R7, R5, R6, 0); // code[1] -> isel_m
1.175 - a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1);// code[2] -> lxarx_m
1.176 - a->cmpb(R7, R5, R6); // code[3] -> bcmp
1.177 - //a->mftgpr(R7, F3); // code[4] -> mftgpr
1.178 - a->popcntb(R7, R5); // code[5] -> popcntb
1.179 - a->popcntw(R7, R5); // code[6] -> popcntw
1.180 - a->fcfids(F3, F4); // code[7] -> fcfids
1.181 - a->vand(VR0, VR0, VR0); // code[8] -> vand
1.182 + // Keep R3_ARG1 unmodified, it contains &field (see below).
1.183 + // Keep R4_ARG2 unmodified, it contains offset = 0 (see below).
1.184 + a->fsqrt(F3, F4); // code[0] -> fsqrt_m
1.185 + a->isel(R7, R5, R6, 0); // code[1] -> isel_m
1.186 + a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[2] -> lxarx_m
1.187 + a->cmpb(R7, R5, R6); // code[3] -> bcmp
1.188 + //a->mftgpr(R7, F3); // code[4] -> mftgpr
1.189 + a->popcntb(R7, R5); // code[5] -> popcntb
1.190 + a->popcntw(R7, R5); // code[6] -> popcntw
1.191 + a->fcfids(F3, F4); // code[7] -> fcfids
1.192 + a->vand(VR0, VR0, VR0); // code[8] -> vand
1.193 a->blr();
1.194
1.195 - // Emit function to set one cache line to zero
1.196 - void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->emit_fd(); // emit function descriptor and get pointer to it
1.197 - a->dcbz(R3_ARG1); // R3_ARG1 = R3 = addr
1.198 + // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it.
1.199 + void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->emit_fd();
1.200 + a->dcbz(R3_ARG1); // R3_ARG1 = addr
1.201 a->blr();
1.202
1.203 uint32_t *code_end = (uint32_t *)a->pc();
1.204 @@ -428,8 +430,8 @@
1.205 }
1.206
1.207 // Measure cache line size.
1.208 - memset(test_area, 0xFF, BUFFER_SIZE); // fill test area with 0xFF
1.209 - (*zero_cacheline_func_ptr)(mid_of_test_area); // call function which executes dcbz to the middle
1.210 + memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF.
1.211 + (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle.
1.212 int count = 0; // count zeroed bytes
1.213 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++;
1.214 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2");
