| 1 |
|
| #define PERL_NO_GET_CONTEXT |
| 2 |
|
| #include "EXTERN.h" |
| 3 |
|
| #include "perl.h" |
| 4 |
|
| #include "XSUB.h" |
| 5 |
|
| #include "ppport.h" |
| 6 |
|
| #include "ndarray.h" |
| 7 |
|
| |
| 8 |
|
| #ifdef HAVE_PDL |
| 9 |
|
| #include <pdlcore.h> |
| 10 |
|
| static Core *PDL; /* PDL core C functions; bootstrapped lazily on first as_pdl_alias */ |
| 11 |
|
| |
| 12 |
|
| /* Keep the mapping alive until PDL releases the external data. */ |
| 13 |
|
| static void nda_pdl_nofree(pdl *p, Size_t param) { |
| 14 |
|
| dTHX; |
| 15 |
|
| p->data = 0; |
| 16 |
|
| SvREFCNT_dec(INT2PTR(SV *, param)); |
| 17 |
|
| } |
| 18 |
|
| |
| 19 |
|
| /* Load PDL::Core and grab its Core struct on first use, so plain use of the module |
| 20 |
|
| without as_pdl_alias does not pull in PDL. */ |
| 21 |
|
| static void nda_boot_pdl(pTHX) { |
| 22 |
|
| SV *coresv; |
| 23 |
|
| if (PDL) return; |
| 24 |
|
| perl_require_pv("PDL/Core.pm"); |
| 25 |
|
| if (SvTRUE(ERRSV)) croak("%s", SvPV_nolen(ERRSV)); |
| 26 |
|
| coresv = get_sv("PDL::SHARE", 0); |
| 27 |
|
| if (!coresv) croak("Data::NDArray::Shared: PDL::Core not loadable (needed for as_pdl_alias)"); |
| 28 |
|
| Core *core = INT2PTR(Core *, SvIV(coresv)); |
| 29 |
|
| if (!core) croak("Data::NDArray::Shared: got a NULL PDL Core pointer"); |
| 30 |
|
| if (core->Version != PDL_CORE_VERSION) |
| 31 |
|
| croak("Data::NDArray::Shared: please rebuild against the installed PDL (Core version mismatch)"); |
| 32 |
|
| PDL = core; |
| 33 |
|
| } |
| 34 |
|
| #endif |
| 35 |
|
| |
| 36 |
|
| #define EXTRACT(sv) \ |
| 37 |
|
| if (!sv_isobject(sv) || !sv_derived_from(sv, "Data::NDArray::Shared")) \ |
| 38 |
|
| croak("Expected a Data::NDArray::Shared object"); \ |
| 39 |
|
| NdaHandle *h = INT2PTR(NdaHandle*, SvIV(SvRV(sv))); \ |
| 40 |
|
| if (!h) croak("Attempted to use a destroyed Data::NDArray::Shared object"); \ |
| 41 |
|
| NdaHandle *h0 = h; PERL_UNUSED_VAR(h0); \ |
| 42 |
|
| sv_2mortal(SvREFCNT_inc(SvRV(sv))) |
| 43 |
|
| |
| 44 |
|
| /* Re-read the handle after a call that can run Perl code (tied/overloaded |
| 45 |
|
| * argument magic, tied-array fetches). That code may call $obj->DESTROY |
| 46 |
|
| * explicitly, which frees the handle and zeroes the IV; EXTRACT's mortal |
| 47 |
|
| * pins the referent only against refcount-driven destruction, not an |
| 48 |
|
| * explicit DESTROY, so the local `h` would dangle. Used only where magic |
| 49 |
|
| * can actually intervene between EXTRACT and the first use of h. |
| 50 |
|
| * The same Perl can also REPLACE the invocant ($obj = 42 mutates ST(0), |
| 51 |
|
| * because Perl passes aliases), hence the SvROK re-check before SvRV. */ |
| 52 |
|
| #define REEXTRACT(sv) \ |
| 53 |
|
| if (!SvROK(sv)) \ |
| 54 |
|
| croak("Data::NDArray::Shared object was replaced during the call"); \ |
| 55 |
|
| h = INT2PTR(NdaHandle*, SvIV(SvRV(sv))); \ |
| 56 |
|
| if (h != h0) croak("Data::NDArray::Shared object replaced or destroyed during the call") |
| 57 |
|
| |
| 58 |
|
| #define MAKE_OBJ(class, handle) \ |
| 59 |
|
| SV *obj = newSViv(PTR2IV(handle)); \ |
| 60 |
|
| SV *ref = newRV_noinc(obj); \ |
| 61 |
|
| sv_bless(ref, gv_stashpv(class, GV_ADD)); \ |
| 62 |
|
| RETVAL = ref |
| 63 |
|
| |
| 64 |
|
| /* ---------------------------------------------------------------- |
| 65 |
|
| * Typed element get: read element e of dtype dt, return the dtype-correct SV |
| 66 |
|
| * (float -> newSVnv, signed -> newSViv, unsigned -> newSVuv). |
| 67 |
|
| * ---------------------------------------------------------------- */ |
| 68 |
1231 |
| static uint32_t nda_dtype_bytes(int dt) { |
| 69 |
1231 |
| switch (dt) { |
| 70 |
1166 |
| case NDA_F64: case NDA_I64: case NDA_U64: return 8; |
| 71 |
43 |
| case NDA_F32: case NDA_I32: case NDA_U32: return 4; |
| 72 |
8 |
| case NDA_I16: case NDA_U16: return 2; |
| 73 |
14 |
| default: return 1; /* I8 / U8 */ |
| 74 |
|
| } |
| 75 |
|
| } |
| 76 |
|
| |
| 77 |
|
| /* Copy element e's raw bytes into buf (caller holds a lock). Building the SV is |
| 78 |
|
| deferred to nda_sv_from_raw so a croak-capable alloc never runs under the lock |
| 79 |
|
| (an OOM longjmp there would strand this process's read lock -> writers hang). */ |
| 80 |
1172 |
| static void nda_read_raw(NdaHandle *h, uint64_t e, char buf[8]) { |
| 81 |
1172 |
| uint32_t nb = nda_dtype_bytes(h->dtype); /* cached: peer hdr->dtype must not resize the element */ |
| 82 |
1172 |
| memcpy(buf, nda_data(h) + e * nb, nb); |
| 83 |
1172 |
| } |
| 84 |
|
| |
| 85 |
|
| /* Build an SV from raw element bytes (NO lock needed). */ |
| 86 |
4041 |
| static SV *nda_sv_from_raw(pTHX_ int dtype, const char *buf) { |
| 87 |
4041 |
| switch (dtype) { |
| 88 |
1610 |
| case NDA_F64: { double v; memcpy(&v, buf, 8); return newSVnv(v); } |
| 89 |
277 |
| case NDA_F32: { float v; memcpy(&v, buf, 4); return newSVnv((double)v); } |
| 90 |
278 |
| case NDA_I64: { int64_t v; memcpy(&v, buf, 8); return newSViv((IV)v); } |
| 91 |
305 |
| case NDA_I32: { int32_t v; memcpy(&v, buf, 4); return newSViv((IV)v); } |
| 92 |
259 |
| case NDA_I16: { int16_t v; memcpy(&v, buf, 2); return newSViv((IV)v); } |
| 93 |
259 |
| case NDA_I8: { int8_t v = (int8_t)buf[0]; return newSViv((IV)v); } |
| 94 |
262 |
| case NDA_U64: { uint64_t v; memcpy(&v, buf, 8); return newSVuv((UV)v); } |
| 95 |
263 |
| case NDA_U32: { uint32_t v; memcpy(&v, buf, 4); return newSVuv((UV)v); } |
| 96 |
261 |
| case NDA_U16: { uint16_t v; memcpy(&v, buf, 2); return newSVuv((UV)v); } |
| 97 |
267 |
| case NDA_U8: { uint8_t v = (uint8_t)buf[0]; return newSVuv((UV)v); } |
| 98 |
|
| } |
| 99 |
0 |
| return newSViv(0); |
| 100 |
|
| } |
| 101 |
|
| |
| 102 |
|
| /* Read + build in one step (only for callers that already build after unlock). */ |
| 103 |
0 |
| static SV *nda_get_sv(pTHX_ NdaHandle *h, uint64_t e) { |
| 104 |
0 |
| char buf[8]; nda_read_raw(h, e, buf); return nda_sv_from_raw(aTHX_ h->dtype, buf); |
| 105 |
|
| } |
| 106 |
|
| |
| 107 |
|
| /* ---------------------------------------------------------------- |
| 108 |
|
| * Typed element set: write element e from an SV, casting/truncating to the |
| 109 |
|
| * element width (float -> SvNV, signed -> (intN)SvIV, unsigned -> (uintN)SvUV). |
| 110 |
|
| * Out-of-range integer values wrap to the element width per C cast rules. |
| 111 |
|
| * ---------------------------------------------------------------- */ |
| 112 |
|
| /* Encode val to its dtype's raw bytes BEFORE locking: SvNV/SvIV/SvUV can croak |
| 113 |
|
| * (overload/FATAL-numeric/tied FETCH), which under the wrlock would leak it. */ |
| 114 |
480 |
| static uint32_t nda_encode(pTHX_ uint8_t dtype, SV *val, char *out) { |
| 115 |
480 |
| switch (dtype) { |
| 116 |
372 |
| case NDA_F64: { double v = (double)SvNV(val); memcpy(out, &v, 8); return 8; } |
| 117 |
11 |
| case NDA_F32: { float v = (float)SvNV(val); memcpy(out, &v, 4); return 4; } |
| 118 |
25 |
| case NDA_I64: { int64_t v = (int64_t)SvIV(val); memcpy(out, &v, 8); return 8; } |
| 119 |
34 |
| case NDA_I32: { int32_t v = (int32_t)SvIV(val); memcpy(out, &v, 4); return 4; } |
| 120 |
4 |
| case NDA_I16: { int16_t v = (int16_t)SvIV(val); memcpy(out, &v, 2); return 2; } |
| 121 |
3 |
| case NDA_I8: { int8_t v = (int8_t)SvIV(val); out[0] = (char)v; return 1; } |
| 122 |
5 |
| case NDA_U64: { uint64_t v = (uint64_t)SvUV(val); memcpy(out, &v, 8); return 8; } |
| 123 |
3 |
| case NDA_U32: { uint32_t v = (uint32_t)SvUV(val); memcpy(out, &v, 4); return 4; } |
| 124 |
7 |
| case NDA_U16: { uint16_t v = (uint16_t)SvUV(val); memcpy(out, &v, 2); return 2; } |
| 125 |
16 |
| case NDA_U8: { uint8_t v = (uint8_t)SvUV(val); out[0] = (char)v; return 1; } |
| 126 |
|
| } |
| 127 |
0 |
| return 0; |
| 128 |
|
| } |
| 129 |
|
| /* Store pre-encoded bytes at element e (call under the write lock). */ |
| 130 |
434 |
| static void nda_store_bytes(NdaHandle *h, uint64_t e, const char *enc, uint32_t nb) { |
| 131 |
434 |
| memcpy(nda_data(h) + e * nb, enc, nb); |
| 132 |
434 |
| } |
| 133 |
|
| |
| 134 |
|
| /* ---------------------------------------------------------------- |
| 135 |
|
| * Fill every element with the typed value of val. |
| 136 |
|
| * ---------------------------------------------------------------- */ |
| 137 |
|
| /* Splat pre-encoded bytes across every element (call under the write lock). |
| 138 |
|
| * The croak-prone numeric conversion is done by nda_encode BEFORE locking. */ |
| 139 |
43 |
| static void nda_fill_bytes(NdaHandle *h, const char *enc, uint32_t nb) { |
| 140 |
43 |
| uint64_t size = h->size, e; /* cached immutable count, not peer hdr->size */ |
| 141 |
43 |
| char *base = nda_data(h); |
| 142 |
43 |
▶ | if (nb == 1) { memset(base, (unsigned char)enc[0], (size_t)size); return; } |
Branch: 1 branch
| Branch | True | False | Total |
| if (nb == 1) { memset(base, (unsigned char)enc[0], (size_t)size); return; } |
4 |
39 |
43 |
|
| 143 |
|
| #define NDA_FILL_WIDTH(CT) do { \ |
| 144 |
|
| CT v; memcpy(&v, enc, sizeof v); \ |
| 145 |
|
| for (e = 0; e < size; e++) memcpy(base + e * sizeof v, &v, sizeof v); \ |
| 146 |
|
| } while (0) |
| 147 |
39 |
| switch (nb) { |
| 148 |
524 |
▶ | case 2: NDA_FILL_WIDTH(uint16_t); break; |
Branch: 1 branch
| Branch | True | False | Total |
| case 2: NDA_FILL_WIDTH(uint16_t); break; |
520 |
4 |
524 |
|
| 149 |
819 |
▶ | case 4: NDA_FILL_WIDTH(uint32_t); break; |
Branch: 1 branch
| Branch | True | False | Total |
| case 4: NDA_FILL_WIDTH(uint32_t); break; |
802 |
17 |
819 |
|
| 150 |
890 |
▶ | case 8: NDA_FILL_WIDTH(uint64_t); break; |
Branch: 1 branch
| Branch | True | False | Total |
| case 8: NDA_FILL_WIDTH(uint64_t); break; |
872 |
18 |
890 |
|
| 151 |
|
| } |
| 152 |
|
| #undef NDA_FILL_WIDTH |
| 153 |
|
| } |
| 154 |
|
| |
| 155 |
|
| /* ---------------------------------------------------------------- |
| 156 |
|
| * In-place scalar op over every element: e = e OP s, in the dtype's arithmetic |
| 157 |
|
| * (float dtypes use double; int dtypes use integer arithmetic in the element |
| 158 |
|
| * width, wrapping per C rules). `op` is '+' or '*'. |
| 159 |
|
| * ---------------------------------------------------------------- */ |
| 160 |
|
| /* Integer arithmetic is done in UT -- an unsigned type of at least `unsigned int` |
| 161 |
|
| * rank -- so overflow is a defined wrap (no signed-overflow UB, and no u16/u8 |
| 162 |
|
| * promotion to signed int) and the narrowing cast back to CT gives the documented |
| 163 |
|
| * two's-complement wrap. UT must be uint32_t for the <=32-bit dtypes, uint64_t |
| 164 |
|
| * for the 64-bit ones. */ |
| 165 |
|
| #define NDA_SCALAR_INT(CT, UT) do { \ |
| 166 |
|
| CT *p = (CT *)base; \ |
| 167 |
|
| if (op == '+') { CT s = (CT)siv; for (e=0;e<size;e++) p[e] = (CT)((UT)p[e] + (UT)s); } \ |
| 168 |
|
| else { CT s = (CT)siv; for (e=0;e<size;e++) p[e] = (CT)((UT)p[e] * (UT)s); } \ |
| 169 |
|
| } while (0) |
| 170 |
|
| #define NDA_SCALAR_UINT(CT, UT) do { \ |
| 171 |
|
| CT *p = (CT *)base; \ |
| 172 |
|
| if (op == '+') { CT s = (CT)suv; for (e=0;e<size;e++) p[e] = (CT)((UT)p[e] + (UT)s); } \ |
| 173 |
|
| else { CT s = (CT)suv; for (e=0;e<size;e++) p[e] = (CT)((UT)p[e] * (UT)s); } \ |
| 174 |
|
| } while (0) |
| 175 |
|
| |
| 176 |
|
| /* Read the scalar into its dtype form BEFORE the wrlock: SvNV/SvIV/SvUV can |
| 177 |
|
| * croak (overload / FATAL-numeric / tied FETCH), which under the lock leaks it. */ |
| 178 |
7 |
| static void nda_read_scalar(pTHX_ uint8_t dt, SV *sv, double *s, int64_t *siv, uint64_t *suv) { |
| 179 |
7 |
▶ | SvGETMAGIC(sv); |
Branch: 2 branches
| Branch | True | False | Total |
| SvGETMAGIC(sv); |
0 |
7 |
7 |
| SvGETMAGIC(sv); |
0 |
0 |
0 |
|
| 180 |
7 |
▶ | if (nda_is_float(dt)) *s = SvNV(sv); |
Branch: 1 branch
| Branch | True | False | Total |
| if (nda_is_float(dt)) *s = SvNV(sv); |
3 |
4 |
7 |
|
| 181 |
4 |
▶ | else if (nda_is_signed(dt)) *siv = SvIV(sv); |
Branch: 1 branch
| Branch | True | False | Total |
| else if (nda_is_signed(dt)) *siv = SvIV(sv); |
4 |
0 |
4 |
|
| 182 |
0 |
| else *suv = SvUV(sv); |
| 183 |
7 |
| } |
| 184 |
7 |
| static void nda_scalar_op_locked(NdaHandle *h, double s, int64_t siv, uint64_t suv, int op) { |
| 185 |
7 |
| uint64_t size = h->size, e; /* cached immutable count, not peer hdr->size */ |
| 186 |
7 |
| char *base = nda_data(h); |
| 187 |
7 |
| uint32_t dt = h->dtype; /* cached: peer hdr->dtype must not resize the element */ |
| 188 |
7 |
▶ | if (nda_is_float(dt)) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (nda_is_float(dt)) { |
3 |
4 |
7 |
|
| 189 |
3 |
▶ | if (dt == NDA_F64) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (dt == NDA_F64) { |
1 |
2 |
3 |
|
| 190 |
1 |
| double *p = (double *)base; |
| 191 |
4 |
▶ | if (op == '+') for (e=0;e<size;e++) p[e] += s; else for (e=0;e<size;e++) p[e] *= s; |
Branch: 3 branches
| Branch | True | False | Total |
| if (op == '+') for (e=0;e<size;e++) p[e] += s; else for (e=0;e<size;e++) p[e] *= s; |
0 |
1 |
1 |
| if (op == '+') for (e=0;e<size;e++) p[e] += s; else for (e=0;e<size;e++) p[e] *= s; |
0 |
0 |
0 |
| if (op == '+') for (e=0;e<size;e++) p[e] += s; else for (e=0;e<size;e++) p[e] *= s; |
3 |
1 |
4 |
|
| 192 |
|
| } else { /* F32 */ |
| 193 |
2 |
| float *p = (float *)base; float fs = (float)s; |
| 194 |
10 |
▶ | if (op == '+') for (e=0;e<size;e++) p[e] += fs; else for (e=0;e<size;e++) p[e] *= fs; |
Branch: 3 branches
| Branch | True | False | Total |
| if (op == '+') for (e=0;e<size;e++) p[e] += fs; else for (e=0;e<size;e++) p[e] *= fs; |
1 |
1 |
2 |
| if (op == '+') for (e=0;e<size;e++) p[e] += fs; else for (e=0;e<size;e++) p[e] *= fs; |
4 |
1 |
5 |
| if (op == '+') for (e=0;e<size;e++) p[e] += fs; else for (e=0;e<size;e++) p[e] *= fs; |
4 |
1 |
5 |
|
| 195 |
|
| } |
| 196 |
4 |
▶ | } else if (nda_is_signed(dt)) { |
Branch: 1 branch
| Branch | True | False | Total |
| } else if (nda_is_signed(dt)) { |
4 |
0 |
4 |
|
| 197 |
4 |
| switch (dt) { |
| 198 |
2 |
▶ | case NDA_I64: NDA_SCALAR_INT(int64_t, uint64_t); break; |
Branch: 3 branches
| Branch | True | False | Total |
| case NDA_I64: NDA_SCALAR_INT(int64_t, uint64_t); break; |
0 |
1 |
1 |
| case NDA_I64: NDA_SCALAR_INT(int64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_I64: NDA_SCALAR_INT(int64_t, uint64_t); break; |
1 |
1 |
2 |
|
| 199 |
12 |
▶ | case NDA_I32: NDA_SCALAR_INT(int32_t, uint32_t); break; |
Branch: 3 branches
| Branch | True | False | Total |
| case NDA_I32: NDA_SCALAR_INT(int32_t, uint32_t); break; |
1 |
2 |
3 |
| case NDA_I32: NDA_SCALAR_INT(int32_t, uint32_t); break; |
4 |
1 |
5 |
| case NDA_I32: NDA_SCALAR_INT(int32_t, uint32_t); break; |
5 |
2 |
7 |
|
| 200 |
0 |
▶ | case NDA_I16: NDA_SCALAR_INT(int16_t, uint32_t); break; |
Branch: 3 branches
| Branch | True | False | Total |
| case NDA_I16: NDA_SCALAR_INT(int16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I16: NDA_SCALAR_INT(int16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I16: NDA_SCALAR_INT(int16_t, uint32_t); break; |
0 |
0 |
0 |
|
| 201 |
0 |
▶ | case NDA_I8: NDA_SCALAR_INT(int8_t, uint32_t); break; |
Branch: 3 branches
| Branch | True | False | Total |
| case NDA_I8: NDA_SCALAR_INT(int8_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I8: NDA_SCALAR_INT(int8_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I8: NDA_SCALAR_INT(int8_t, uint32_t); break; |
0 |
0 |
0 |
|
| 202 |
|
| } |
| 203 |
|
| } else { |
| 204 |
0 |
| switch (dt) { |
| 205 |
0 |
▶ | case NDA_U64: NDA_SCALAR_UINT(uint64_t, uint64_t); break; |
Branch: 3 branches
| Branch | True | False | Total |
| case NDA_U64: NDA_SCALAR_UINT(uint64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_U64: NDA_SCALAR_UINT(uint64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_U64: NDA_SCALAR_UINT(uint64_t, uint64_t); break; |
0 |
0 |
0 |
|
| 206 |
0 |
▶ | case NDA_U32: NDA_SCALAR_UINT(uint32_t, uint32_t); break; |
Branch: 3 branches
| Branch | True | False | Total |
| case NDA_U32: NDA_SCALAR_UINT(uint32_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U32: NDA_SCALAR_UINT(uint32_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U32: NDA_SCALAR_UINT(uint32_t, uint32_t); break; |
0 |
0 |
0 |
|
| 207 |
0 |
▶ | case NDA_U16: NDA_SCALAR_UINT(uint16_t, uint32_t); break; |
Branch: 3 branches
| Branch | True | False | Total |
| case NDA_U16: NDA_SCALAR_UINT(uint16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U16: NDA_SCALAR_UINT(uint16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U16: NDA_SCALAR_UINT(uint16_t, uint32_t); break; |
0 |
0 |
0 |
|
| 208 |
0 |
▶ | case NDA_U8: NDA_SCALAR_UINT(uint8_t, uint32_t); break; |
Branch: 3 branches
| Branch | True | False | Total |
| case NDA_U8: NDA_SCALAR_UINT(uint8_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U8: NDA_SCALAR_UINT(uint8_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U8: NDA_SCALAR_UINT(uint8_t, uint32_t); break; |
0 |
0 |
0 |
|
| 209 |
|
| } |
| 210 |
|
| } |
| 211 |
7 |
| } |
| 212 |
|
| |
| 213 |
|
| /* ---------------------------------------------------------------- |
| 214 |
|
| * In-place element-wise array op: a[i] = a[i] OP b[i] (same dtype + size). |
| 215 |
|
| * `op` is '+', '-' or '*'. Both buffers are interpreted in the shared dtype. |
| 216 |
|
| * ---------------------------------------------------------------- */ |
| 217 |
|
| /* float dtypes: native arithmetic (overflow -> +-inf, well-defined) */ |
| 218 |
|
| #define NDA_EW_FLT(CT) do { \ |
| 219 |
|
| CT *pa = (CT *)da; const CT *pb = (const CT *)db; \ |
| 220 |
|
| if (op == '+') for (e=0;e<size;e++) pa[e] = (CT)(pa[e] + pb[e]); \ |
| 221 |
|
| else if (op == '-') for (e=0;e<size;e++) pa[e] = (CT)(pa[e] - pb[e]); \ |
| 222 |
|
| else for (e=0;e<size;e++) pa[e] = (CT)(pa[e] * pb[e]); \ |
| 223 |
|
| } while (0) |
| 224 |
|
| /* integer dtypes: compute in UT (unsigned, >= int rank) for defined wrap -- see |
| 225 |
|
| * the NDA_SCALAR_INT note above. */ |
| 226 |
|
| #define NDA_EW_INT(CT, UT) do { \ |
| 227 |
|
| CT *pa = (CT *)da; const CT *pb = (const CT *)db; \ |
| 228 |
|
| if (op == '+') for (e=0;e<size;e++) pa[e] = (CT)((UT)pa[e] + (UT)pb[e]); \ |
| 229 |
|
| else if (op == '-') for (e=0;e<size;e++) pa[e] = (CT)((UT)pa[e] - (UT)pb[e]); \ |
| 230 |
|
| else for (e=0;e<size;e++) pa[e] = (CT)((UT)pa[e] * (UT)pb[e]); \ |
| 231 |
|
| } while (0) |
| 232 |
|
| |
| 233 |
27 |
| static void nda_elementwise_op_locked(NdaHandle *a, NdaHandle *b, int op) { |
| 234 |
27 |
| uint64_t size = a->size, e; /* cached immutable count (== b->size, checked), not peer hdr->size */ |
| 235 |
27 |
| char *da = nda_data(a); |
| 236 |
27 |
| const char *db = nda_data(b); |
| 237 |
27 |
| switch (a->dtype) { /* cached: peer hdr->dtype must not resize the element */ |
| 238 |
163 |
▶ | case NDA_F64: NDA_EW_FLT(double); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_F64: NDA_EW_FLT(double); break; |
5 |
8 |
13 |
| case NDA_F64: NDA_EW_FLT(double); break; |
54 |
5 |
59 |
| case NDA_F64: NDA_EW_FLT(double); break; |
4 |
4 |
8 |
| case NDA_F64: NDA_EW_FLT(double); break; |
48 |
4 |
52 |
| case NDA_F64: NDA_EW_FLT(double); break; |
48 |
4 |
52 |
|
| 239 |
10 |
▶ | case NDA_F32: NDA_EW_FLT(float); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_F32: NDA_EW_FLT(float); break; |
0 |
2 |
2 |
| case NDA_F32: NDA_EW_FLT(float); break; |
0 |
0 |
0 |
| case NDA_F32: NDA_EW_FLT(float); break; |
1 |
1 |
2 |
| case NDA_F32: NDA_EW_FLT(float); break; |
4 |
1 |
5 |
| case NDA_F32: NDA_EW_FLT(float); break; |
4 |
1 |
5 |
|
| 240 |
0 |
▶ | case NDA_I64: NDA_EW_INT(int64_t, uint64_t); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_I64: NDA_EW_INT(int64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_I64: NDA_EW_INT(int64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_I64: NDA_EW_INT(int64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_I64: NDA_EW_INT(int64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_I64: NDA_EW_INT(int64_t, uint64_t); break; |
0 |
0 |
0 |
|
| 241 |
24 |
▶ | case NDA_I32: NDA_EW_INT(int32_t, uint32_t); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_I32: NDA_EW_INT(int32_t, uint32_t); break; |
4 |
4 |
8 |
| case NDA_I32: NDA_EW_INT(int32_t, uint32_t); break; |
8 |
4 |
12 |
| case NDA_I32: NDA_EW_INT(int32_t, uint32_t); break; |
2 |
2 |
4 |
| case NDA_I32: NDA_EW_INT(int32_t, uint32_t); break; |
4 |
2 |
6 |
| case NDA_I32: NDA_EW_INT(int32_t, uint32_t); break; |
4 |
2 |
6 |
|
| 242 |
0 |
▶ | case NDA_I16: NDA_EW_INT(int16_t, uint32_t); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_I16: NDA_EW_INT(int16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I16: NDA_EW_INT(int16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I16: NDA_EW_INT(int16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I16: NDA_EW_INT(int16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I16: NDA_EW_INT(int16_t, uint32_t); break; |
0 |
0 |
0 |
|
| 243 |
0 |
▶ | case NDA_I8: NDA_EW_INT(int8_t, uint32_t); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_I8: NDA_EW_INT(int8_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I8: NDA_EW_INT(int8_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I8: NDA_EW_INT(int8_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I8: NDA_EW_INT(int8_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_I8: NDA_EW_INT(int8_t, uint32_t); break; |
0 |
0 |
0 |
|
| 244 |
0 |
▶ | case NDA_U64: NDA_EW_INT(uint64_t, uint64_t); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_U64: NDA_EW_INT(uint64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_U64: NDA_EW_INT(uint64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_U64: NDA_EW_INT(uint64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_U64: NDA_EW_INT(uint64_t, uint64_t); break; |
0 |
0 |
0 |
| case NDA_U64: NDA_EW_INT(uint64_t, uint64_t); break; |
0 |
0 |
0 |
|
| 245 |
0 |
▶ | case NDA_U32: NDA_EW_INT(uint32_t, uint32_t); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_U32: NDA_EW_INT(uint32_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U32: NDA_EW_INT(uint32_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U32: NDA_EW_INT(uint32_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U32: NDA_EW_INT(uint32_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U32: NDA_EW_INT(uint32_t, uint32_t); break; |
0 |
0 |
0 |
|
| 246 |
3 |
▶ | case NDA_U16: NDA_EW_INT(uint16_t, uint32_t); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_U16: NDA_EW_INT(uint16_t, uint32_t); break; |
0 |
1 |
1 |
| case NDA_U16: NDA_EW_INT(uint16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U16: NDA_EW_INT(uint16_t, uint32_t); break; |
0 |
1 |
1 |
| case NDA_U16: NDA_EW_INT(uint16_t, uint32_t); break; |
0 |
0 |
0 |
| case NDA_U16: NDA_EW_INT(uint16_t, uint32_t); break; |
2 |
1 |
3 |
|
| 247 |
8 |
▶ | case NDA_U8: NDA_EW_INT(uint8_t, uint32_t); break; |
Branch: 5 branches
| Branch | True | False | Total |
| case NDA_U8: NDA_EW_INT(uint8_t, uint32_t); break; |
1 |
2 |
3 |
| case NDA_U8: NDA_EW_INT(uint8_t, uint32_t); break; |
2 |
1 |
3 |
| case NDA_U8: NDA_EW_INT(uint8_t, uint32_t); break; |
1 |
1 |
2 |
| case NDA_U8: NDA_EW_INT(uint8_t, uint32_t); break; |
2 |
1 |
3 |
| case NDA_U8: NDA_EW_INT(uint8_t, uint32_t); break; |
1 |
1 |
2 |
|
| 248 |
|
| } |
| 249 |
27 |
| } |
| 250 |
|
| |
| 251 |
|
| /* Acquire receiver `a`'s WRITE lock and other `b`'s READ lock in a |
| 252 |
|
| * globally-consistent order keyed on each array's shared identity (array_id), |
| 253 |
|
| * NOT the process-local handle pointer, so two unrelated processes mapping the |
| 254 |
|
| * same pair agree on order and cannot deadlock. a and b must be DISTINCT |
| 255 |
|
| * underlying arrays (the caller has already handled the same-array case). |
| 256 |
|
| * |
| 257 |
|
| * `b` may be a frozen/readonly (PROT_READ) handle: its reader-slot/wlock |
| 258 |
|
| * bookkeeping lives in the header region, which such a handle cannot WRITE to, |
| 259 |
|
| * so rdlock/rdunlock on it would SIGSEGV. Once sealed, b's data can never |
| 260 |
|
| * change (every mutator on it is refused), so skipping its lock entirely is |
| 261 |
|
| * safe -- only `a` (never readonly; the caller already refused that) takes a |
| 262 |
|
| * lock. `a` is always the caller's own (mutable) handle, so array_id ordering |
| 263 |
|
| * is only needed -- and only meaningful -- when both sides actually lock. */ |
| 264 |
34 |
| static int nda_array_before(const NdaHandle *a, const NdaHandle *b) { |
| 265 |
34 |
▶ | if (a->hdr->array_id != b->hdr->array_id) |
Branch: 1 branch
| Branch | True | False | Total |
| if (a->hdr->array_id != b->hdr->array_id) |
24 |
10 |
34 |
|
| 266 |
24 |
| return a->hdr->array_id < b->hdr->array_id; |
| 267 |
10 |
▶ | if (a->device != b->device) return a->device < b->device; |
Branch: 1 branch
| Branch | True | False | Total |
| if (a->device != b->device) return a->device < b->device; |
4 |
6 |
10 |
|
| 268 |
6 |
| return a->inode < b->inode; /* copied files retain array_id */ |
| 269 |
|
| } |
| 270 |
|
| |
| 271 |
20 |
| static void nda_lock_pair(NdaHandle *a, NdaHandle *b) { |
| 272 |
20 |
▶ | if (b->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (b->readonly) { |
3 |
17 |
20 |
|
| 273 |
3 |
| nda_rwlock_wrlock(a); |
| 274 |
3 |
| return; |
| 275 |
|
| } |
| 276 |
17 |
▶ | if (nda_array_before(a, b)) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (nda_array_before(a, b)) { |
7 |
10 |
17 |
|
| 277 |
7 |
| nda_rwlock_wrlock(a); |
| 278 |
7 |
| nda_rwlock_rdlock(b); |
| 279 |
|
| } else { |
| 280 |
10 |
| nda_rwlock_rdlock(b); |
| 281 |
10 |
| nda_rwlock_wrlock(a); |
| 282 |
|
| } |
| 283 |
|
| } |
| 284 |
20 |
| static void nda_unlock_pair(NdaHandle *a, NdaHandle *b) { |
| 285 |
20 |
▶ | if (b->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (b->readonly) { |
3 |
17 |
20 |
|
| 286 |
3 |
| nda_rwlock_wrunlock(a); |
| 287 |
3 |
| return; |
| 288 |
|
| } |
| 289 |
17 |
▶ | if (nda_array_before(a, b)) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (nda_array_before(a, b)) { |
7 |
10 |
17 |
|
| 290 |
7 |
| nda_rwlock_rdunlock(b); |
| 291 |
7 |
| nda_rwlock_wrunlock(a); |
| 292 |
|
| } else { |
| 293 |
10 |
| nda_rwlock_wrunlock(a); |
| 294 |
10 |
| nda_rwlock_rdunlock(b); |
| 295 |
|
| } |
| 296 |
|
| } |
| 297 |
|
| |
| 298 |
23 |
| static int nda_same_array(pTHX_ NdaHandle *a, NdaHandle *b) { |
| 299 |
23 |
▶ | if (a->hdr->array_id != b->hdr->array_id) return 0; |
Branch: 1 branch
| Branch | True | False | Total |
| if (a->hdr->array_id != b->hdr->array_id) return 0; |
15 |
8 |
23 |
|
| 300 |
8 |
| NdaHandle *pair[2] = { a, b }; |
| 301 |
24 |
▶ | for (int i = 0; i < 2; i++) { |
Branch: 1 branch
| Branch | True | False | Total |
| for (int i = 0; i < 2; i++) { |
16 |
8 |
24 |
|
| 302 |
16 |
| NdaHandle *h = pair[i]; |
| 303 |
16 |
▶ | if (!h->inode && h->backing_fd >= 0) { |
Branch: 2 branches
| Branch | True | False | Total |
| if (!h->inode && h->backing_fd >= 0) { |
3 |
13 |
16 |
| if (!h->inode && h->backing_fd >= 0) { |
3 |
0 |
3 |
|
| 304 |
|
| struct stat st; |
| 305 |
3 |
▶ | if (fstat(h->backing_fd, &st) < 0) croak("Data::NDArray::Shared: fstat: %s", strerror(errno)); |
Branch: 1 branch
| Branch | True | False | Total |
| if (fstat(h->backing_fd, &st) < 0) croak("Data::NDArray::Shared: fstat: %s", strerror(errno)); |
0 |
3 |
3 |
|
| 306 |
3 |
| h->device = st.st_dev; |
| 307 |
3 |
| h->inode = st.st_ino; |
| 308 |
|
| } |
| 309 |
|
| } |
| 310 |
8 |
▶ | return a->inode && a->inode == b->inode && a->device == b->device; |
Branch: 3 branches
| Branch | True | False | Total |
| return a->inode && a->inode == b->inode && a->device == b->device; |
8 |
0 |
8 |
| return a->inode && a->inode == b->inode && a->device == b->device; |
3 |
5 |
8 |
| return a->inode && a->inode == b->inode && a->device == b->device; |
3 |
0 |
3 |
|
| 311 |
|
| } |
| 312 |
|
| |
| 313 |
|
| /* Shared body of add/subtract/multiply: validate `other`, take the locks in a |
| 314 |
|
| * deadlock-free id order (or a single wrlock for the same mapping), apply the |
| 315 |
|
| * element-wise op, bump stat_ops, and unlock. `op` is '+', '-' or '*'; `who` |
| 316 |
|
| * is the fully-qualified method name used in croak messages. */ |
| 317 |
40 |
| static void nda_do_elementwise(pTHX_ SV *self_sv, NdaHandle *h, SV *other, int op, const char *who) { |
| 318 |
40 |
| NdaHandle *h0 = h; PERL_UNUSED_VAR(h0); /* identity anchor for the REEXTRACT below */ |
| 319 |
40 |
▶ | if (h->readonly) croak("%s: array is frozen (read-only)", who); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("%s: array is frozen (read-only)", who); |
4 |
36 |
40 |
|
| 320 |
36 |
▶ | if (!sv_isobject(other) || !sv_derived_from(other, "Data::NDArray::Shared")) |
Branch: 2 branches
| Branch | True | False | Total |
| if (!sv_isobject(other) || !sv_derived_from(other, "Data::NDArray::Shared")) |
33 |
3 |
36 |
| if (!sv_isobject(other) || !sv_derived_from(other, "Data::NDArray::Shared")) |
0 |
33 |
33 |
|
| 321 |
3 |
| croak("%s: expected a Data::NDArray::Shared object", who); |
| 322 |
33 |
| NdaHandle *o = INT2PTR(NdaHandle*, SvIV(SvRV(other))); |
| 323 |
33 |
▶ | if (!o) croak("Attempted to use a destroyed Data::NDArray::Shared object"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (!o) croak("Attempted to use a destroyed Data::NDArray::Shared object"); |
0 |
33 |
33 |
|
| 324 |
|
| /* sv_isobject/sv_derived_from above begin with SvGETMAGIC(other), so a tied |
| 325 |
|
| * `other` can have run Perl that destroyed self before h is used below. |
| 326 |
|
| * `h` is this function's own parameter, so re-reading it here is enough; |
| 327 |
|
| * `o` was read after the magic and needs no re-read. */ |
| 328 |
33 |
▶ | REEXTRACT(self_sv); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self_sv); |
0 |
33 |
33 |
| REEXTRACT(self_sv); |
0 |
33 |
33 |
|
| 329 |
33 |
▶ | if (o->dtype != h->dtype) /* cached: the kernel interprets both in the cached dtype */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (o->dtype != h->dtype) /* cached: the kernel interprets both in the cached dtype */ |
3 |
30 |
33 |
|
| 330 |
3 |
| croak("%s: dtype mismatch", who); |
| 331 |
30 |
▶ | if (o->size != h->size) /* cached immutable counts, not peer hdr->size */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (o->size != h->size) /* cached immutable counts, not peer hdr->size */ |
3 |
27 |
30 |
|
| 332 |
3 |
| croak("%s: size mismatch (%" UVuf " vs %" UVuf ")", |
| 333 |
|
| who, (UV)h->size, (UV)o->size); |
| 334 |
27 |
▶ | if (o == h || nda_same_array(aTHX_ h, o)) { |
Branch: 2 branches
| Branch | True | False | Total |
| if (o == h || nda_same_array(aTHX_ h, o)) { |
23 |
4 |
27 |
| if (o == h || nda_same_array(aTHX_ h, o)) { |
3 |
20 |
23 |
|
| 335 |
7 |
| nda_rwlock_wrlock(h); |
| 336 |
7 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); croak("%s: array is frozen (read-only)", who); } |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); croak("%s: array is frozen (read-only)", who); } |
0 |
7 |
7 |
|
| 337 |
7 |
| nda_elementwise_op_locked(h, h, op); /* self: +=->double, -=->zero, *=->square */ |
| 338 |
7 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 339 |
7 |
| nda_rwlock_wrunlock(h); |
| 340 |
|
| } else { |
| 341 |
|
| /* o may be a frozen/readonly (PROT_READ) other -- nda_lock_pair/ |
| 342 |
|
| * nda_unlock_pair skip its lock in that case (see their comment); |
| 343 |
|
| * its data is immutable so reading it lock-free is safe. */ |
| 344 |
20 |
| nda_lock_pair(h, o); |
| 345 |
20 |
▶ | if (h->hdr->sealed) { nda_unlock_pair(h, o); croak("%s: array is frozen (read-only)", who); } |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_unlock_pair(h, o); croak("%s: array is frozen (read-only)", who); } |
0 |
20 |
20 |
|
| 346 |
20 |
| nda_elementwise_op_locked(h, o, op); |
| 347 |
20 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 348 |
20 |
| nda_unlock_pair(h, o); |
| 349 |
|
| } |
| 350 |
27 |
| } |
| 351 |
|
| |
| 352 |
|
| /* Resolve the uniform (path, dtype, @shape) argument convention shared by new |
| 353 |
|
| * and new_memfd, matching the rest of the Data::*::Shared family: |
| 354 |
|
| * ($path_or_undef, $dtype, @shape) |
| 355 |
|
| * ST(1) is ALWAYS the path/memfd label (an SV that may carry undef for an |
| 356 |
|
| * anonymous mapping), ST(2) is ALWAYS the dtype name string, and ST(3).. are |
| 357 |
|
| * the dimensions. Sets *out_label, *out_dt, fills dims[] and returns ndim. |
| 358 |
|
| * Croaks (no resource held) on a missing/unknown dtype or bad dims. */ |
| 359 |
142 |
| static uint32_t nda_resolve_ctor_args(pTHX_ I32 ax, I32 items, const char *who, |
| 360 |
|
| SV **out_label, int *out_dt, |
| 361 |
|
| uint64_t dims[NDA_MAX_DIMS]) { |
| 362 |
142 |
| SV **sp_mark = PL_stack_base + ax; /* ST(0) == class */ |
| 363 |
142 |
▶ | if (items < 3) croak("%s: dtype required", who); |
Branch: 1 branch
| Branch | True | False | Total |
| if (items < 3) croak("%s: dtype required", who); |
0 |
142 |
142 |
|
| 364 |
142 |
| SV *label = sp_mark[1]; /* path/name SV (may be undef) */ |
| 365 |
142 |
| SV *dsv = sp_mark[2]; /* dtype name */ |
| 366 |
142 |
| STRLEN dlen; const char *dname = SvPVbyte(dsv, dlen); |
| 367 |
142 |
| int dt = nda_dtype_from_name(dname, dlen); |
| 368 |
142 |
▶ | if (dt < 0) croak("%s: unknown dtype", who); |
Branch: 1 branch
| Branch | True | False | Total |
| if (dt < 0) croak("%s: unknown dtype", who); |
2 |
140 |
142 |
|
| 369 |
|
| /* parse @shape from ST(3).. */ |
| 370 |
140 |
| I32 nd = items - 3; |
| 371 |
140 |
▶ | if (nd < 1) croak("%s: at least one dimension required", who); |
Branch: 1 branch
| Branch | True | False | Total |
| if (nd < 1) croak("%s: at least one dimension required", who); |
1 |
139 |
140 |
|
| 372 |
139 |
▶ | if (nd > NDA_MAX_DIMS) croak("%s: too many dimensions (max %d)", who, NDA_MAX_DIMS); |
Branch: 1 branch
| Branch | True | False | Total |
| if (nd > NDA_MAX_DIMS) croak("%s: too many dimensions (max %d)", who, NDA_MAX_DIMS); |
1 |
138 |
139 |
|
| 373 |
320 |
▶ | for (I32 i = 0; i < nd; i++) { |
Branch: 1 branch
| Branch | True | False | Total |
| for (I32 i = 0; i < nd; i++) { |
184 |
136 |
320 |
|
| 374 |
184 |
| IV iv = SvIV(sp_mark[3 + i]); |
| 375 |
184 |
▶ | if (iv < 1) croak("%s: dimension %d must be >= 1", who, (int)i); |
Branch: 1 branch
| Branch | True | False | Total |
| if (iv < 1) croak("%s: dimension %d must be >= 1", who, (int)i); |
2 |
182 |
184 |
|
| 376 |
182 |
| dims[i] = (uint64_t)iv; |
| 377 |
|
| } |
| 378 |
136 |
| *out_label = label; |
| 379 |
136 |
| *out_dt = dt; |
| 380 |
136 |
| return (uint32_t)nd; |
| 381 |
|
| } |
| 382 |
|
| |
| 383 |
|
| MODULE = Data::NDArray::Shared PACKAGE = Data::NDArray::Shared |
| 384 |
|
| |
| 385 |
|
| PROTOTYPES: DISABLE |
| 386 |
|
| |
| 387 |
|
| SV * |
| 388 |
|
| new(class, ...) |
| 389 |
|
| const char *class |
| 390 |
|
| PREINIT: |
| 391 |
|
| char errbuf[NDA_ERR_BUFLEN]; |
| 392 |
|
| uint64_t dims[NDA_MAX_DIMS]; |
| 393 |
|
| CODE: |
| 394 |
|
| { |
| 395 |
|
| SV *label; int dt; |
| 396 |
|
| /* Opt-in cross-user sharing: an optional trailing \%opts (e.g. |
| 397 |
|
| { mode => 0660 }) sets the backing-file mode; default 0600. @shape |
| 398 |
|
| is greedy, so mode cannot be a bare trailing positional (0600 is a |
| 399 |
|
| valid u8 dimension) -- it rides in a hashref, unambiguous and inert |
| 400 |
|
| for every existing ($path,$dtype,@shape) call. */ |
| 401 |
137 |
| mode_t mode = 0600; |
| 402 |
137 |
| I32 eff_items = items; |
| 403 |
137 |
▶ | if (items > 3) SvGETMAGIC(ST(items - 1)); /* a tied/overloaded scalar may FETCH to the opts hashref */ |
Branch: 3 branches
| Branch | True | False | Total |
| if (items > 3) SvGETMAGIC(ST(items - 1)); /* a tied/overloaded scalar may FETCH to the opts hashref */ |
135 |
2 |
137 |
| if (items > 3) SvGETMAGIC(ST(items - 1)); /* a tied/overloaded scalar may FETCH to the opts hashref */ |
0 |
135 |
135 |
| if (items > 3) SvGETMAGIC(ST(items - 1)); /* a tied/overloaded scalar may FETCH to the opts hashref */ |
0 |
0 |
0 |
|
| 404 |
137 |
▶ | if (items > 3 && SvROK(ST(items - 1)) |
Branch: 2 branches
| Branch | True | False | Total |
| if (items > 3 && SvROK(ST(items - 1)) |
135 |
2 |
137 |
| if (items > 3 && SvROK(ST(items - 1)) |
0 |
135 |
135 |
|
| 405 |
0 |
▶ | && SvTYPE(SvRV(ST(items - 1))) == SVt_PVHV) { |
Branch: 1 branch
| Branch | True | False | Total |
| && SvTYPE(SvRV(ST(items - 1))) == SVt_PVHV) { |
0 |
0 |
0 |
|
| 406 |
0 |
| HV *opts = (HV *)SvRV(ST(items - 1)); |
| 407 |
0 |
| SV **msv = hv_fetchs(opts, "mode", 0); |
| 408 |
0 |
▶ | if (msv && (SvGETMAGIC(*msv), SvOK(*msv))) mode = (mode_t)SvUV(*msv); |
Branch: 4 branches
| Branch | True | False | Total |
| if (msv && (SvGETMAGIC(*msv), SvOK(*msv))) mode = (mode_t)SvUV(*msv); |
0 |
0 |
0 |
| if (msv && (SvGETMAGIC(*msv), SvOK(*msv))) mode = (mode_t)SvUV(*msv); |
0 |
0 |
0 |
| if (msv && (SvGETMAGIC(*msv), SvOK(*msv))) mode = (mode_t)SvUV(*msv); |
0 |
0 |
0 |
| if (msv && (SvGETMAGIC(*msv), SvOK(*msv))) mode = (mode_t)SvUV(*msv); |
0 |
0 |
0 |
|
| 409 |
0 |
| eff_items = items - 1; |
| 410 |
|
| } |
| 411 |
137 |
| uint32_t ndim = nda_resolve_ctor_args(aTHX_ ax, eff_items, |
| 412 |
|
| "Data::NDArray::Shared->new", &label, &dt, dims); /* croaks before any alloc */ |
| 413 |
166 |
▶ | const char *p = (SvGETMAGIC(label), SvOK(label)) ? SvPV_nolen(label) : NULL; |
Branch: 3 branches
| Branch | True | False | Total |
| const char *p = (SvGETMAGIC(label), SvOK(label)) ? SvPV_nolen(label) : NULL; |
0 |
131 |
131 |
| const char *p = (SvGETMAGIC(label), SvOK(label)) ? SvPV_nolen(label) : NULL; |
0 |
0 |
0 |
| const char *p = (SvGETMAGIC(label), SvOK(label)) ? SvPV_nolen(label) : NULL; |
35 |
96 |
131 |
|
| 414 |
131 |
| NdaHandle *h = nda_create(p, dt, dims, ndim, mode, errbuf); |
| 415 |
131 |
▶ | if (!h) croak("Data::NDArray::Shared->new: %s", errbuf[0] ? errbuf : "out of memory"); |
Branch: 2 branches
| Branch | True | False | Total |
| if (!h) croak("Data::NDArray::Shared->new: %s", errbuf[0] ? errbuf : "out of memory"); |
6 |
125 |
131 |
| if (!h) croak("Data::NDArray::Shared->new: %s", errbuf[0] ? errbuf : "out of memory"); |
6 |
0 |
6 |
|
| 416 |
125 |
| MAKE_OBJ(class, h); |
| 417 |
|
| } |
| 418 |
|
| OUTPUT: |
| 419 |
|
| RETVAL |
| 420 |
|
| |
| 421 |
|
| SV * |
| 422 |
|
| new_memfd(class, ...) |
| 423 |
|
| const char *class |
| 424 |
|
| PREINIT: |
| 425 |
|
| char errbuf[NDA_ERR_BUFLEN]; |
| 426 |
|
| uint64_t dims[NDA_MAX_DIMS]; |
| 427 |
|
| CODE: |
| 428 |
|
| { |
| 429 |
|
| SV *label; int dt; |
| 430 |
5 |
| uint32_t ndim = nda_resolve_ctor_args(aTHX_ ax, items, |
| 431 |
|
| "Data::NDArray::Shared->new_memfd", &label, &dt, dims); |
| 432 |
6 |
▶ | const char *nm = (SvGETMAGIC(label), SvOK(label)) ? SvPV_nolen(label) : NULL; |
Branch: 3 branches
| Branch | True | False | Total |
| const char *nm = (SvGETMAGIC(label), SvOK(label)) ? SvPV_nolen(label) : NULL; |
0 |
5 |
5 |
| const char *nm = (SvGETMAGIC(label), SvOK(label)) ? SvPV_nolen(label) : NULL; |
0 |
0 |
0 |
| const char *nm = (SvGETMAGIC(label), SvOK(label)) ? SvPV_nolen(label) : NULL; |
1 |
4 |
5 |
|
| 433 |
5 |
| NdaHandle *h = nda_create_memfd(nm, dt, dims, ndim, errbuf); |
| 434 |
5 |
▶ | if (!h) croak("Data::NDArray::Shared->new_memfd: %s", errbuf[0] ? errbuf : "out of memory"); |
Branch: 2 branches
| Branch | True | False | Total |
| if (!h) croak("Data::NDArray::Shared->new_memfd: %s", errbuf[0] ? errbuf : "out of memory"); |
0 |
5 |
5 |
| if (!h) croak("Data::NDArray::Shared->new_memfd: %s", errbuf[0] ? errbuf : "out of memory"); |
0 |
0 |
0 |
|
| 435 |
5 |
| MAKE_OBJ(class, h); |
| 436 |
|
| } |
| 437 |
|
| OUTPUT: |
| 438 |
|
| RETVAL |
| 439 |
|
| |
| 440 |
|
| SV * |
| 441 |
|
| new_from_fd(class, fd) |
| 442 |
|
| const char *class |
| 443 |
|
| int fd |
| 444 |
|
| PREINIT: |
| 445 |
|
| char errbuf[NDA_ERR_BUFLEN]; |
| 446 |
|
| CODE: |
| 447 |
5 |
| NdaHandle *h = nda_open_fd(fd, errbuf); |
| 448 |
5 |
▶ | if (!h) croak("Data::NDArray::Shared->new_from_fd: %s", errbuf[0] ? errbuf : "out of memory"); |
Branch: 2 branches
| Branch | True | False | Total |
| if (!h) croak("Data::NDArray::Shared->new_from_fd: %s", errbuf[0] ? errbuf : "out of memory"); |
2 |
3 |
5 |
| if (!h) croak("Data::NDArray::Shared->new_from_fd: %s", errbuf[0] ? errbuf : "out of memory"); |
2 |
0 |
2 |
|
| 449 |
3 |
| MAKE_OBJ(class, h); |
| 450 |
|
| OUTPUT: |
| 451 |
|
| RETVAL |
| 452 |
|
| |
| 453 |
|
| SV * |
| 454 |
|
| new_readonly(class, path) |
| 455 |
|
| const char *class |
| 456 |
|
| SV *path |
| 457 |
|
| PREINIT: |
| 458 |
|
| char errbuf[NDA_ERR_BUFLEN]; |
| 459 |
|
| CODE: |
| 460 |
|
| /* Open a FROZEN (sealed) file read-only: O_RDONLY + PROT_READ, no lock |
| 461 |
|
| ever. A sealed file is immutable and no read path writes the mapping, |
| 462 |
|
| so queries take no reader-slot / rwlock traffic and any number of |
| 463 |
|
| processes can share one PROT_READ mapping (same architecture; the |
| 464 |
|
| magic rejects a wrong-endian file). */ |
| 465 |
13 |
▶ | const char *p = (SvGETMAGIC(path), SvOK(path)) ? SvPV_nolen(path) : NULL; |
Branch: 3 branches
| Branch | True | False | Total |
| const char *p = (SvGETMAGIC(path), SvOK(path)) ? SvPV_nolen(path) : NULL; |
0 |
7 |
7 |
| const char *p = (SvGETMAGIC(path), SvOK(path)) ? SvPV_nolen(path) : NULL; |
0 |
0 |
0 |
| const char *p = (SvGETMAGIC(path), SvOK(path)) ? SvPV_nolen(path) : NULL; |
6 |
1 |
7 |
|
| 466 |
7 |
▶ | if (!p) croak("Data::NDArray::Shared->new_readonly: path is required"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (!p) croak("Data::NDArray::Shared->new_readonly: path is required"); |
1 |
6 |
7 |
|
| 467 |
6 |
| NdaHandle *h = nda_open_readonly(p, errbuf); |
| 468 |
6 |
▶ | if (!h) croak("Data::NDArray::Shared->new_readonly: %s", errbuf); |
Branch: 1 branch
| Branch | True | False | Total |
| if (!h) croak("Data::NDArray::Shared->new_readonly: %s", errbuf); |
2 |
4 |
6 |
|
| 469 |
4 |
| MAKE_OBJ(class, h); |
| 470 |
|
| OUTPUT: |
| 471 |
|
| RETVAL |
| 472 |
|
| |
| 473 |
|
| void |
| 474 |
|
| DESTROY(self) |
| 475 |
|
| SV *self |
| 476 |
|
| CODE: |
| 477 |
139 |
▶ | if (sv_isobject(self) && sv_derived_from(self, "Data::NDArray::Shared")) { |
Branch: 2 branches
| Branch | True | False | Total |
| if (sv_isobject(self) && sv_derived_from(self, "Data::NDArray::Shared")) { |
139 |
0 |
139 |
| if (sv_isobject(self) && sv_derived_from(self, "Data::NDArray::Shared")) { |
139 |
0 |
139 |
|
| 478 |
139 |
| NdaHandle *h = INT2PTR(NdaHandle*, SvIV(SvRV(self))); |
| 479 |
139 |
▶ | if (h) { sv_setiv(SvRV(self), 0); nda_destroy(h); } /* null first: activates EXTRACT's use-after-destroy croak + makes a double DESTROY a no-op */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (h) { sv_setiv(SvRV(self), 0); nda_destroy(h); } /* null first: activates EXTRACT's use-after-destroy croak + makes a double DESTROY a no-op */ |
137 |
2 |
139 |
|
| 480 |
|
| } |
| 481 |
|
| |
| 482 |
|
| SV * |
| 483 |
|
| get(self, ...) |
| 484 |
|
| SV *self |
| 485 |
|
| PREINIT: |
| 486 |
251 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
251 |
0 |
251 |
| EXTRACT(self); |
0 |
251 |
251 |
| EXTRACT(self); |
1 |
250 |
251 |
|
| 487 |
|
| uint64_t idx[NDA_MAX_DIMS]; |
| 488 |
|
| CODE: |
| 489 |
|
| { |
| 490 |
250 |
| uint32_t nix = (uint32_t)(items - 1); |
| 491 |
250 |
▶ | if (nix > NDA_MAX_DIMS) |
Branch: 1 branch
| Branch | True | False | Total |
| if (nix > NDA_MAX_DIMS) |
0 |
250 |
250 |
|
| 492 |
0 |
| croak("Data::NDArray::Shared->get: too many indices (%u)", nix); |
| 493 |
959 |
▶ | for (uint32_t d = 0; d < nix; d++) /* read the args before locking */ |
Branch: 1 branch
| Branch | True | False | Total |
| for (uint32_t d = 0; d < nix; d++) /* read the args before locking */ |
709 |
250 |
959 |
|
| 494 |
709 |
| idx[d] = (uint64_t)SvUV(ST(1 + d)); |
| 495 |
250 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
250 |
250 |
| REEXTRACT(self); |
0 |
250 |
250 |
|
| 496 |
|
| char _raw[8]; uint32_t ndim; |
| 497 |
250 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
17 |
233 |
250 |
|
| 498 |
|
| /* Frozen: reshape (the only thing that can change ndim/shape/strides) |
| 499 |
|
| is refused, so they can never change -- safe to read lock-free. |
| 500 |
|
| Taking the rwlock here would WRITE into the PROT_READ-mapped header |
| 501 |
|
| (reader-slot claim) and SIGSEGV. */ |
| 502 |
17 |
| ndim = h->hdr->ndim; |
| 503 |
17 |
▶ | if (nix != ndim) |
Branch: 1 branch
| Branch | True | False | Total |
| if (nix != ndim) |
1 |
16 |
17 |
|
| 504 |
1 |
| croak("Data::NDArray::Shared->get: expected %u indices, got %u", ndim, nix); |
| 505 |
46 |
▶ | for (uint32_t d = 0; d < ndim; d++) |
Branch: 1 branch
| Branch | True | False | Total |
| for (uint32_t d = 0; d < ndim; d++) |
31 |
15 |
46 |
|
| 506 |
31 |
▶ | if (idx[d] >= h->hdr->shape[d]) |
Branch: 1 branch
| Branch | True | False | Total |
| if (idx[d] >= h->hdr->shape[d]) |
1 |
30 |
31 |
|
| 507 |
1 |
| croak("Data::NDArray::Shared->get: index %u = %" UVuf " out of range (dim size %" UVuf ")", |
| 508 |
|
| d, (UV)idx[d], (UV)h->hdr->shape[d]); |
| 509 |
15 |
| uint64_t off = nda_flat_offset(h, idx, ndim); |
| 510 |
15 |
▶ | if (off >= h->size) /* cached immutable count, not peer hdr->size */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (off >= h->size) /* cached immutable count, not peer hdr->size */ |
0 |
15 |
15 |
|
| 511 |
0 |
| croak("Data::NDArray::Shared->get: computed offset out of range"); |
| 512 |
15 |
| nda_read_raw(h, off, _raw); |
| 513 |
|
| } else { |
| 514 |
|
| /* Validate shape + compute the flat offset UNDER the read lock: a concurrent |
| 515 |
|
| reshape() mutates ndim/shape/strides under the write lock, so doing this |
| 516 |
|
| lock-free could bound-check one shape yet offset with another -> OOB. */ |
| 517 |
233 |
| nda_rwlock_rdlock(h); |
| 518 |
233 |
| ndim = h->hdr->ndim; |
| 519 |
233 |
▶ | if (nix != ndim) { nda_rwlock_rdunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (nix != ndim) { nda_rwlock_rdunlock(h); |
2 |
231 |
233 |
|
| 520 |
2 |
| croak("Data::NDArray::Shared->get: expected %u indices, got %u", ndim, nix); } |
| 521 |
898 |
▶ | for (uint32_t d = 0; d < ndim; d++) |
Branch: 1 branch
| Branch | True | False | Total |
| for (uint32_t d = 0; d < ndim; d++) |
670 |
228 |
898 |
|
| 522 |
670 |
▶ | if (idx[d] >= h->hdr->shape[d]) { uint64_t sz = h->hdr->shape[d]; nda_rwlock_rdunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (idx[d] >= h->hdr->shape[d]) { uint64_t sz = h->hdr->shape[d]; nda_rwlock_rdunlock(h); |
3 |
667 |
670 |
|
| 523 |
3 |
| croak("Data::NDArray::Shared->get: index %u = %" UVuf " out of range (dim size %" UVuf ")", |
| 524 |
|
| d, (UV)idx[d], (UV)sz); } |
| 525 |
228 |
| uint64_t off = nda_flat_offset(h, idx, ndim); |
| 526 |
|
| /* Layer B: strides[] are read from the shared segment; a local peer with |
| 527 |
|
| write access to the backing file could corrupt them so an in-range |
| 528 |
|
| multi-index still yields a flat offset past the element count -> OOB |
| 529 |
|
| read. Canonical row-major strides always keep off < size, so this |
| 530 |
|
| never fires for good data. */ |
| 531 |
228 |
▶ | if (off >= h->size) { nda_rwlock_rdunlock(h); /* cached immutable count, not peer hdr->size */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (off >= h->size) { nda_rwlock_rdunlock(h); /* cached immutable count, not peer hdr->size */ |
0 |
228 |
228 |
|
| 532 |
0 |
| croak("Data::NDArray::Shared->get: computed offset out of range"); } |
| 533 |
228 |
| nda_read_raw(h, off, _raw); /* copy element under the lock */ |
| 534 |
228 |
| nda_rwlock_rdunlock(h); |
| 535 |
|
| } |
| 536 |
243 |
| RETVAL = nda_sv_from_raw(aTHX_ h->dtype, _raw); /* build SV after unlock */ |
| 537 |
|
| } |
| 538 |
|
| OUTPUT: |
| 539 |
|
| RETVAL |
| 540 |
|
| |
| 541 |
|
| void |
| 542 |
|
| set(self, ...) |
| 543 |
|
| SV *self |
| 544 |
|
| PREINIT: |
| 545 |
223 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
223 |
0 |
223 |
| EXTRACT(self); |
0 |
223 |
223 |
| EXTRACT(self); |
0 |
223 |
223 |
|
| 546 |
|
| uint64_t idx[NDA_MAX_DIMS]; |
| 547 |
|
| CODE: |
| 548 |
|
| { |
| 549 |
223 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->set: array is frozen (read-only)"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->set: array is frozen (read-only)"); |
2 |
221 |
223 |
|
| 550 |
221 |
▶ | if (items < 2) croak("Data::NDArray::Shared->set: expected indices + value"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (items < 2) croak("Data::NDArray::Shared->set: expected indices + value"); |
0 |
221 |
221 |
|
| 551 |
221 |
| uint32_t nix = (uint32_t)(items - 2); /* self + ndim indices + value */ |
| 552 |
221 |
▶ | if (nix > NDA_MAX_DIMS) |
Branch: 1 branch
| Branch | True | False | Total |
| if (nix > NDA_MAX_DIMS) |
0 |
221 |
221 |
|
| 553 |
0 |
| croak("Data::NDArray::Shared->set: too many indices (%u)", nix); |
| 554 |
221 |
| SV *val = ST(items - 1); |
| 555 |
872 |
▶ | for (uint32_t d = 0; d < nix; d++) /* read the args before locking */ |
Branch: 1 branch
| Branch | True | False | Total |
| for (uint32_t d = 0; d < nix; d++) /* read the args before locking */ |
651 |
221 |
872 |
|
| 556 |
651 |
| idx[d] = (uint64_t)SvUV(ST(1 + d)); |
| 557 |
221 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
221 |
221 |
| REEXTRACT(self); |
0 |
221 |
221 |
|
| 558 |
221 |
| char enc[8]; uint32_t nb = nda_encode(aTHX_ h->dtype, val, enc); /* encode val lock-free */ |
| 559 |
|
| /* Validate + compute the offset UNDER the write lock (see get: serializes the |
| 560 |
|
| shape read + offset against a concurrent reshape to prevent an OOB index). */ |
| 561 |
221 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
221 |
221 |
| REEXTRACT(self); |
0 |
221 |
221 |
|
| 562 |
221 |
| nda_rwlock_wrlock(h); |
| 563 |
221 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
1 |
220 |
221 |
|
| 564 |
1 |
| croak("Data::NDArray::Shared->set: array is frozen (read-only)"); } |
| 565 |
220 |
| uint32_t ndim = h->hdr->ndim; |
| 566 |
220 |
▶ | if (nix != ndim) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (nix != ndim) { nda_rwlock_wrunlock(h); |
1 |
219 |
220 |
|
| 567 |
1 |
| croak("Data::NDArray::Shared->set: expected %u indices + value, got %u args", ndim, nix); } |
| 568 |
865 |
▶ | for (uint32_t d = 0; d < ndim; d++) |
Branch: 1 branch
| Branch | True | False | Total |
| for (uint32_t d = 0; d < ndim; d++) |
647 |
218 |
865 |
|
| 569 |
647 |
▶ | if (idx[d] >= h->hdr->shape[d]) { uint64_t sz = h->hdr->shape[d]; nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (idx[d] >= h->hdr->shape[d]) { uint64_t sz = h->hdr->shape[d]; nda_rwlock_wrunlock(h); |
1 |
646 |
647 |
|
| 570 |
1 |
| croak("Data::NDArray::Shared->set: index %u = %" UVuf " out of range (dim size %" UVuf ")", |
| 571 |
|
| d, (UV)idx[d], (UV)sz); } |
| 572 |
218 |
| uint64_t off = nda_flat_offset(h, idx, ndim); |
| 573 |
|
| /* Layer B: bound the strides-derived flat offset against the element |
| 574 |
|
| count before writing -- corrupted strides in the shared segment must |
| 575 |
|
| not drive an OOB write (see get; never fires for canonical data). */ |
| 576 |
218 |
▶ | if (off >= h->size) { nda_rwlock_wrunlock(h); /* cached immutable count, not peer hdr->size */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (off >= h->size) { nda_rwlock_wrunlock(h); /* cached immutable count, not peer hdr->size */ |
0 |
218 |
218 |
|
| 577 |
0 |
| croak("Data::NDArray::Shared->set: computed offset out of range"); } |
| 578 |
218 |
| nda_store_bytes(h, off, enc, nb); |
| 579 |
218 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 580 |
218 |
| nda_rwlock_wrunlock(h); |
| 581 |
|
| } |
| 582 |
|
| |
| 583 |
|
| SV * |
| 584 |
|
| get_flat(self, e) |
| 585 |
|
| SV *self |
| 586 |
|
| UV e |
| 587 |
|
| PREINIT: |
| 588 |
913 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
913 |
0 |
913 |
| EXTRACT(self); |
0 |
913 |
913 |
| EXTRACT(self); |
0 |
913 |
913 |
|
| 589 |
|
| CODE: |
| 590 |
913 |
▶ | if (e >= h->size) /* cached immutable count, not peer hdr->size */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (e >= h->size) /* cached immutable count, not peer hdr->size */ |
2 |
911 |
913 |
|
| 591 |
2 |
| croak("Data::NDArray::Shared->get_flat: index %" UVuf " out of range (size %" UVuf ")", |
| 592 |
|
| e, (UV)h->size); |
| 593 |
|
| { |
| 594 |
|
| char _raw[8]; |
| 595 |
911 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
14 |
897 |
911 |
|
| 596 |
14 |
| nda_read_raw(h, (uint64_t)e, _raw); |
| 597 |
|
| } else { |
| 598 |
897 |
| nda_rwlock_rdlock(h); |
| 599 |
897 |
| nda_read_raw(h, (uint64_t)e, _raw); |
| 600 |
897 |
| nda_rwlock_rdunlock(h); |
| 601 |
|
| } |
| 602 |
911 |
| RETVAL = nda_sv_from_raw(aTHX_ h->dtype, _raw); |
| 603 |
|
| } |
| 604 |
|
| OUTPUT: |
| 605 |
|
| RETVAL |
| 606 |
|
| |
| 607 |
|
| void |
| 608 |
|
| set_flat(self, e, val) |
| 609 |
|
| SV *self |
| 610 |
|
| UV e |
| 611 |
|
| SV *val |
| 612 |
|
| PREINIT: |
| 613 |
219 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
219 |
0 |
219 |
| EXTRACT(self); |
0 |
219 |
219 |
| EXTRACT(self); |
0 |
219 |
219 |
|
| 614 |
|
| CODE: |
| 615 |
219 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->set_flat: array is frozen (read-only)"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->set_flat: array is frozen (read-only)"); |
2 |
217 |
219 |
|
| 616 |
217 |
▶ | if (e >= h->size) /* cached immutable count, not peer hdr->size */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (e >= h->size) /* cached immutable count, not peer hdr->size */ |
1 |
216 |
217 |
|
| 617 |
1 |
| croak("Data::NDArray::Shared->set_flat: index %" UVuf " out of range (size %" UVuf ")", |
| 618 |
|
| e, (UV)h->size); |
| 619 |
216 |
| char enc[8]; uint32_t nb = nda_encode(aTHX_ h->dtype, val, enc); /* encode before locking */ |
| 620 |
216 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
216 |
216 |
| REEXTRACT(self); |
0 |
216 |
216 |
|
| 621 |
216 |
| nda_rwlock_wrlock(h); |
| 622 |
216 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
0 |
216 |
216 |
|
| 623 |
0 |
| croak("Data::NDArray::Shared->set_flat: array is frozen (read-only)"); } |
| 624 |
216 |
| nda_store_bytes(h, (uint64_t)e, enc, nb); |
| 625 |
216 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 626 |
216 |
| nda_rwlock_wrunlock(h); |
| 627 |
|
| |
| 628 |
|
| SV * |
| 629 |
|
| fill(self, val) |
| 630 |
|
| SV *self |
| 631 |
|
| SV *val |
| 632 |
|
| PREINIT: |
| 633 |
45 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
45 |
0 |
45 |
| EXTRACT(self); |
0 |
45 |
45 |
| EXTRACT(self); |
0 |
45 |
45 |
|
| 634 |
|
| CODE: |
| 635 |
45 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->fill: array is frozen (read-only)"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->fill: array is frozen (read-only)"); |
2 |
43 |
45 |
|
| 636 |
43 |
| char enc[8]; uint32_t nb = nda_encode(aTHX_ h->dtype, val, enc); /* encode before locking */ |
| 637 |
43 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
43 |
43 |
| REEXTRACT(self); |
0 |
43 |
43 |
|
| 638 |
43 |
| nda_rwlock_wrlock(h); |
| 639 |
43 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
0 |
43 |
43 |
|
| 640 |
0 |
| croak("Data::NDArray::Shared->fill: array is frozen (read-only)"); } |
| 641 |
43 |
| nda_fill_bytes(h, enc, nb); |
| 642 |
43 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 643 |
43 |
| nda_rwlock_wrunlock(h); |
| 644 |
43 |
| SvREFCNT_inc(self); |
| 645 |
43 |
| RETVAL = self; |
| 646 |
|
| OUTPUT: |
| 647 |
|
| RETVAL |
| 648 |
|
| |
| 649 |
|
| SV * |
| 650 |
|
| zero(self) |
| 651 |
|
| SV *self |
| 652 |
|
| PREINIT: |
| 653 |
3 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
3 |
0 |
3 |
| EXTRACT(self); |
0 |
3 |
3 |
| EXTRACT(self); |
0 |
3 |
3 |
|
| 654 |
|
| CODE: |
| 655 |
3 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->zero: array is frozen (read-only)"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->zero: array is frozen (read-only)"); |
2 |
1 |
3 |
|
| 656 |
1 |
| nda_rwlock_wrlock(h); |
| 657 |
1 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
0 |
1 |
1 |
|
| 658 |
0 |
| croak("Data::NDArray::Shared->zero: array is frozen (read-only)"); } |
| 659 |
1 |
| memset(nda_data(h), 0, (size_t)(h->size * h->itemsize)); /* cached immutable geometry, not peer hdr fields */ |
| 660 |
1 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 661 |
1 |
| nda_rwlock_wrunlock(h); |
| 662 |
1 |
| SvREFCNT_inc(self); |
| 663 |
1 |
| RETVAL = self; |
| 664 |
|
| OUTPUT: |
| 665 |
|
| RETVAL |
| 666 |
|
| |
| 667 |
|
| SV * |
| 668 |
|
| reshape(self, ...) |
| 669 |
|
| SV *self |
| 670 |
|
| PREINIT: |
| 671 |
9 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
9 |
0 |
9 |
| EXTRACT(self); |
0 |
9 |
9 |
| EXTRACT(self); |
0 |
9 |
9 |
|
| 672 |
|
| uint64_t dims[NDA_MAX_DIMS]; |
| 673 |
|
| uint64_t strides[NDA_MAX_DIMS]; |
| 674 |
|
| CODE: |
| 675 |
|
| { |
| 676 |
9 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->reshape: array is frozen (read-only)"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->reshape: array is frozen (read-only)"); |
2 |
7 |
9 |
|
| 677 |
7 |
| I32 nd = items - 1; |
| 678 |
7 |
▶ | if (nd < 1) croak("Data::NDArray::Shared->reshape: at least one dimension required"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (nd < 1) croak("Data::NDArray::Shared->reshape: at least one dimension required"); |
0 |
7 |
7 |
|
| 679 |
7 |
▶ | if (nd > NDA_MAX_DIMS) croak("Data::NDArray::Shared->reshape: too many dimensions (max %d)", NDA_MAX_DIMS); |
Branch: 1 branch
| Branch | True | False | Total |
| if (nd > NDA_MAX_DIMS) croak("Data::NDArray::Shared->reshape: too many dimensions (max %d)", NDA_MAX_DIMS); |
1 |
6 |
7 |
|
| 680 |
6 |
| uint64_t newsize = 1; |
| 681 |
14 |
▶ | for (I32 i = 0; i < nd; i++) { |
Branch: 1 branch
| Branch | True | False | Total |
| for (I32 i = 0; i < nd; i++) { |
10 |
4 |
14 |
|
| 682 |
10 |
| IV iv = SvIV(ST(1 + i)); |
| 683 |
10 |
▶ | if (iv < 1) croak("Data::NDArray::Shared->reshape: dimension %d must be >= 1", (int)i); |
Branch: 1 branch
| Branch | True | False | Total |
| if (iv < 1) croak("Data::NDArray::Shared->reshape: dimension %d must be >= 1", (int)i); |
2 |
8 |
10 |
|
| 684 |
8 |
| dims[i] = (uint64_t)iv; |
| 685 |
8 |
▶ | if (dims[i] > UINT64_MAX / newsize) croak("Data::NDArray::Shared->reshape: shape too large"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (dims[i] > UINT64_MAX / newsize) croak("Data::NDArray::Shared->reshape: shape too large"); |
0 |
8 |
8 |
|
| 686 |
8 |
| newsize *= dims[i]; |
| 687 |
|
| } |
| 688 |
4 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
4 |
4 |
| REEXTRACT(self); |
0 |
4 |
4 |
|
| 689 |
4 |
▶ | if (newsize != h->size) /* cached immutable count: reshape must preserve element count */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (newsize != h->size) /* cached immutable count: reshape must preserve element count */ |
1 |
3 |
4 |
|
| 690 |
1 |
| croak("Data::NDArray::Shared->reshape: total size %" UVuf " does not match current size %" UVuf, |
| 691 |
|
| (UV)newsize, (UV)h->size); |
| 692 |
|
| /* row-major strides for the new shape */ |
| 693 |
3 |
| strides[nd - 1] = 1; |
| 694 |
5 |
▶ | for (int d = nd - 2; d >= 0; d--) strides[d] = strides[d + 1] * dims[d + 1]; |
Branch: 1 branch
| Branch | True | False | Total |
| for (int d = nd - 2; d >= 0; d--) strides[d] = strides[d + 1] * dims[d + 1]; |
2 |
3 |
5 |
|
| 695 |
3 |
| nda_rwlock_wrlock(h); |
| 696 |
3 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
0 |
3 |
3 |
|
| 697 |
0 |
| croak("Data::NDArray::Shared->reshape: array is frozen (read-only)"); } |
| 698 |
3 |
| h->hdr->ndim = (uint32_t)nd; |
| 699 |
8 |
▶ | for (I32 i = 0; i < nd; i++) { h->hdr->shape[i] = dims[i]; h->hdr->strides[i] = strides[i]; } |
Branch: 1 branch
| Branch | True | False | Total |
| for (I32 i = 0; i < nd; i++) { h->hdr->shape[i] = dims[i]; h->hdr->strides[i] = strides[i]; } |
5 |
3 |
8 |
|
| 700 |
3 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 701 |
3 |
| nda_rwlock_wrunlock(h); |
| 702 |
3 |
| SvREFCNT_inc(self); |
| 703 |
3 |
| RETVAL = self; |
| 704 |
|
| } |
| 705 |
|
| OUTPUT: |
| 706 |
|
| RETVAL |
| 707 |
|
| |
| 708 |
|
| NV |
| 709 |
|
| sum(self) |
| 710 |
|
| SV *self |
| 711 |
|
| PREINIT: |
| 712 |
12 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
12 |
0 |
12 |
| EXTRACT(self); |
0 |
12 |
12 |
| EXTRACT(self); |
0 |
12 |
12 |
|
| 713 |
|
| CODE: |
| 714 |
12 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
4 |
8 |
12 |
|
| 715 |
4 |
| RETVAL = nda_sum_locked(h); /* frozen: immutable, no lock needed */ |
| 716 |
|
| } else { |
| 717 |
8 |
| nda_rwlock_rdlock(h); |
| 718 |
8 |
| RETVAL = nda_sum_locked(h); |
| 719 |
8 |
| nda_rwlock_rdunlock(h); |
| 720 |
|
| } |
| 721 |
|
| OUTPUT: |
| 722 |
|
| RETVAL |
| 723 |
|
| |
| 724 |
|
| NV |
| 725 |
|
| mean(self) |
| 726 |
|
| SV *self |
| 727 |
|
| PREINIT: |
| 728 |
6 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
6 |
0 |
6 |
| EXTRACT(self); |
0 |
6 |
6 |
| EXTRACT(self); |
0 |
6 |
6 |
|
| 729 |
|
| CODE: |
| 730 |
|
| { |
| 731 |
|
| double acc; |
| 732 |
6 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
1 |
5 |
6 |
|
| 733 |
1 |
| acc = nda_sum_locked(h); /* frozen: immutable, no lock needed */ |
| 734 |
|
| } else { |
| 735 |
5 |
| nda_rwlock_rdlock(h); |
| 736 |
5 |
| acc = nda_sum_locked(h); |
| 737 |
5 |
| nda_rwlock_rdunlock(h); |
| 738 |
|
| } |
| 739 |
6 |
▶ | RETVAL = acc / (double)h->size; /* cached immutable count (>= 1); matches the summed span */ |
Branch: 1 branch
| Branch | True | False | Total |
| RETVAL = acc / (double)h->size; /* cached immutable count (>= 1); matches the summed span */ |
0 |
6 |
6 |
|
| 740 |
|
| } |
| 741 |
|
| OUTPUT: |
| 742 |
|
| RETVAL |
| 743 |
|
| |
| 744 |
|
| SV * |
| 745 |
|
| min(self) |
| 746 |
|
| SV *self |
| 747 |
|
| PREINIT: |
| 748 |
9 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
9 |
0 |
9 |
| EXTRACT(self); |
0 |
9 |
9 |
| EXTRACT(self); |
0 |
9 |
9 |
|
| 749 |
|
| CODE: |
| 750 |
|
| { |
| 751 |
|
| uint64_t best; char _raw[8]; |
| 752 |
9 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
1 |
8 |
9 |
|
| 753 |
1 |
| best = nda_argextreme_locked(h, 0); /* frozen: immutable, no lock needed */ |
| 754 |
1 |
| nda_read_raw(h, best, _raw); |
| 755 |
|
| } else { |
| 756 |
8 |
| nda_rwlock_rdlock(h); |
| 757 |
8 |
| best = nda_argextreme_locked(h, 0); /* compare in native dtype */ |
| 758 |
8 |
| nda_read_raw(h, best, _raw); /* copy the min element under the lock */ |
| 759 |
8 |
| nda_rwlock_rdunlock(h); |
| 760 |
|
| } |
| 761 |
9 |
| RETVAL = nda_sv_from_raw(aTHX_ h->dtype, _raw); /* build SV after unlock */ |
| 762 |
|
| } |
| 763 |
|
| OUTPUT: |
| 764 |
|
| RETVAL |
| 765 |
|
| |
| 766 |
|
| SV * |
| 767 |
|
| max(self) |
| 768 |
|
| SV *self |
| 769 |
|
| PREINIT: |
| 770 |
9 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
9 |
0 |
9 |
| EXTRACT(self); |
0 |
9 |
9 |
| EXTRACT(self); |
0 |
9 |
9 |
|
| 771 |
|
| CODE: |
| 772 |
|
| { |
| 773 |
|
| uint64_t best; char _raw[8]; |
| 774 |
9 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
1 |
8 |
9 |
|
| 775 |
1 |
| best = nda_argextreme_locked(h, 1); /* frozen: immutable, no lock needed */ |
| 776 |
1 |
| nda_read_raw(h, best, _raw); |
| 777 |
|
| } else { |
| 778 |
8 |
| nda_rwlock_rdlock(h); |
| 779 |
8 |
| best = nda_argextreme_locked(h, 1); /* compare in native dtype */ |
| 780 |
8 |
| nda_read_raw(h, best, _raw); /* copy the max element under the lock */ |
| 781 |
8 |
| nda_rwlock_rdunlock(h); |
| 782 |
|
| } |
| 783 |
9 |
| RETVAL = nda_sv_from_raw(aTHX_ h->dtype, _raw); /* build SV after unlock */ |
| 784 |
|
| } |
| 785 |
|
| OUTPUT: |
| 786 |
|
| RETVAL |
| 787 |
|
| |
| 788 |
|
| SV * |
| 789 |
|
| add_scalar(self, s) |
| 790 |
|
| SV *self |
| 791 |
|
| SV *s |
| 792 |
|
| PREINIT: |
| 793 |
4 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
4 |
0 |
4 |
| EXTRACT(self); |
0 |
4 |
4 |
| EXTRACT(self); |
0 |
4 |
4 |
|
| 794 |
|
| CODE: |
| 795 |
4 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->add_scalar: array is frozen (read-only)"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->add_scalar: array is frozen (read-only)"); |
2 |
2 |
4 |
|
| 796 |
2 |
| double sd=0; int64_t si=0; uint64_t su=0; |
| 797 |
2 |
| nda_read_scalar(aTHX_ h->dtype, s, &sd, &si, &su); /* pre-read lock-free */ |
| 798 |
2 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
2 |
2 |
| REEXTRACT(self); |
0 |
2 |
2 |
|
| 799 |
2 |
| nda_rwlock_wrlock(h); |
| 800 |
2 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
0 |
2 |
2 |
|
| 801 |
0 |
| croak("Data::NDArray::Shared->add_scalar: array is frozen (read-only)"); } |
| 802 |
2 |
| nda_scalar_op_locked(h, sd, si, su, '+'); |
| 803 |
2 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 804 |
2 |
| nda_rwlock_wrunlock(h); |
| 805 |
2 |
| SvREFCNT_inc(self); |
| 806 |
2 |
| RETVAL = self; |
| 807 |
|
| OUTPUT: |
| 808 |
|
| RETVAL |
| 809 |
|
| |
| 810 |
|
| SV * |
| 811 |
|
| mul_scalar(self, s) |
| 812 |
|
| SV *self |
| 813 |
|
| SV *s |
| 814 |
|
| PREINIT: |
| 815 |
7 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
7 |
0 |
7 |
| EXTRACT(self); |
0 |
7 |
7 |
| EXTRACT(self); |
0 |
7 |
7 |
|
| 816 |
|
| CODE: |
| 817 |
7 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->mul_scalar: array is frozen (read-only)"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->mul_scalar: array is frozen (read-only)"); |
2 |
5 |
7 |
|
| 818 |
5 |
| double sd=0; int64_t si=0; uint64_t su=0; |
| 819 |
5 |
| nda_read_scalar(aTHX_ h->dtype, s, &sd, &si, &su); /* pre-read lock-free */ |
| 820 |
5 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
5 |
5 |
| REEXTRACT(self); |
0 |
5 |
5 |
|
| 821 |
5 |
| nda_rwlock_wrlock(h); |
| 822 |
5 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
0 |
5 |
5 |
|
| 823 |
0 |
| croak("Data::NDArray::Shared->mul_scalar: array is frozen (read-only)"); } |
| 824 |
5 |
| nda_scalar_op_locked(h, sd, si, su, '*'); |
| 825 |
5 |
| __atomic_fetch_add(&h->hdr->stat_ops, 1, __ATOMIC_RELAXED); |
| 826 |
5 |
| nda_rwlock_wrunlock(h); |
| 827 |
5 |
| SvREFCNT_inc(self); |
| 828 |
5 |
| RETVAL = self; |
| 829 |
|
| OUTPUT: |
| 830 |
|
| RETVAL |
| 831 |
|
| |
| 832 |
|
| SV * |
| 833 |
|
| add(self, other) |
| 834 |
|
| SV *self |
| 835 |
|
| SV *other |
| 836 |
|
| PREINIT: |
| 837 |
15 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
15 |
0 |
15 |
| EXTRACT(self); |
0 |
15 |
15 |
| EXTRACT(self); |
0 |
15 |
15 |
|
| 838 |
|
| CODE: |
| 839 |
15 |
| nda_do_elementwise(aTHX_ self, h, other, '+', "Data::NDArray::Shared->add"); |
| 840 |
10 |
| SvREFCNT_inc(self); |
| 841 |
10 |
| RETVAL = self; |
| 842 |
|
| OUTPUT: |
| 843 |
|
| RETVAL |
| 844 |
|
| |
| 845 |
|
| SV * |
| 846 |
|
| subtract(self, other) |
| 847 |
|
| SV *self |
| 848 |
|
| SV *other |
| 849 |
|
| PREINIT: |
| 850 |
12 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
12 |
0 |
12 |
| EXTRACT(self); |
0 |
12 |
12 |
| EXTRACT(self); |
0 |
12 |
12 |
|
| 851 |
|
| CODE: |
| 852 |
12 |
| nda_do_elementwise(aTHX_ self, h, other, '-', "Data::NDArray::Shared->subtract"); |
| 853 |
8 |
| SvREFCNT_inc(self); |
| 854 |
8 |
| RETVAL = self; |
| 855 |
|
| OUTPUT: |
| 856 |
|
| RETVAL |
| 857 |
|
| |
| 858 |
|
| SV * |
| 859 |
|
| multiply(self, other) |
| 860 |
|
| SV *self |
| 861 |
|
| SV *other |
| 862 |
|
| PREINIT: |
| 863 |
13 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
13 |
0 |
13 |
| EXTRACT(self); |
0 |
13 |
13 |
| EXTRACT(self); |
0 |
13 |
13 |
|
| 864 |
|
| CODE: |
| 865 |
13 |
| nda_do_elementwise(aTHX_ self, h, other, '*', "Data::NDArray::Shared->multiply"); |
| 866 |
9 |
| SvREFCNT_inc(self); |
| 867 |
9 |
| RETVAL = self; |
| 868 |
|
| OUTPUT: |
| 869 |
|
| RETVAL |
| 870 |
|
| |
| 871 |
|
| SV * |
| 872 |
|
| to_list(self) |
| 873 |
|
| SV *self |
| 874 |
|
| PREINIT: |
| 875 |
59 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
59 |
0 |
59 |
| EXTRACT(self); |
0 |
59 |
59 |
| EXTRACT(self); |
0 |
59 |
59 |
|
| 876 |
|
| CODE: |
| 877 |
|
| { |
| 878 |
59 |
| uint64_t size = h->size, e; /* cached immutable count, not peer hdr->size */ |
| 879 |
59 |
| int dt = h->dtype; |
| 880 |
59 |
| uint32_t nb = nda_dtype_bytes(dt); |
| 881 |
59 |
| char *snap = NULL; |
| 882 |
59 |
| AV *av = newAV(); |
| 883 |
59 |
▶ | if (size) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (size) { |
59 |
0 |
59 |
|
| 884 |
59 |
| av_extend(av, (SSize_t)(size - 1)); /* pre-extend BEFORE the lock */ |
| 885 |
59 |
| Newx(snap, size * (STRLEN)nb, char); |
| 886 |
59 |
| SAVEFREEPV(snap); /* freed on scope exit / croak */ |
| 887 |
|
| } |
| 888 |
59 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
2 |
57 |
59 |
|
| 889 |
2 |
▶ | if (size) memcpy(snap, nda_data(h), (size_t)(size * nb)); /* frozen: immutable, no lock needed */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (size) memcpy(snap, nda_data(h), (size_t)(size * nb)); /* frozen: immutable, no lock needed */ |
2 |
0 |
2 |
|
| 890 |
|
| } else { |
| 891 |
57 |
| nda_rwlock_rdlock(h); |
| 892 |
57 |
▶ | if (size) memcpy(snap, nda_data(h), (size_t)(size * nb)); /* snapshot raw bytes under the lock */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (size) memcpy(snap, nda_data(h), (size_t)(size * nb)); /* snapshot raw bytes under the lock */ |
57 |
0 |
57 |
|
| 893 |
57 |
| nda_rwlock_rdunlock(h); |
| 894 |
|
| } |
| 895 |
2928 |
▶ | for (e = 0; e < size; e++) /* build the SVs AFTER unlocking */ |
Branch: 1 branch
| Branch | True | False | Total |
| for (e = 0; e < size; e++) /* build the SVs AFTER unlocking */ |
2869 |
59 |
2928 |
|
| 896 |
2869 |
| av_store(av, (SSize_t)e, nda_sv_from_raw(aTHX_ dt, snap + e * nb)); |
| 897 |
59 |
| RETVAL = newRV_noinc((SV *)av); |
| 898 |
|
| } |
| 899 |
|
| OUTPUT: |
| 900 |
|
| RETVAL |
| 901 |
|
| |
| 902 |
|
| void |
| 903 |
|
| shape(self) |
| 904 |
|
| SV *self |
| 905 |
|
| PREINIT: |
| 906 |
30 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
30 |
0 |
30 |
| EXTRACT(self); |
0 |
30 |
30 |
| EXTRACT(self); |
0 |
30 |
30 |
|
| 907 |
|
| PPCODE: |
| 908 |
|
| { |
| 909 |
|
| uint64_t snap[NDA_MAX_DIMS]; uint32_t ndim, d; |
| 910 |
|
| /* Snapshot ndim+shape under the read lock: reshape mutates them under |
| 911 |
|
| * the write lock, so an unlocked read could tear (a new ndim with a |
| 912 |
|
| * stale shape[]). Copy out, unlock, then build the SVs. Frozen: reshape |
| 913 |
|
| * is refused, so ndim/shape can never change -- read lock-free (the |
| 914 |
|
| * rwlock would WRITE into the PROT_READ header and SIGSEGV). */ |
| 915 |
30 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
3 |
27 |
30 |
|
| 916 |
3 |
| ndim = h->hdr->ndim; |
| 917 |
3 |
▶ | if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
0 |
3 |
3 |
|
| 918 |
9 |
▶ | for (d = 0; d < ndim; d++) snap[d] = (uint64_t)h->hdr->shape[d]; |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) snap[d] = (uint64_t)h->hdr->shape[d]; |
6 |
3 |
9 |
|
| 919 |
|
| } else { |
| 920 |
27 |
| nda_rwlock_rdlock(h); |
| 921 |
27 |
| ndim = h->hdr->ndim; |
| 922 |
27 |
▶ | if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
0 |
27 |
27 |
|
| 923 |
64 |
▶ | for (d = 0; d < ndim; d++) snap[d] = (uint64_t)h->hdr->shape[d]; |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) snap[d] = (uint64_t)h->hdr->shape[d]; |
37 |
27 |
64 |
|
| 924 |
27 |
| nda_rwlock_rdunlock(h); |
| 925 |
|
| } |
| 926 |
30 |
▶ | EXTEND(SP, (SSize_t)ndim); |
Branch: 1 branch
| Branch | True | False | Total |
| EXTEND(SP, (SSize_t)ndim); |
0 |
30 |
30 |
|
| 927 |
73 |
▶ | for (d = 0; d < ndim; d++) |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) |
43 |
30 |
73 |
|
| 928 |
43 |
| PUSHs(sv_2mortal(newSVuv((UV)snap[d]))); |
| 929 |
|
| } |
| 930 |
|
| |
| 931 |
|
| void |
| 932 |
|
| strides(self) |
| 933 |
|
| SV *self |
| 934 |
|
| PREINIT: |
| 935 |
6 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
6 |
0 |
6 |
| EXTRACT(self); |
0 |
6 |
6 |
| EXTRACT(self); |
0 |
6 |
6 |
|
| 936 |
|
| PPCODE: |
| 937 |
|
| { |
| 938 |
|
| uint64_t snap[NDA_MAX_DIMS]; uint32_t ndim, d; |
| 939 |
|
| /* Snapshot ndim+strides under the read lock (see shape): reshape mutates |
| 940 |
|
| * them under the write lock; an unlocked read could tear. Frozen: lock-free |
| 941 |
|
| * (see shape). */ |
| 942 |
6 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
1 |
5 |
6 |
|
| 943 |
1 |
| ndim = h->hdr->ndim; |
| 944 |
1 |
▶ | if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
0 |
1 |
1 |
|
| 945 |
3 |
▶ | for (d = 0; d < ndim; d++) snap[d] = (uint64_t)h->hdr->strides[d]; |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) snap[d] = (uint64_t)h->hdr->strides[d]; |
2 |
1 |
3 |
|
| 946 |
|
| } else { |
| 947 |
5 |
| nda_rwlock_rdlock(h); |
| 948 |
5 |
| ndim = h->hdr->ndim; |
| 949 |
5 |
▶ | if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
0 |
5 |
5 |
|
| 950 |
15 |
▶ | for (d = 0; d < ndim; d++) snap[d] = (uint64_t)h->hdr->strides[d]; |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) snap[d] = (uint64_t)h->hdr->strides[d]; |
10 |
5 |
15 |
|
| 951 |
5 |
| nda_rwlock_rdunlock(h); |
| 952 |
|
| } |
| 953 |
6 |
▶ | EXTEND(SP, (SSize_t)ndim); |
Branch: 1 branch
| Branch | True | False | Total |
| EXTEND(SP, (SSize_t)ndim); |
0 |
6 |
6 |
|
| 954 |
18 |
▶ | for (d = 0; d < ndim; d++) |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) |
12 |
6 |
18 |
|
| 955 |
12 |
| PUSHs(sv_2mortal(newSVuv((UV)snap[d]))); |
| 956 |
|
| } |
| 957 |
|
| |
| 958 |
|
| UV |
| 959 |
|
| ndim(self) |
| 960 |
|
| SV *self |
| 961 |
|
| PREINIT: |
| 962 |
4 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
4 |
0 |
4 |
| EXTRACT(self); |
0 |
4 |
4 |
| EXTRACT(self); |
0 |
4 |
4 |
|
| 963 |
|
| CODE: |
| 964 |
4 |
▶ | RETVAL = (UV)h->hdr->ndim; /* current rank (1..8); reshape() can change it */ |
Branch: 1 branch
| Branch | True | False | Total |
| RETVAL = (UV)h->hdr->ndim; /* current rank (1..8); reshape() can change it */ |
0 |
4 |
4 |
|
| 965 |
|
| OUTPUT: |
| 966 |
|
| RETVAL |
| 967 |
|
| |
| 968 |
|
| UV |
| 969 |
|
| size(self) |
| 970 |
|
| SV *self |
| 971 |
|
| PREINIT: |
| 972 |
10 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
10 |
0 |
10 |
| EXTRACT(self); |
0 |
10 |
10 |
| EXTRACT(self); |
0 |
10 |
10 |
|
| 973 |
|
| CODE: |
| 974 |
10 |
▶ | RETVAL = (UV)h->hdr->size; /* immutable after creation -- lock-free */ |
Branch: 1 branch
| Branch | True | False | Total |
| RETVAL = (UV)h->hdr->size; /* immutable after creation -- lock-free */ |
0 |
10 |
10 |
|
| 975 |
|
| OUTPUT: |
| 976 |
|
| RETVAL |
| 977 |
|
| |
| 978 |
|
| UV |
| 979 |
|
| itemsize(self) |
| 980 |
|
| SV *self |
| 981 |
|
| PREINIT: |
| 982 |
5 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
5 |
0 |
5 |
| EXTRACT(self); |
0 |
5 |
5 |
| EXTRACT(self); |
0 |
5 |
5 |
|
| 983 |
|
| CODE: |
| 984 |
5 |
▶ | RETVAL = (UV)h->hdr->itemsize; /* immutable after creation -- lock-free */ |
Branch: 1 branch
| Branch | True | False | Total |
| RETVAL = (UV)h->hdr->itemsize; /* immutable after creation -- lock-free */ |
0 |
5 |
5 |
|
| 985 |
|
| OUTPUT: |
| 986 |
|
| RETVAL |
| 987 |
|
| |
| 988 |
|
| # Raw contiguous data region as a byte string (read-locked snapshot copy). The |
| 989 |
|
| # bytes are row-major C-order; pair with shape()/dtype() to interpret them. |
| 990 |
|
| SV * |
| 991 |
|
| buffer(self) |
| 992 |
|
| SV *self |
| 993 |
|
| PREINIT: |
| 994 |
22 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
22 |
0 |
22 |
| EXTRACT(self); |
0 |
22 |
22 |
| EXTRACT(self); |
0 |
22 |
22 |
|
| 995 |
|
| CODE: |
| 996 |
|
| { |
| 997 |
22 |
| uint64_t bytes = h->size * h->itemsize; /* cached immutable geometry, not peer hdr fields */ |
| 998 |
22 |
| char *base = nda_data(h); |
| 999 |
22 |
| RETVAL = newSVpvn("", 0); |
| 1000 |
22 |
▶ | (void)SvGROW(RETVAL, (STRLEN)(bytes + 1)); /* size the buffer BEFORE the lock */ |
Branch: 2 branches
| Branch | True | False | Total |
| (void)SvGROW(RETVAL, (STRLEN)(bytes + 1)); /* size the buffer BEFORE the lock */ |
22 |
0 |
22 |
| (void)SvGROW(RETVAL, (STRLEN)(bytes + 1)); /* size the buffer BEFORE the lock */ |
11 |
11 |
22 |
|
| 1001 |
22 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
3 |
19 |
22 |
|
| 1002 |
3 |
| Copy(base, SvPVX(RETVAL), bytes, char); /* frozen: immutable, no lock needed */ |
| 1003 |
|
| } else { |
| 1004 |
19 |
| nda_rwlock_rdlock(h); |
| 1005 |
19 |
| Copy(base, SvPVX(RETVAL), bytes, char); |
| 1006 |
19 |
| nda_rwlock_rdunlock(h); |
| 1007 |
|
| } |
| 1008 |
22 |
| SvCUR_set(RETVAL, (STRLEN)bytes); |
| 1009 |
22 |
| *SvEND(RETVAL) = '\0'; |
| 1010 |
|
| } |
| 1011 |
|
| OUTPUT: |
| 1012 |
|
| RETVAL |
| 1013 |
|
| |
| 1014 |
|
| # Overwrite the whole data region from a byte string of exactly size*itemsize |
| 1015 |
|
| # bytes (write-locked). Used by from_pdl/update_from_pdl. |
| 1016 |
|
| void |
| 1017 |
|
| update_from_bytes(self, src) |
| 1018 |
|
| SV *self |
| 1019 |
|
| SV *src |
| 1020 |
|
| PREINIT: |
| 1021 |
20 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
20 |
0 |
20 |
| EXTRACT(self); |
0 |
20 |
20 |
| EXTRACT(self); |
0 |
20 |
20 |
|
| 1022 |
|
| CODE: |
| 1023 |
|
| { |
| 1024 |
20 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->update_from_bytes: array is frozen (read-only)"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->update_from_bytes: array is frozen (read-only)"); |
2 |
18 |
20 |
|
| 1025 |
|
| STRLEN slen; |
| 1026 |
18 |
| const char *sbytes = SvPVbyte(src, slen); /* resolve + any croak BEFORE the lock */ |
| 1027 |
18 |
▶ | REEXTRACT(self); |
Branch: 2 branches
| Branch | True | False | Total |
| REEXTRACT(self); |
0 |
18 |
18 |
| REEXTRACT(self); |
0 |
18 |
18 |
|
| 1028 |
18 |
| uint64_t bytes = h->size * h->itemsize; /* cached immutable geometry, not peer hdr fields */ |
| 1029 |
|
| char *base; |
| 1030 |
18 |
▶ | if ((uint64_t)slen != bytes) |
Branch: 1 branch
| Branch | True | False | Total |
| if ((uint64_t)slen != bytes) |
1 |
17 |
18 |
|
| 1031 |
1 |
| croak("Data::NDArray::Shared->update_from_bytes: got %" UVuf |
| 1032 |
|
| " bytes, expected %" UVuf, (UV)slen, (UV)bytes); |
| 1033 |
17 |
| base = nda_data(h); |
| 1034 |
17 |
| nda_rwlock_wrlock(h); |
| 1035 |
17 |
▶ | if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->hdr->sealed) { nda_rwlock_wrunlock(h); |
0 |
17 |
17 |
|
| 1036 |
0 |
| croak("Data::NDArray::Shared->update_from_bytes: array is frozen (read-only)"); } |
| 1037 |
17 |
| Copy(sbytes, base, bytes, char); |
| 1038 |
17 |
| nda_rwlock_wrunlock(h); |
| 1039 |
|
| } |
| 1040 |
|
| |
| 1041 |
|
| # Zero-copy: build a PDL ndarray whose data IS our shared mmap, via PDL's C API |
| 1042 |
|
| # (PDL_DONTTOUCHDATA -- PDL never frees or reallocates the mapping, the only way to |
| 1043 |
|
| # alias external memory PDL won't silently detach from). The PDL C path is compiled |
| 1044 |
|
| # only when the module was built against PDL; otherwise this croaks. Returns SV* |
| 1045 |
|
| # (not pdl*) so the XSUB needs no PDL typemap and the no-PDL build still compiles. |
| 1046 |
|
| # `datatype` is the PDL type number; `dims_av` is in PDL order (fastest axis first). |
| 1047 |
|
| SV * |
| 1048 |
|
| _alias_pdl_create(self, datatype, dims_av) |
| 1049 |
|
| SV *self |
| 1050 |
|
| int datatype |
| 1051 |
|
| AV *dims_av |
| 1052 |
|
| PREINIT: |
| 1053 |
3 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
3 |
0 |
3 |
| EXTRACT(self); |
0 |
3 |
3 |
| EXTRACT(self); |
0 |
3 |
3 |
|
| 1054 |
|
| CODE: |
| 1055 |
|
| #ifdef HAVE_PDL |
| 1056 |
|
| { |
| 1057 |
|
| /* PDL's readonly flag does not cover every write path. */ |
| 1058 |
|
| if (h->readonly || h->hdr->sealed) |
| 1059 |
|
| croak("Data::NDArray::Shared->as_pdl_alias: array is frozen (read-only); " |
| 1060 |
|
| "use to_pdl for a safe copy instead"); |
| 1061 |
|
| IV nd = av_len(dims_av) + 1, i; |
| 1062 |
|
| PDL_Indx dims[NDA_MAX_DIMS]; |
| 1063 |
|
| pdl_error err; |
| 1064 |
|
| pdl *p; |
| 1065 |
|
| nda_boot_pdl(aTHX); |
| 1066 |
|
| if (nd < 1 || nd > NDA_MAX_DIMS) |
| 1067 |
|
| croak("Data::NDArray::Shared: bad dim count %" IVdf, nd); |
| 1068 |
|
| for (i = 0; i < nd; i++) { |
| 1069 |
|
| SV **e = av_fetch(dims_av, (SSize_t)i, 0); |
| 1070 |
|
| dims[i] = (PDL_Indx)(e ? SvIV(*e) : 0); |
| 1071 |
|
| } |
| 1072 |
|
| p = PDL->pdlnew(); |
| 1073 |
|
| if (!p) croak("Data::NDArray::Shared: PDL->pdlnew failed"); |
| 1074 |
|
| p->datatype = (pdl_datatypes)datatype; /* set type before setdims so nbytes is right */ |
| 1075 |
|
| err = PDL->setdims(p, dims, (PDL_Indx)nd); |
| 1076 |
|
| if (err.error) { PDL->destroy(p); croak("Data::NDArray::Shared: PDL->setdims failed"); } |
| 1077 |
|
| REEXTRACT(self); |
| 1078 |
|
| if (h->readonly || h->hdr->sealed) { |
| 1079 |
|
| PDL->destroy(p); |
| 1080 |
|
| croak("Data::NDArray::Shared->as_pdl_alias: array is frozen (read-only); " |
| 1081 |
|
| "use to_pdl for a safe copy instead"); |
| 1082 |
|
| } |
| 1083 |
|
| p->data = nda_data(h); /* alias the shared mmap */ |
| 1084 |
|
| p->state |= PDL_DONTTOUCHDATA | PDL_ALLOCATED; |
| 1085 |
|
| SV *owner = newSVsv(self); |
| 1086 |
|
| err = PDL->add_deletedata_magic(p, nda_pdl_nofree, (Size_t)PTR2UV(owner)); |
| 1087 |
|
| if (err.error) { |
| 1088 |
|
| SvREFCNT_dec(owner); |
| 1089 |
|
| PDL->destroy(p); |
| 1090 |
|
| croak("Data::NDArray::Shared: PDL->add_deletedata_magic failed"); |
| 1091 |
|
| } |
| 1092 |
|
| RETVAL = newSV(0); |
| 1093 |
|
| PDL->SetSV_PDL(RETVAL, p); /* wrap the pdl as its Perl object */ |
| 1094 |
|
| } |
| 1095 |
|
| #else |
| 1096 |
|
| PERL_UNUSED_VAR(datatype); |
| 1097 |
|
| PERL_UNUSED_VAR(dims_av); |
| 1098 |
3 |
| croak("Data::NDArray::Shared: as_pdl_alias needs the module built with PDL " |
| 1099 |
|
| "installed; reinstall with PDL present (to_pdl/update_from_pdl need no rebuild)"); |
| 1100 |
|
| RETVAL = &PL_sv_undef; /* not reached (croak does not return) */ |
| 1101 |
|
| #endif |
| 1102 |
|
| OUTPUT: |
| 1103 |
|
| RETVAL |
| 1104 |
|
| |
| 1105 |
|
| SV * |
| 1106 |
|
| dtype(self) |
| 1107 |
|
| SV *self |
| 1108 |
|
| PREINIT: |
| 1109 |
53 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
53 |
0 |
53 |
| EXTRACT(self); |
0 |
53 |
53 |
| EXTRACT(self); |
0 |
53 |
53 |
|
| 1110 |
|
| CODE: |
| 1111 |
53 |
| RETVAL = newSVpv(nda_name_tab[h->dtype], 0); /* immutable */ |
| 1112 |
|
| OUTPUT: |
| 1113 |
|
| RETVAL |
| 1114 |
|
| |
| 1115 |
|
| void |
| 1116 |
|
| freeze(self) |
| 1117 |
|
| SV *self |
| 1118 |
|
| PREINIT: |
| 1119 |
5 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
5 |
0 |
5 |
| EXTRACT(self); |
0 |
5 |
5 |
| EXTRACT(self); |
0 |
5 |
5 |
|
| 1120 |
|
| CODE: |
| 1121 |
5 |
▶ | if (h->readonly) croak("Data::NDArray::Shared->freeze: cannot freeze a read-only handle"); |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) croak("Data::NDArray::Shared->freeze: cannot freeze a read-only handle"); |
2 |
3 |
5 |
|
| 1122 |
3 |
▶ | if (nda_freeze(h) != 0) croak("Data::NDArray::Shared->freeze: msync: %s", strerror(errno)); |
Branch: 1 branch
| Branch | True | False | Total |
| if (nda_freeze(h) != 0) croak("Data::NDArray::Shared->freeze: msync: %s", strerror(errno)); |
0 |
3 |
3 |
|
| 1123 |
3 |
| h->readonly = 1; /* this handle now rejects mutation too (the file is sealed) */ |
| 1124 |
|
| |
| 1125 |
|
| UV |
| 1126 |
|
| frozen(self) |
| 1127 |
|
| SV *self |
| 1128 |
|
| PREINIT: |
| 1129 |
3 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
3 |
0 |
3 |
| EXTRACT(self); |
0 |
3 |
3 |
| EXTRACT(self); |
0 |
3 |
3 |
|
| 1130 |
|
| CODE: |
| 1131 |
3 |
▶ | RETVAL = h->hdr->sealed ? 1 : 0; |
Branch: 1 branch
| Branch | True | False | Total |
| RETVAL = h->hdr->sealed ? 1 : 0; |
0 |
3 |
3 |
|
| 1132 |
|
| OUTPUT: |
| 1133 |
|
| RETVAL |
| 1134 |
|
| |
| 1135 |
|
| UV |
| 1136 |
|
| readonly(self) |
| 1137 |
|
| SV *self |
| 1138 |
|
| PREINIT: |
| 1139 |
4 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
4 |
0 |
4 |
| EXTRACT(self); |
0 |
4 |
4 |
| EXTRACT(self); |
0 |
4 |
4 |
|
| 1140 |
|
| CODE: |
| 1141 |
4 |
▶ | RETVAL = h->readonly ? 1 : 0; |
Branch: 1 branch
| Branch | True | False | Total |
| RETVAL = h->readonly ? 1 : 0; |
0 |
4 |
4 |
|
| 1142 |
|
| OUTPUT: |
| 1143 |
|
| RETVAL |
| 1144 |
|
| |
| 1145 |
|
| SV * |
| 1146 |
|
| stats(self) |
| 1147 |
|
| SV *self |
| 1148 |
|
| PREINIT: |
| 1149 |
2 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
2 |
0 |
2 |
| EXTRACT(self); |
0 |
2 |
2 |
| EXTRACT(self); |
0 |
2 |
2 |
|
| 1150 |
|
| CODE: |
| 1151 |
|
| { |
| 1152 |
|
| uint32_t dtype, ndim, itemsize, d; |
| 1153 |
|
| uint64_t size, ops, shp[NDA_MAX_DIMS]; |
| 1154 |
|
| uint8_t sealed; |
| 1155 |
|
| /* Snapshot under the lock; build the HV after releasing it so an OOM |
| 1156 |
|
| in newHV/newSV* can never strand the lock. Frozen: immutable, no |
| 1157 |
|
| lock needed (the rwlock would WRITE into the PROT_READ header). */ |
| 1158 |
2 |
▶ | if (h->readonly) { |
Branch: 1 branch
| Branch | True | False | Total |
| if (h->readonly) { |
1 |
1 |
2 |
|
| 1159 |
1 |
| dtype = h->dtype; /* cached + in-range: safe to index nda_name_tab */ |
| 1160 |
1 |
| ndim = h->hdr->ndim; |
| 1161 |
1 |
▶ | if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
0 |
1 |
1 |
|
| 1162 |
1 |
| itemsize = h->hdr->itemsize; |
| 1163 |
1 |
| size = h->hdr->size; |
| 1164 |
1 |
| ops = h->hdr->stat_ops; |
| 1165 |
1 |
| sealed = h->hdr->sealed; |
| 1166 |
3 |
▶ | for (d = 0; d < ndim; d++) shp[d] = h->hdr->shape[d]; |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) shp[d] = h->hdr->shape[d]; |
2 |
1 |
3 |
|
| 1167 |
|
| } else { |
| 1168 |
1 |
| nda_rwlock_rdlock(h); |
| 1169 |
1 |
| dtype = h->dtype; /* cached + in-range: safe to index nda_name_tab */ |
| 1170 |
1 |
| ndim = h->hdr->ndim; |
| 1171 |
1 |
▶ | if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
Branch: 1 branch
| Branch | True | False | Total |
| if (ndim > NDA_MAX_DIMS) ndim = NDA_MAX_DIMS; /* clamp a peer-corrupted ndim */ |
0 |
1 |
1 |
|
| 1172 |
1 |
| itemsize = h->hdr->itemsize; |
| 1173 |
1 |
| size = h->hdr->size; |
| 1174 |
1 |
| ops = h->hdr->stat_ops; |
| 1175 |
1 |
| sealed = h->hdr->sealed; |
| 1176 |
3 |
▶ | for (d = 0; d < ndim; d++) shp[d] = h->hdr->shape[d]; |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) shp[d] = h->hdr->shape[d]; |
2 |
1 |
3 |
|
| 1177 |
1 |
| nda_rwlock_rdunlock(h); |
| 1178 |
|
| } |
| 1179 |
|
| |
| 1180 |
2 |
| AV *shape_av = newAV(); |
| 1181 |
|
| /* Guard the low side: a peer-corrupted ndim==0 would wrap ndim-1 to |
| 1182 |
|
| UINT32_MAX -> a ~34GB av_extend -> uncatchable OOM abort. ndim==0 is |
| 1183 |
|
| handled gracefully (empty shape), matching shape()/strides(). */ |
| 1184 |
2 |
▶ | if (ndim > 0) av_extend(shape_av, (SSize_t)(ndim - 1)); |
Branch: 1 branch
| Branch | True | False | Total |
| if (ndim > 0) av_extend(shape_av, (SSize_t)(ndim - 1)); |
2 |
0 |
2 |
|
| 1185 |
6 |
▶ | for (d = 0; d < ndim; d++) av_store(shape_av, (SSize_t)d, newSVuv((UV)shp[d])); |
Branch: 1 branch
| Branch | True | False | Total |
| for (d = 0; d < ndim; d++) av_store(shape_av, (SSize_t)d, newSVuv((UV)shp[d])); |
4 |
2 |
6 |
|
| 1186 |
|
| |
| 1187 |
2 |
| HV *hv = newHV(); |
| 1188 |
2 |
| hv_stores(hv, "dtype", newSVpv(nda_name_tab[dtype], 0)); |
| 1189 |
2 |
| hv_stores(hv, "ndim", newSVuv((UV)ndim)); |
| 1190 |
2 |
| hv_stores(hv, "size", newSVuv((UV)size)); |
| 1191 |
2 |
| hv_stores(hv, "itemsize", newSVuv((UV)itemsize)); |
| 1192 |
2 |
| hv_stores(hv, "shape", newRV_noinc((SV *)shape_av)); |
| 1193 |
2 |
| hv_stores(hv, "ops", newSVuv((UV)ops)); |
| 1194 |
2 |
| hv_stores(hv, "mmap_size", newSVuv((UV)h->mmap_size)); |
| 1195 |
2 |
| hv_stores(hv, "frozen", newSVuv(sealed ? 1 : 0)); |
| 1196 |
2 |
| hv_stores(hv, "readonly", newSVuv(h->readonly ? 1 : 0)); |
| 1197 |
2 |
| RETVAL = newRV_noinc((SV *)hv); |
| 1198 |
|
| } |
| 1199 |
|
| OUTPUT: |
| 1200 |
|
| RETVAL |
| 1201 |
|
| |
| 1202 |
|
| SV * |
| 1203 |
|
| path(self) |
| 1204 |
|
| SV *self |
| 1205 |
|
| PREINIT: |
| 1206 |
5 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
5 |
0 |
5 |
| EXTRACT(self); |
0 |
5 |
5 |
| EXTRACT(self); |
0 |
5 |
5 |
|
| 1207 |
|
| CODE: |
| 1208 |
5 |
▶ | RETVAL = h->path ? newSVpv(h->path, 0) : &PL_sv_undef; |
Branch: 1 branch
| Branch | True | False | Total |
| RETVAL = h->path ? newSVpv(h->path, 0) : &PL_sv_undef; |
3 |
2 |
5 |
|
| 1209 |
|
| OUTPUT: |
| 1210 |
|
| RETVAL |
| 1211 |
|
| |
| 1212 |
|
| int |
| 1213 |
|
| memfd(self) |
| 1214 |
|
| SV *self |
| 1215 |
|
| PREINIT: |
| 1216 |
7 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
7 |
0 |
7 |
| EXTRACT(self); |
0 |
7 |
7 |
| EXTRACT(self); |
0 |
7 |
7 |
|
| 1217 |
|
| CODE: |
| 1218 |
7 |
▶ | RETVAL = h->backing_fd; |
Branch: 1 branch
| Branch | True | False | Total |
| RETVAL = h->backing_fd; |
0 |
7 |
7 |
|
| 1219 |
|
| OUTPUT: |
| 1220 |
|
| RETVAL |
| 1221 |
|
| |
| 1222 |
|
| void |
| 1223 |
|
| sync(self) |
| 1224 |
|
| SV *self |
| 1225 |
|
| PREINIT: |
| 1226 |
3 |
▶ | EXTRACT(self); |
Branch: 3 branches
| Branch | True | False | Total |
| EXTRACT(self); |
3 |
0 |
3 |
| EXTRACT(self); |
0 |
3 |
3 |
| EXTRACT(self); |
0 |
3 |
3 |
|
| 1227 |
|
| CODE: |
| 1228 |
3 |
▶ | if (!h->readonly && nda_msync(h) != 0) croak("sync: %s", strerror(errno)); |
Branch: 2 branches
| Branch | True | False | Total |
| if (!h->readonly && nda_msync(h) != 0) croak("sync: %s", strerror(errno)); |
2 |
1 |
3 |
| if (!h->readonly && nda_msync(h) != 0) croak("sync: %s", strerror(errno)); |
0 |
2 |
2 |
|
| 1229 |
|
| |
| 1230 |
|
| void |
| 1231 |
|
| unlink(self, ...) |
| 1232 |
|
| SV *self |
| 1233 |
|
| CODE: |
| 1234 |
5 |
▶ | if (sv_isobject(self) && sv_derived_from(self, "Data::NDArray::Shared")) { |
Branch: 2 branches
| Branch | True | False | Total |
| if (sv_isobject(self) && sv_derived_from(self, "Data::NDArray::Shared")) { |
2 |
1 |
3 |
| if (sv_isobject(self) && sv_derived_from(self, "Data::NDArray::Shared")) { |
2 |
0 |
2 |
|
| 1235 |
2 |
| NdaHandle *h = INT2PTR(NdaHandle*, SvIV(SvRV(self))); |
| 1236 |
2 |
▶ | if (h && h->path && unlink(h->path) != 0 && errno != ENOENT) |
Branch: 4 branches
| Branch | True | False | Total |
| if (h && h->path && unlink(h->path) != 0 && errno != ENOENT) |
2 |
0 |
2 |
| if (h && h->path && unlink(h->path) != 0 && errno != ENOENT) |
2 |
0 |
2 |
| if (h && h->path && unlink(h->path) != 0 && errno != ENOENT) |
0 |
2 |
2 |
| if (h && h->path && unlink(h->path) != 0 && errno != ENOENT) |
0 |
0 |
0 |
|
| 1237 |
0 |
| croak("Data::NDArray::Shared->unlink(%s): %s", h->path, strerror(errno)); |
| 1238 |
1 |
▶ | } else if (items >= 2 && (SvGETMAGIC(ST(1)), SvOK(ST(1)))) { |
Branch: 4 branches
| Branch | True | False | Total |
| } else if (items >= 2 && (SvGETMAGIC(ST(1)), SvOK(ST(1)))) { |
1 |
0 |
1 |
| } else if (items >= 2 && (SvGETMAGIC(ST(1)), SvOK(ST(1)))) { |
0 |
1 |
1 |
| } else if (items >= 2 && (SvGETMAGIC(ST(1)), SvOK(ST(1)))) { |
0 |
0 |
0 |
| } else if (items >= 2 && (SvGETMAGIC(ST(1)), SvOK(ST(1)))) { |
1 |
0 |
1 |
|
| 1239 |
|
| { |
| 1240 |
1 |
| const char *up = SvPV_nolen(ST(1)); |
| 1241 |
1 |
▶ | if (unlink(up) != 0 && errno != ENOENT) |
Branch: 2 branches
| Branch | True | False | Total |
| if (unlink(up) != 0 && errno != ENOENT) |
0 |
1 |
1 |
| if (unlink(up) != 0 && errno != ENOENT) |
0 |
0 |
0 |
|
| 1242 |
0 |
| croak("Data::NDArray::Shared->unlink(%s): %s", up, strerror(errno)); |
| 1243 |
|
| } |
| 1244 |
|
| } |