symjit 2.19.2__tar.gz → 2.22.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {symjit-2.19.2 → symjit-2.22.0}/Cargo.toml +5 -4
- {symjit-2.19.2/python/symjit.egg-info → symjit-2.22.0}/PKG-INFO +1 -1
- {symjit-2.19.2 → symjit-2.22.0}/pyproject.toml +1 -1
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/__init__.py +65 -40
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/engine.py +5 -3
- {symjit-2.19.2 → symjit-2.22.0/python/symjit.egg-info}/PKG-INFO +1 -1
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit.egg-info/SOURCES.txt +5 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit.egg-info/dependency_links.txt +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit.egg-info/top_level.txt +0 -0
- symjit-2.22.0/rust/allocator.rs +649 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/amd/asm.rs +2 -2
- {symjit-2.19.2 → symjit-2.22.0}/rust/amd/complex.rs +153 -205
- symjit-2.22.0/rust/amd/f64x8.rs +477 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/amd/fused.rs +70 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/amd/mod.rs +48 -30
- {symjit-2.19.2 → symjit-2.22.0}/rust/amd/scalar.rs +123 -132
- {symjit-2.19.2 → symjit-2.22.0}/rust/amd/sse.rs +110 -122
- {symjit-2.19.2 → symjit-2.22.0}/rust/amd/vector.rs +214 -264
- symjit-2.22.0/rust/amd/vector_f64x8.rs +859 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/applet.rs +171 -11
- {symjit-2.19.2 → symjit-2.22.0}/rust/arm/complex.rs +95 -116
- symjit-2.22.0/rust/arm/funclet_tests.rs +132 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/arm/macros.rs +52 -154
- symjit-2.22.0/rust/arm/mod.rs +391 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/arm/scalar.rs +85 -119
- {symjit-2.19.2 → symjit-2.22.0}/rust/arm/vector.rs +243 -222
- {symjit-2.19.2 → symjit-2.22.0}/rust/bin.rs +19 -1
- {symjit-2.19.2 → symjit-2.22.0}/rust/block.rs +47 -23
- {symjit-2.19.2 → symjit-2.22.0}/rust/builder.rs +106 -79
- {symjit-2.19.2 → symjit-2.22.0}/rust/compactor.rs +29 -10
- {symjit-2.19.2 → symjit-2.22.0}/rust/compiler.rs +252 -152
- {symjit-2.19.2 → symjit-2.22.0}/rust/complexify.rs +15 -10
- {symjit-2.19.2 → symjit-2.22.0}/rust/composer.rs +11 -10
- {symjit-2.19.2 → symjit-2.22.0}/rust/config.rs +403 -60
- {symjit-2.19.2 → symjit-2.22.0}/rust/defuns.rs +6 -4
- {symjit-2.19.2 → symjit-2.22.0}/rust/expr.rs +16 -18
- {symjit-2.19.2 → symjit-2.22.0}/rust/generator.rs +29 -5
- symjit-2.22.0/rust/lib.rs +956 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/matrix.rs +4 -1
- {symjit-2.19.2 → symjit-2.22.0}/rust/mir.rs +161 -136
- symjit-2.19.2/rust/lib.rs → symjit-2.22.0/rust/mod.rs +4 -4
- {symjit-2.19.2 → symjit-2.22.0}/rust/model.rs +13 -11
- {symjit-2.19.2 → symjit-2.22.0}/rust/node.rs +105 -60
- symjit-2.22.0/rust/operation.rs +57 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/parser.rs +1 -1
- {symjit-2.19.2 → symjit-2.22.0}/rust/riscv64/macros.rs +1 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/riscv64/mod.rs +149 -94
- {symjit-2.19.2 → symjit-2.22.0}/rust/runnable.rs +117 -44
- {symjit-2.19.2 → symjit-2.22.0}/rust/serializer.rs +4 -4
- {symjit-2.19.2 → symjit-2.22.0}/rust/statement.rs +21 -10
- {symjit-2.19.2 → symjit-2.22.0}/rust/symbol.rs +19 -1
- {symjit-2.19.2 → symjit-2.22.0}/rust/utils.rs +39 -2
- symjit-2.19.2/rust/allocator.rs +0 -1105
- symjit-2.19.2/rust/arm/mod.rs +0 -184
- {symjit-2.19.2 → symjit-2.22.0}/LICENSE +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/MANIFEST.in +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/README.md +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/composer.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/func.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/ode.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/pyengine/__init__.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/pyengine/amd.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/pyengine/arm.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/pyengine/assembler.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/pyengine/tree.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/pyengine/vectorizer.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/python/symjit/structure.py +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/assembler.rs +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/code.rs +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/instruction.rs +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/machine.rs +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/memory.rs +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/rust/types.rs +0 -0
- {symjit-2.19.2 → symjit-2.22.0}/setup.cfg +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
[package]
|
|
2
2
|
name = "symjit"
|
|
3
|
-
version = "2.
|
|
3
|
+
version = "2.22.0"
|
|
4
4
|
edition = "2021"
|
|
5
5
|
authors = ["Shahriar Iravanian <siravan@svtsim.com>"]
|
|
6
6
|
license = "MIT"
|
|
@@ -19,17 +19,18 @@ libc = "0.2"
|
|
|
19
19
|
# libm = "0.2"
|
|
20
20
|
region = "3.0.2"
|
|
21
21
|
wasmtime-jit-icache-coherence = "*"
|
|
22
|
-
|
|
22
|
+
num-traits = "0.2"
|
|
23
23
|
rayon = "1.1"
|
|
24
24
|
spec_math = "0.1.6"
|
|
25
|
-
petgraph= "0.8.3"
|
|
25
|
+
# petgraph= "0.8.3"
|
|
26
26
|
hashbrown = "0.16.0"
|
|
27
|
-
num-traits = "0.2"
|
|
27
|
+
# num-traits = "0.2"
|
|
28
28
|
num-complex = "0.4"
|
|
29
29
|
wide = "1.1"
|
|
30
30
|
toml= "1.1"
|
|
31
31
|
hugepage-rs = "0.1"
|
|
32
32
|
lazy_static = "1.5"
|
|
33
|
+
rand = "0.10"
|
|
33
34
|
|
|
34
35
|
[target.'cfg(windows)'.dependencies]
|
|
35
36
|
windows-sys = { version = "0.61", features = ["Win32_System_Memory"] }
|
|
@@ -35,14 +35,15 @@ def compile_func(
|
|
|
35
35
|
obs=None,
|
|
36
36
|
ty="native",
|
|
37
37
|
use_simd=True,
|
|
38
|
+
enable_simd512=False,
|
|
38
39
|
use_threads=True,
|
|
39
40
|
cse=True,
|
|
40
41
|
fastmath=True,
|
|
41
42
|
backend="rust",
|
|
42
43
|
opt_level=2,
|
|
43
44
|
defuns=None,
|
|
44
|
-
sanitize=True,
|
|
45
45
|
dtype="float64",
|
|
46
|
+
compact=True,
|
|
46
47
|
compress=False,
|
|
47
48
|
fast_complex=True,
|
|
48
49
|
huge=False,
|
|
@@ -77,14 +78,16 @@ def compile_func(
|
|
|
77
78
|
dtype (default `float64`): the data type. Possibilities are `float64` and `complex128`.
|
|
78
79
|
use_simd (default `True`): generates SIMD code for vectorized operations. Currently supports
|
|
79
80
|
AVX on x86-64 and NEON on aarch64 systems.
|
|
81
|
+
enable_avx512 (default `False`): allows using AVX512 (f64x8) SIMD type when available.
|
|
80
82
|
use_threads (default `True`): use multi-threading to speed up parallel operations when called
|
|
81
83
|
on numpy arrays.
|
|
82
84
|
cse (default `True`): performs common-subexpression elimination.
|
|
83
|
-
fastmath (default
|
|
84
|
-
fast_complex (default True): use f64x2 SIMD instructions for complex operations.
|
|
85
|
-
parallel_mul (default True): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
86
|
-
huge (default False): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
87
|
-
compress (default False): compress binary size by converting inline operations to calls.
|
|
85
|
+
fastmath (default `True`): use fastmath floating point operations, especially fused multiply-addition.
|
|
86
|
+
fast_complex (default `True`): use f64x2 SIMD instructions for complex operations.
|
|
87
|
+
parallel_mul (default `True`): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
88
|
+
huge (default `False`): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
89
|
+
compress (default `False`): compress binary size by converting inline operations to calls.
|
|
90
|
+
compact (default `True`): compacts the stack frame by reusing temporary variables.
|
|
88
91
|
opt_level (default 2): optimization level (0, 1, 2, or 3). Broadly the numbers are parallel to -O0, -O1, -O2, and -O3
|
|
89
92
|
options to gcc and clang. Level-0 performs minimum amount of optimization. Level-1 does peephole optimization.
|
|
90
93
|
Levels 2 and 3 use improved graph-coloring algorithms for better register allocation.
|
|
@@ -116,8 +119,9 @@ def compile_func(
|
|
|
116
119
|
fastmath=fastmath,
|
|
117
120
|
opt_level=opt_level,
|
|
118
121
|
defuns=defuns,
|
|
119
|
-
|
|
122
|
+
enable_simd512=enable_simd512,
|
|
120
123
|
dtype=dtype,
|
|
124
|
+
compact=compact,
|
|
121
125
|
compress=compress,
|
|
122
126
|
fast_complex=fast_complex,
|
|
123
127
|
huge=huge,
|
|
@@ -157,18 +161,19 @@ def compile_ode(
|
|
|
157
161
|
params=None,
|
|
158
162
|
ty="native",
|
|
159
163
|
use_simd=True,
|
|
164
|
+
enable_simd512=False,
|
|
160
165
|
use_threads=True,
|
|
161
166
|
cse=True,
|
|
162
167
|
fastmath=True,
|
|
163
168
|
backend="rust",
|
|
164
169
|
opt_level=2,
|
|
165
170
|
defuns=None,
|
|
166
|
-
sanitize=True,
|
|
167
171
|
dtype="float64",
|
|
172
|
+
compact=True,
|
|
173
|
+
compress=False,
|
|
168
174
|
fast_complex=True,
|
|
169
175
|
huge=False,
|
|
170
176
|
parallel_mul=True,
|
|
171
|
-
compress=False,
|
|
172
177
|
):
|
|
173
178
|
"""Compile a symbolic ODE model into an executable form suitable for
|
|
174
179
|
passung to scipy.integrate.solve_ivp.
|
|
@@ -187,14 +192,16 @@ def compile_ode(
|
|
|
187
192
|
dtype (default `float64`): the data type. Possibilities are `float64` and `complex128`.
|
|
188
193
|
use_simd (default `True`): generates SIMD code for vectorized operations. Currently
|
|
189
194
|
supports AVX on x86-64 and NEON on aarch64 systems.
|
|
195
|
+
enable_avx512 (default `False`): allows using AVX512 (f64x8) SIMD type when available.
|
|
190
196
|
use_threads (default `True`): use multi-threading to speed up parallel operations
|
|
191
197
|
when called on numpy arrays.
|
|
192
198
|
cse (default `True`): performs common-subexpression elimination.
|
|
193
199
|
fastmath (default `True`): use fastmath floating point operations, especially fused multiply-addition.
|
|
194
|
-
fast_complex (default True): use f64x2 SIMD instructions for complex operations.
|
|
195
|
-
parallel_mul (default True): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
196
|
-
huge (default False): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
197
|
-
compress (default False): compress binary size by converting inline operations to calls.
|
|
200
|
+
fast_complex (default `True`): use f64x2 SIMD instructions for complex operations.
|
|
201
|
+
parallel_mul (default `True`): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
202
|
+
huge (default `False`): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
203
|
+
compress (default `False`): compress binary size by converting inline operations to calls.
|
|
204
|
+
compact (default `True`): compacts the stack frame by reusing temporary variables.
|
|
198
205
|
opt_level (default 2): optimization level (0, 1, 2, or 3). Broadly the numbers are parallel to -O0, -O1, -O2, and -O3
|
|
199
206
|
options to gcc and clang. Level-0 performs minimum amount of optimization. Level-1 does peephole optimization.
|
|
200
207
|
Levels 2 and 3 use improved graph-coloring algorithms for better register allocation.
|
|
@@ -228,17 +235,18 @@ def compile_ode(
|
|
|
228
235
|
model,
|
|
229
236
|
ty=ty,
|
|
230
237
|
use_simd=use_simd,
|
|
238
|
+
enable_simd512=enable_simd512,
|
|
231
239
|
use_threads=use_threads,
|
|
232
240
|
cse=cse,
|
|
233
241
|
fastmath=fastmath,
|
|
234
242
|
opt_level=opt_level,
|
|
235
243
|
defuns=defuns,
|
|
236
|
-
sanitize=sanitize,
|
|
237
244
|
dtype=dtype,
|
|
238
245
|
fast_complex=fast_complex,
|
|
239
246
|
huge=huge,
|
|
240
247
|
parallel_mul=parallel_mul,
|
|
241
248
|
compress=compress,
|
|
249
|
+
compact=compact,
|
|
242
250
|
)
|
|
243
251
|
elif can_use_python(backend):
|
|
244
252
|
model = pyengine.tree.model_ode(iv, states, odes, params)
|
|
@@ -259,18 +267,19 @@ def compile_jac(
|
|
|
259
267
|
params=None,
|
|
260
268
|
ty="native",
|
|
261
269
|
use_simd=True,
|
|
270
|
+
enable_simd512=False,
|
|
262
271
|
use_threads=True,
|
|
263
272
|
cse=True,
|
|
264
273
|
fastmath=True,
|
|
265
274
|
backend="rust",
|
|
266
275
|
opt_level=2,
|
|
267
276
|
defuns=None,
|
|
268
|
-
sanitize=True,
|
|
269
277
|
dtype="float64",
|
|
270
278
|
fast_complex=True,
|
|
271
279
|
huge=False,
|
|
272
280
|
parallel_mul=True,
|
|
273
281
|
compress=False,
|
|
282
|
+
compact=True,
|
|
274
283
|
):
|
|
275
284
|
"""Genenrates and compiles Jacobian for an ODE system.
|
|
276
285
|
iv: a single symbol, the independent variable.
|
|
@@ -284,14 +293,16 @@ def compile_jac(
|
|
|
284
293
|
dtype (default `float64`): the data type. Possibilities are `float64` and `complex128`.
|
|
285
294
|
use_simd (default `True`): generates SIMD code for vectorized operations. Currently
|
|
286
295
|
supports AVX on x86-64 and NEON on aarch64 systems.
|
|
296
|
+
enable_avx512 (default `False`): allows using AVX512 (f64x8) SIMD type when available.
|
|
287
297
|
use_threads (default `True`): use multi-threading to speed up parallel operations when called
|
|
288
298
|
on numpy arrays.
|
|
289
299
|
cse (default `True`): performs common-subexpression elimination.
|
|
290
300
|
fastmath (default `True`): use fastmath floating point operations, especially fused multiply-addition.
|
|
291
|
-
fast_complex (default True): use f64x2 SIMD instructions for complex operations.
|
|
292
|
-
parallel_mul (default True): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
293
|
-
huge (default False): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
294
|
-
compress (default False): compress binary size by converting inline operations to calls.
|
|
301
|
+
fast_complex (default `True`): use f64x2 SIMD instructions for complex operations.
|
|
302
|
+
parallel_mul (default `True`): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
303
|
+
huge (default `False`): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
304
|
+
compress (default `False`): compress binary size by converting inline operations to calls.
|
|
305
|
+
compact (default `True`): compacts the stack frame by reusing temporary variables.
|
|
295
306
|
opt_level (default 2): optimization level (0, 1, 2, or 3). Broadly the numbers are parallel to -O0, -O1, -O2, and -O3
|
|
296
307
|
options to gcc and clang. Level-0 performs minimum amount of optimization. Level-1 does peephole optimization.
|
|
297
308
|
Levels 2 and 3 use improved graph-coloring algorithms for better register allocation.
|
|
@@ -311,17 +322,18 @@ def compile_jac(
|
|
|
311
322
|
model,
|
|
312
323
|
ty=ty,
|
|
313
324
|
use_simd=use_simd,
|
|
325
|
+
enable_simd512=enable_simd512,
|
|
314
326
|
use_threads=use_threads,
|
|
315
327
|
cse=cse,
|
|
316
328
|
fastmath=fastmath,
|
|
317
329
|
opt_level=opt_level,
|
|
318
330
|
defuns=defuns,
|
|
319
|
-
sanitize=sanitize,
|
|
320
331
|
dtype=dtype,
|
|
321
332
|
fast_complex=fast_complex,
|
|
322
333
|
huge=huge,
|
|
323
334
|
parallel_mul=parallel_mul,
|
|
324
335
|
compress=compress,
|
|
336
|
+
compact=compact,
|
|
325
337
|
)
|
|
326
338
|
elif can_use_python(backend):
|
|
327
339
|
model = pyengine.tree.model_jac(iv, states, odes, params)
|
|
@@ -356,16 +368,17 @@ def compile_json(
|
|
|
356
368
|
model,
|
|
357
369
|
ty="native",
|
|
358
370
|
use_simd=True,
|
|
371
|
+
enable_simd512=False,
|
|
359
372
|
use_threads=True,
|
|
360
373
|
cse=True,
|
|
361
374
|
fastmath=True,
|
|
362
375
|
opt_level=2,
|
|
363
|
-
sanitize=True,
|
|
364
376
|
dtype="float64",
|
|
365
377
|
fast_complex=True,
|
|
366
378
|
huge=False,
|
|
367
379
|
parallel_mul=True,
|
|
368
380
|
compress=False,
|
|
381
|
+
compact=True,
|
|
369
382
|
backend="rust",
|
|
370
383
|
):
|
|
371
384
|
"""Compiles CellML models
|
|
@@ -380,17 +393,18 @@ def compile_json(
|
|
|
380
393
|
model,
|
|
381
394
|
ty=ty,
|
|
382
395
|
use_simd=use_simd,
|
|
396
|
+
enable_simd512=enable_simd512,
|
|
383
397
|
use_threads=use_threads,
|
|
384
398
|
cse=cse,
|
|
385
399
|
fastmath=fastmath,
|
|
386
400
|
opt_level=opt_level,
|
|
387
401
|
defuns=None,
|
|
388
|
-
sanitize=sanitize,
|
|
389
402
|
dtype=dtype,
|
|
390
403
|
fast_complex=fast_complex,
|
|
391
404
|
huge=huge,
|
|
392
405
|
parallel_mul=parallel_mul,
|
|
393
406
|
compress=compress,
|
|
407
|
+
compact=compact,
|
|
394
408
|
)
|
|
395
409
|
|
|
396
410
|
if compiler.count_diffs == 0:
|
|
@@ -411,13 +425,13 @@ def compile_evaluator(
|
|
|
411
425
|
evaluator,
|
|
412
426
|
ty="native",
|
|
413
427
|
use_simd=True,
|
|
428
|
+
enable_simd512=False,
|
|
414
429
|
use_threads=True,
|
|
415
|
-
cse=
|
|
430
|
+
cse=True,
|
|
416
431
|
fastmath=True,
|
|
417
432
|
backend="rust",
|
|
418
|
-
opt_level=
|
|
433
|
+
opt_level=3,
|
|
419
434
|
defuns=None,
|
|
420
|
-
sanitize=True,
|
|
421
435
|
dtype="float64",
|
|
422
436
|
num_params=1,
|
|
423
437
|
order="c",
|
|
@@ -427,6 +441,7 @@ def compile_evaluator(
|
|
|
427
441
|
huge=False,
|
|
428
442
|
parallel_mul=True,
|
|
429
443
|
compress=False,
|
|
444
|
+
compact=True,
|
|
430
445
|
):
|
|
431
446
|
"""Compiles an Evaluator object generated by Symbolica.
|
|
432
447
|
|
|
@@ -446,13 +461,15 @@ def compile_evaluator(
|
|
|
446
461
|
dtype (default `float64`): the data type. Possibilities are `float64` and `complex128`.
|
|
447
462
|
use_simd (default `True`): generates SIMD code for vectorized operations. Currently supports
|
|
448
463
|
AVX on x86-64 and NEON on aarch64 systems.
|
|
464
|
+
enable_avx512 (default `False`): allows using AVX512 (f64x8) SIMD type when available.
|
|
449
465
|
use_threads (default `False`): currently not supported for the Symbolica bridge.
|
|
450
466
|
cse (default `False`): performs common-subexpression elimination.
|
|
451
|
-
fastmath (default False): use fastmath floating point operations, especially fused multiply-addition.
|
|
452
|
-
fast_complex (default True): use f64x2 SIMD instructions for complex operations.
|
|
453
|
-
parallel_mul (default True): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
454
|
-
huge (default False): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
455
|
-
compress (default False): compress binary size by converting inline operations to calls.
|
|
467
|
+
fastmath (default `False`): use fastmath floating point operations, especially fused multiply-addition.
|
|
468
|
+
fast_complex (default `True`): use f64x2 SIMD instructions for complex operations.
|
|
469
|
+
parallel_mul (default `True`): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
470
|
+
huge (default `False`): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
471
|
+
compress (default `False`): compress binary size by converting inline operations to calls.
|
|
472
|
+
compact (default `True`): compacts the stack frame by reusing temporary variables.
|
|
456
473
|
opt_level (default 2): optimization level (0, 1, 2, or 3). Broadly the numbers are parallel to -O0, -O1, -O2, and -O3
|
|
457
474
|
options to gcc and clang. Level-0 performs minimum amount of optimization. Level-1 does peephole optimization.
|
|
458
475
|
Levels 2 and 3 use improved graph-coloring algorithms for better register allocation.
|
|
@@ -485,12 +502,12 @@ def compile_evaluator(
|
|
|
485
502
|
model,
|
|
486
503
|
ty=ty,
|
|
487
504
|
use_simd=use_simd,
|
|
505
|
+
enable_simd512=enable_simd512,
|
|
488
506
|
use_threads=use_threads,
|
|
489
507
|
cse=cse,
|
|
490
508
|
fastmath=fastmath,
|
|
491
509
|
opt_level=opt_level,
|
|
492
510
|
defuns=defuns,
|
|
493
|
-
sanitize=sanitize,
|
|
494
511
|
dtype=dtype,
|
|
495
512
|
action="translate",
|
|
496
513
|
convert=False,
|
|
@@ -502,18 +519,19 @@ def compile_evaluator(
|
|
|
502
519
|
huge=huge,
|
|
503
520
|
parallel_mul=parallel_mul,
|
|
504
521
|
compress=compress,
|
|
522
|
+
compact=compact,
|
|
505
523
|
)
|
|
506
524
|
else: # order == "fortran"
|
|
507
525
|
compiler = engine.RustyCompiler(
|
|
508
526
|
model,
|
|
509
527
|
ty=ty,
|
|
510
528
|
use_simd=use_simd,
|
|
529
|
+
enable_simd512=enable_simd512,
|
|
511
530
|
use_threads=use_threads,
|
|
512
531
|
cse=cse,
|
|
513
532
|
fastmath=fastmath,
|
|
514
533
|
opt_level=opt_level,
|
|
515
534
|
defuns=defuns,
|
|
516
|
-
sanitize=sanitize,
|
|
517
535
|
dtype=dtype,
|
|
518
536
|
action="translate",
|
|
519
537
|
convert=False,
|
|
@@ -525,6 +543,7 @@ def compile_evaluator(
|
|
|
525
543
|
huge=huge,
|
|
526
544
|
parallel_mul=parallel_mul,
|
|
527
545
|
compress=compress,
|
|
546
|
+
compact=compact,
|
|
528
547
|
)
|
|
529
548
|
|
|
530
549
|
if dtype == "complex128":
|
|
@@ -549,8 +568,9 @@ def compile_composer(
|
|
|
549
568
|
ty="native",
|
|
550
569
|
dtype="float64",
|
|
551
570
|
use_simd=True,
|
|
571
|
+
enable_simd512=False,
|
|
552
572
|
use_threads=True,
|
|
553
|
-
cse=
|
|
573
|
+
cse=True,
|
|
554
574
|
fastmath=True,
|
|
555
575
|
backend="rust",
|
|
556
576
|
opt_level=2,
|
|
@@ -560,6 +580,7 @@ def compile_composer(
|
|
|
560
580
|
huge=False,
|
|
561
581
|
parallel_mul=True,
|
|
562
582
|
compress=False,
|
|
583
|
+
compact=True,
|
|
563
584
|
):
|
|
564
585
|
"""Compiles a Composer object.
|
|
565
586
|
|
|
@@ -579,13 +600,15 @@ def compile_composer(
|
|
|
579
600
|
dtype (default `float64`): the data type. Possibilities are `float64` and `complex128`.
|
|
580
601
|
use_simd (default `True`): generates SIMD code for vectorized operations. Currently supports
|
|
581
602
|
AVX on x86-64 and NEON on aarch64 systems.
|
|
582
|
-
|
|
583
|
-
|
|
584
|
-
|
|
585
|
-
|
|
586
|
-
|
|
587
|
-
|
|
588
|
-
|
|
603
|
+
enable_avx512 (default `False`): allows using AVX512 (f64x8) SIMD type when available.
|
|
604
|
+
use_threads (default `True`): currently not supported for the Symbolica bridge.
|
|
605
|
+
cse (default `True`): performs common-subexpression elimination.
|
|
606
|
+
fastmath (default `False`): use fastmath floating point operations, especially fused multiply-addition.
|
|
607
|
+
fast_complex (default `True`): use f64x2 SIMD instructions for complex operations.
|
|
608
|
+
parallel_mul (default `True`): try f64x4 SIMD indtructions to convert serial to parallel complex multiplications.
|
|
609
|
+
huge (default `False`): use huge (2 MB) pages instead of the standard 4 KB ones (only on Linux x86-64 machines).
|
|
610
|
+
compress (default `False`): compress binary size by converting inline operations to calls.
|
|
611
|
+
compact (default `True`): compacts the stack frame by reusing temporary variables.
|
|
589
612
|
opt_level (default 2): optimization level (0, 1, 2, or 3). Broadly the numbers are parallel to -O0, -O1, -O2, and -O3
|
|
590
613
|
options to gcc and clang. Level-0 performs minimum amount of optimization. Level-1 does peephole optimization.
|
|
591
614
|
Levels 2 and 3 use improved graph-coloring algorithms for better register allocation.
|
|
@@ -614,6 +637,7 @@ def compile_composer(
|
|
|
614
637
|
model,
|
|
615
638
|
ty=ty,
|
|
616
639
|
use_simd=use_simd,
|
|
640
|
+
enable_simd512=enable_simd512,
|
|
617
641
|
use_threads=use_threads,
|
|
618
642
|
cse=cse,
|
|
619
643
|
fastmath=fastmath,
|
|
@@ -630,6 +654,7 @@ def compile_composer(
|
|
|
630
654
|
huge=huge,
|
|
631
655
|
parallel_mul=parallel_mul,
|
|
632
656
|
compress=compress,
|
|
657
|
+
compact=compact,
|
|
633
658
|
)
|
|
634
659
|
|
|
635
660
|
|
|
@@ -323,13 +323,13 @@ class RustyCompiler:
|
|
|
323
323
|
model,
|
|
324
324
|
ty="native",
|
|
325
325
|
use_simd=True,
|
|
326
|
+
enable_simd512=False,
|
|
326
327
|
use_threads=True,
|
|
327
328
|
cse=True,
|
|
328
329
|
fastmath=True,
|
|
329
330
|
opt_level=1,
|
|
330
331
|
convert=True,
|
|
331
332
|
defuns=None,
|
|
332
|
-
sanitize=True,
|
|
333
333
|
dtype="float64",
|
|
334
334
|
action="compile",
|
|
335
335
|
file="",
|
|
@@ -337,7 +337,8 @@ class RustyCompiler:
|
|
|
337
337
|
order="fortran",
|
|
338
338
|
simd_branch=False,
|
|
339
339
|
fast_complex=True,
|
|
340
|
-
direct=
|
|
340
|
+
direct=False,
|
|
341
|
+
compact=True,
|
|
341
342
|
compress=False,
|
|
342
343
|
huge=False,
|
|
343
344
|
parallel_mul=True,
|
|
@@ -363,10 +364,11 @@ class RustyCompiler:
|
|
|
363
364
|
| (0x00000002 if use_threads else 0)
|
|
364
365
|
| (0x00000004 if cse else 0)
|
|
365
366
|
| (0x00000008 if fastmath else 0)
|
|
366
|
-
| (0x00000010 if
|
|
367
|
+
| (0x00000010 if enable_simd512 else 0)
|
|
367
368
|
| (0x00000020 if dtype == "complex128" else 0)
|
|
368
369
|
| (0x00000040 if order == "c" else 0)
|
|
369
370
|
| (0x00000080 if simd_branch else 0)
|
|
371
|
+
| (0x00001000 if compact else 0)
|
|
370
372
|
| (0x00002000 if compress else 0)
|
|
371
373
|
| (0x00004000 if direct else 0)
|
|
372
374
|
| (0x00008000 if fast_complex else 0)
|
|
@@ -40,8 +40,10 @@ rust/machine.rs
|
|
|
40
40
|
rust/matrix.rs
|
|
41
41
|
rust/memory.rs
|
|
42
42
|
rust/mir.rs
|
|
43
|
+
rust/mod.rs
|
|
43
44
|
rust/model.rs
|
|
44
45
|
rust/node.rs
|
|
46
|
+
rust/operation.rs
|
|
45
47
|
rust/parser.rs
|
|
46
48
|
rust/runnable.rs
|
|
47
49
|
rust/serializer.rs
|
|
@@ -51,12 +53,15 @@ rust/types.rs
|
|
|
51
53
|
rust/utils.rs
|
|
52
54
|
rust/amd/asm.rs
|
|
53
55
|
rust/amd/complex.rs
|
|
56
|
+
rust/amd/f64x8.rs
|
|
54
57
|
rust/amd/fused.rs
|
|
55
58
|
rust/amd/mod.rs
|
|
56
59
|
rust/amd/scalar.rs
|
|
57
60
|
rust/amd/sse.rs
|
|
58
61
|
rust/amd/vector.rs
|
|
62
|
+
rust/amd/vector_f64x8.rs
|
|
59
63
|
rust/arm/complex.rs
|
|
64
|
+
rust/arm/funclet_tests.rs
|
|
60
65
|
rust/arm/macros.rs
|
|
61
66
|
rust/arm/mod.rs
|
|
62
67
|
rust/arm/scalar.rs
|
|
File without changes
|
|
File without changes
|