| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* -*- c++ -*- */ | ||
| 2 | /* | ||
| 3 | * Copyright 2011-2012 Free Software Foundation, Inc. | ||
| 4 | * | ||
| 5 | * This file is part of VOLK | ||
| 6 | * | ||
| 7 | * SPDX-License-Identifier: LGPL-3.0-or-later | ||
| 8 | */ | ||
| 9 | |||
| 10 | |||
| 11 | #include <stdio.h> | ||
| 12 | #include <stdlib.h> | ||
| 13 | #include <string.h> | ||
| 14 | |||
| 15 | #include <volk/volk_prefs.h> | ||
| 16 | #include <volk_rank_archs.h> | ||
| 17 | |||
| 18 | 1334 | int volk_get_index(const char* impl_names[], // list of implementations by name | |
| 19 | const size_t n_impls, // number of implementations available | ||
| 20 | const char* impl_name // the implementation name to find | ||
| 21 | ) | ||
| 22 | { | ||
| 23 | unsigned int i; | ||
| 24 |
1/2✓ Branch 0 taken 4832 times.
✗ Branch 1 not taken.
|
4832 | for (i = 0; i < n_impls; i++) { |
| 25 |
2/2✓ Branch 0 taken 1334 times.
✓ Branch 1 taken 3498 times.
|
4832 | if (!strncmp(impl_names[i], impl_name, 20)) { |
| 26 | 1334 | return i; | |
| 27 | } | ||
| 28 | } | ||
| 29 | // TODO return -1; | ||
| 30 | // something terrible should happen here | ||
| 31 | ✗ | fprintf(stderr, "Volk warning: no arch found, returning generic impl\n"); | |
| 32 | ✗ | return volk_get_index(impl_names, n_impls, "generic"); // but we'll fake it for now | |
| 33 | } | ||
| 34 | |||
| 35 | 16 | int volk_rank_archs(const char* kern_name, // name of the kernel to rank | |
| 36 | const char* impl_names[], // list of implementations by name | ||
| 37 | const int* impl_deps, // requirement mask per implementation | ||
| 38 | const bool* alignment, // alignment status of each implementation | ||
| 39 | size_t n_impls, // number of implementations available | ||
| 40 | const bool align // if false, filter aligned implementations | ||
| 41 | ) | ||
| 42 | { | ||
| 43 | size_t i; | ||
| 44 | static volk_arch_pref_t* volk_arch_prefs; | ||
| 45 | static size_t n_arch_prefs = 0; | ||
| 46 | static int prefs_loaded = 0; | ||
| 47 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 12 times.
|
16 | if (!prefs_loaded) { |
| 48 | 4 | n_arch_prefs = volk_load_preferences(&volk_arch_prefs); | |
| 49 | 4 | prefs_loaded = 1; | |
| 50 | } | ||
| 51 | |||
| 52 | // If we've defined VOLK_GENERIC to be anything, always return the | ||
| 53 | // 'generic' kernel. Used in GR's QA code. | ||
| 54 | 16 | char* gen_env = getenv("VOLK_GENERIC"); | |
| 55 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 16 times.
|
16 | if (gen_env) { |
| 56 | ✗ | return volk_get_index(impl_names, n_impls, "generic"); | |
| 57 | } | ||
| 58 | |||
| 59 | // now look for the function name in the prefs list | ||
| 60 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 16 times.
|
16 | for (i = 0; i < n_arch_prefs; i++) { |
| 61 | ✗ | if (!strncmp(kern_name, | |
| 62 | ✗ | volk_arch_prefs[i].name, | |
| 63 | sizeof(volk_arch_prefs[i].name))) // found it | ||
| 64 | { | ||
| 65 | ✗ | const char* impl_name = | |
| 66 | ✗ | align ? volk_arch_prefs[i].impl_a : volk_arch_prefs[i].impl_u; | |
| 67 | ✗ | return volk_get_index(impl_names, n_impls, impl_name); | |
| 68 | } | ||
| 69 | } | ||
| 70 | |||
| 71 | // return the best index with the largest deps | ||
| 72 | 16 | size_t best_index_a = 0; | |
| 73 | 16 | size_t best_index_u = 0; | |
| 74 | 16 | int best_value_a = -1; | |
| 75 | 16 | int best_value_u = -1; | |
| 76 |
2/2✓ Branch 0 taken 100 times.
✓ Branch 1 taken 16 times.
|
116 | for (i = 0; i < n_impls; i++) { |
| 77 | 100 | const signed val = impl_deps[i]; | |
| 78 |
4/4✓ Branch 0 taken 44 times.
✓ Branch 1 taken 56 times.
✓ Branch 2 taken 20 times.
✓ Branch 3 taken 24 times.
|
100 | if (alignment[i] && val > best_value_a) { |
| 79 | 20 | best_index_a = i; | |
| 80 | 20 | best_value_a = val; | |
| 81 | } | ||
| 82 |
4/4✓ Branch 0 taken 56 times.
✓ Branch 1 taken 44 times.
✓ Branch 2 taken 28 times.
✓ Branch 3 taken 28 times.
|
100 | if (!alignment[i] && val > best_value_u) { |
| 83 | 28 | best_index_u = i; | |
| 84 | 28 | best_value_u = val; | |
| 85 | } | ||
| 86 | } | ||
| 87 | |||
| 88 | // when align and we found a best aligned, use it | ||
| 89 |
3/4✓ Branch 0 taken 8 times.
✓ Branch 1 taken 8 times.
✓ Branch 2 taken 8 times.
✗ Branch 3 not taken.
|
16 | if (align && best_value_a != -1) |
| 90 | 8 | return best_index_a; | |
| 91 | |||
| 92 | // otherwise return the best unaligned | ||
| 93 | 8 | return best_index_u; | |
| 94 | } | ||
| 95 |