| 1 | // This file is part of meshoptimizer library; see meshoptimizer.h for version/license details |
|---|---|
| 2 | #include "meshoptimizer.h" |
| 3 | |
| 4 | #include <assert.h> |
| 5 | #include <string.h> |
| 6 | |
| 7 | meshopt_VertexCacheStatistics meshopt_analyzeVertexCache(const unsigned int* indices, size_t index_count, size_t vertex_count, unsigned int cache_size, unsigned int warp_size, unsigned int primgroup_size) |
| 8 | { |
| 9 | assert(index_count % 3 == 0); |
| 10 | assert(cache_size >= 3); |
| 11 | assert(warp_size == 0 || warp_size >= 3); |
| 12 | |
| 13 | meshopt_Allocator allocator; |
| 14 | |
| 15 | meshopt_VertexCacheStatistics result = {}; |
| 16 | |
| 17 | unsigned int warp_offset = 0; |
| 18 | unsigned int primgroup_offset = 0; |
| 19 | |
| 20 | unsigned int* cache_timestamps = allocator.allocate<unsigned int>(vertex_count); |
| 21 | memset(cache_timestamps, 0, vertex_count * sizeof(unsigned int)); |
| 22 | |
| 23 | unsigned int timestamp = cache_size + 1; |
| 24 | |
| 25 | for (size_t i = 0; i < index_count; i += 3) |
| 26 | { |
| 27 | unsigned int a = indices[i + 0], b = indices[i + 1], c = indices[i + 2]; |
| 28 | assert(a < vertex_count && b < vertex_count && c < vertex_count); |
| 29 | |
| 30 | bool ac = (timestamp - cache_timestamps[a]) > cache_size; |
| 31 | bool bc = (timestamp - cache_timestamps[b]) > cache_size; |
| 32 | bool cc = (timestamp - cache_timestamps[c]) > cache_size; |
| 33 | |
| 34 | // flush cache if triangle doesn't fit into warp or into the primitive buffer |
| 35 | if ((primgroup_size && primgroup_offset == primgroup_size) || (warp_size && warp_offset + ac + bc + cc > warp_size)) |
| 36 | { |
| 37 | result.warps_executed += warp_offset > 0; |
| 38 | |
| 39 | warp_offset = 0; |
| 40 | primgroup_offset = 0; |
| 41 | |
| 42 | // reset cache |
| 43 | timestamp += cache_size + 1; |
| 44 | } |
| 45 | |
| 46 | // update cache and add vertices to warp |
| 47 | for (int j = 0; j < 3; ++j) |
| 48 | { |
| 49 | unsigned int index = indices[i + j]; |
| 50 | |
| 51 | if (timestamp - cache_timestamps[index] > cache_size) |
| 52 | { |
| 53 | cache_timestamps[index] = timestamp++; |
| 54 | result.vertices_transformed++; |
| 55 | warp_offset++; |
| 56 | } |
| 57 | } |
| 58 | |
| 59 | primgroup_offset++; |
| 60 | } |
| 61 | |
| 62 | size_t unique_vertex_count = 0; |
| 63 | |
| 64 | for (size_t i = 0; i < vertex_count; ++i) |
| 65 | unique_vertex_count += cache_timestamps[i] > 0; |
| 66 | |
| 67 | result.warps_executed += warp_offset > 0; |
| 68 | |
| 69 | result.acmr = index_count == 0 ? 0 : float(result.vertices_transformed) / float(index_count / 3); |
| 70 | result.atvr = unique_vertex_count == 0 ? 0 : float(result.vertices_transformed) / float(unique_vertex_count); |
| 71 | |
| 72 | return result; |
| 73 | } |
| 74 |