superintervals 0.2.2__tar.gz → 0.2.3__tar.gz

{superintervals-0.2.2/src/superintervals.egg-info → superintervals-0.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: superintervals
-Version: 0.2.2
+Version: 0.2.3
 Summary: Rapid interval intersections
 Author: Kez Cleal
 Author-email: Kez Cleal <clealk@cardiff.ac.uk>

{superintervals-0.2.2 → superintervals-0.2.3}/README.md

@@ -91,11 +91,11 @@ exceptions. Coitrees-s was faster for one test (ONT reads, sorted DB53 reads).
 
 Datasets:
 
-    1. `rna / anno` RNA-seq reads and annotations from cgranges repository
-    2. `ONT reads` nanopore alignments from sample PAO33946 chr1, converted to bed format
-    3. `DB53 reads` paired-end reads from sample DB53, NCBI BioProject PRJNA417592, chr1, converted to bed format
-    4. `mito-b, mito-a` paired-end reads from sample DB53 chrM, converted to bed format (mito-b and mito-a are the same)
-    5. `genes` UCSC genes from hg19
+- `rna / anno` RNA-seq reads and annotations from cgranges repository
+- `ONT reads` nanopore alignments from sample PAO33946 chr1, converted to bed format
+- `DB53 reads` paired-end reads from sample DB53, NCBI BioProject PRJNA417592, chr1, converted to bed format
+- `mito-b, mito-a` paired-end reads from sample DB53 chrM, converted to bed format (mito-b and mito-a are the same)
+- `genes` UCSC genes from hg19
 
 Test programs use internal timers and print data to stdout, measuring the index time, and time to find all intersections. Other steps such as file IO are ignored. Test programs also only assess chr1 bed records - other chromosomes are ignored. For 'chrM' records, the M was replaced with 1 using sed. Data were assessed in position sorted and random order. Datasets can be found on the Releases page, and the test/run_tools.sh script has instructions for how to repeat the benchmark.
 

{superintervals-0.2.2 → superintervals-0.2.3}/pyproject.toml

@@ -8,7 +8,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "superintervals"
-version = "0.2.2"
+version = "0.2.3"
 description = "Rapid interval intersections"
 dependencies = ['Cython']
 authors = [{name = "Kez Cleal", email = "clealk@cardiff.ac.uk"}]

{superintervals-0.2.2 → superintervals-0.2.3/src/superintervals.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: superintervals
-Version: 0.2.2
+Version: 0.2.3
 Summary: Rapid interval intersections
 Author: Kez Cleal
 Author-email: Kez Cleal <clealk@cardiff.ac.uk>

{superintervals-0.2.2 → superintervals-0.2.3}/src/superintervals.hpp

@@ -6,10 +6,14 @@
 #include <climits>
 #include <iostream>
 #include <limits>
-#if defined(__AVX2__)
-    #include <immintrin.h>
-#elif defined(__ARM_NEON__)
-    #include <arm_neon.h>
+#ifndef SI_NOSIMD
+    #if defined(__AVX2__)
+        #include <immintrin.h>
+    #elif defined(__ARM_NEON__) || defined(__aarch64__)
+        #include <arm_neon.h>
+    #else
+        #define SI_NOSIMD
+    #endif
 #endif
 
 /**
@@ -282,25 +286,37 @@ class SuperIntervals {
         size_t found = 0;
         size_t i = idx;
 
-#ifdef __AVX2__
+#ifdef SI_NOSIMD
+        constexpr size_t block = 16;
+#elif defined(__AVX2__)
         __m256i start_vec = _mm256_set1_epi32(start);
         constexpr size_t simd_width = 256 / (sizeof(S) * 8);
         constexpr size_t block = simd_width * 4;
-#elif defined __ARM_NEON
+#elif defined(__ARM_NEON__) || defined(__aarch64__)
         int32x4_t start_vec = vdupq_n_s32(start);
         constexpr size_t simd_width = 128 / (sizeof(S) * 8);
         uint32x4_t ones = vdupq_n_u32(1);
         constexpr size_t block = simd_width * 4;
-#else
-        constexpr size_t block = 16;
 #endif
 
         while (i > 0) {
             if (start <= ends[i]) {
                 ++found;
                 --i;
+#ifdef SI_NOSIMD
+            while (i > block) {  // Rely on compiler auto vectorize
+                    size_t count = 0;
+                    for (size_t j = i; j > i - block; --j) {
+                        count += (start <= ends[j]) ? 1 : 0;
+                    }
+                    found += count;
+                    i -= block;
+                    if (count < block && start > ends[i + 1]) {  // check for a branch
+                        break;
+                    }
+                }
 
-#ifdef __AVX2__
+#elif defined(__AVX2__)
                 while (i > block) {
                     size_t count = 0;
                     for (size_t j = i; j > i - block; j -= simd_width) {
@@ -315,7 +331,7 @@ class SuperIntervals {
                         break;
                     }
                 }
-#elif defined(__ARM_NEON__)
+#elif defined(__ARM_NEON__) || defined(__aarch64__)
                 while (i > block) {
                     size_t count = 0;
                     uint32x4_t mask, bool_mask;
@@ -332,19 +348,6 @@ class SuperIntervals {
                         break;
                     }
                 }
-#else  // Rely on compiler auto vectorize
-                while (i > block) {
-                    size_t count = 0;
-                    for (size_t j = i; j > i - block; --j) {
-                        count += (start <= ends[j]) ? 1 : 0;
-                    }
-                    found += count;
-                    i -= block;
-                    if (count < block && start > ends[i + 1]) {  // check for a branch
-//                    if (count < block) {  // check for a branch
-                        break;
-                    }
-                }
 #endif
             } else {
                 if (branch[i] >= i) {
@@ -419,17 +422,17 @@ class SuperIntervals {
         size_t found = 0;
         size_t i = idx;
 
-#ifdef __AVX2__
+#ifdef SI_NOSIMD
+        constexpr size_t block = 16;
+#elif defined(__AVX2__)
         __m256i start_vec = _mm256_set1_epi32(point);
         constexpr size_t simd_width = 256 / (sizeof(S) * 8);
         constexpr size_t block = simd_width * 4;
-#elif defined __ARM_NEON
+#elif defined(__ARM_NEON__) || defined(__aarch64__)
         int32x4_t start_vec = vdupq_n_s32(point);
         constexpr size_t simd_width = 128 / (sizeof(S) * 8);
         uint32x4_t ones = vdupq_n_u32(1);
         constexpr size_t block = simd_width * 4;
-#else
-        constexpr size_t block = 16;
 #endif
 
         while (i > 0) {
@@ -437,7 +440,19 @@ class SuperIntervals {
                 ++found;
                 --i;
 
-#ifdef __AVX2__
+#ifdef SI_NOSIMD
+                while (i > block) {
+                    size_t count = 0;
+                    for (size_t j = i; j > i - block; --j) {
+                        count += (point <= ends[j]) ? 1 : 0;
+                    }
+                    found += count;
+                    i -= block;
+                    if (count < block) {  // go check for a branch
+                        break;
+                    }
+                }
+#elif defined(__AVX2__)
                 while (i > block) {
                     size_t count = 0;
                     for (size_t j = i; j > i - block; j -= simd_width) {
@@ -452,7 +467,7 @@ class SuperIntervals {
                         break;
                     }
                 }
-#elif defined __ARM_NEON
+#elif defined(__ARM_NEON__) || defined(__aarch64__)
                 while (i > block) {
                     size_t count = 0;
                     uint32x4_t bool_mask;
@@ -468,18 +483,6 @@ class SuperIntervals {
                         break;
                     }
                 }
-#else
-                while (i > block) {
-                    size_t count = 0;
-                    for (size_t j = i; j > i - block; --j) {
-                        count += (point <= ends[j]) ? 1 : 0;
-                    }
-                    found += count;
-                    i -= block;
-                    if (count < block) {  // go check for a branch
-                        break;
-                    }
-                }
 #endif
             } else {
                 if (branch[i] >= i) {