Add more sorting algorithms

Makefile

@@ -143,6 +143,7 @@ include libc/time/time.mk			# │
 include libc/stdio/stdio.mk			# │
 include third_party/libcxx/libcxx.mk		# │
 include net/net.mk				# │
+include third_party/vqsort/vqsort.mk		# │
 include libc/log/log.mk				# │
 include third_party/bzip2/bzip2.mk		# │
 include dsp/core/core.mk			# │

examples/examples.mk

@@ -89,6 +89,7 @@ EXAMPLES_DIRECTDEPS =								\
 	THIRD_PARTY_SED								\
 	THIRD_PARTY_STB								\
 	THIRD_PARTY_TR								\
+	THIRD_PARTY_VQSORT							\
 	THIRD_PARTY_XED								\
 	THIRD_PARTY_ZLIB							\
 	TOOL_BUILD_LIB								\

examples/vqsort.c

@@ -0,0 +1,25 @@
+#if 0
+/*─────────────────────────────────────────────────────────────────╗
+│ To the extent possible under law, Justine Tunney has waived      │
+│ all copyright and related or neighboring rights to this file,    │
+│ as it is written in the following disclaimers:                   │
+│   • http://unlicense.org/                                        │
+│   • http://creativecommons.org/publicdomain/zero/1.0/            │
+╚─────────────────────────────────────────────────────────────────*/
+#endif
+#include "third_party/vqsort/vqsort.h"
+#include "libc/macros.internal.h"
+#include "libc/stdio/stdio.h"
+#include "third_party/vqsort/vqsort.h"
+
+// how to sort one gigabyte of 64-bit integers per second
+
+int main(int argc, char *argv[]) {
+  int64_t A[] = {9, 3, -3, 5, 23, 7};
+  vqsort_int64(A, ARRAYLEN(A));
+  for (int i = 0; i < ARRAYLEN(A); ++i) {
+    if (i) printf(" ");
+    printf("%ld", A[i]);
+  }
+  printf("\n");
+}

libc/mem/alg.h

@@ -34,6 +34,9 @@ int _tarjan(int, const int (*)[2], int, int[], int[], int *)
 char *_replacestr(const char *, const char *, const char *)
     paramsnonnull() __algalloc;
 
+bool radix_sort_int32(int32_t *, size_t);
+bool radix_sort_int64(int64_t *, size_t);
+
 COSMOPOLITAN_C_END_
 #endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
 #endif /* COSMOPOLITAN_LIBC_ALG_ALG_H_ */

libc/mem/radix_sort_int32.c

@@ -0,0 +1,101 @@
+/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
+│vi: set net ft=c ts=2 sts=2 sw=2 fenc=utf-8                                :vi│
+╞══════════════════════════════════════════════════════════════════════════════╡
+│ Copyright 2023 Justine Alexandra Roberts Tunney                              │
+│                                                                              │
+│ Permission to use, copy, modify, and/or distribute this software for         │
+│ any purpose with or without fee is hereby granted, provided that the         │
+│ above copyright notice and this permission notice appear in all copies.      │
+│                                                                              │
+│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL                │
+│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED                │
+│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE             │
+│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL         │
+│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR        │
+│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER               │
+│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
+│ PERFORMANCE OF THIS SOFTWARE.                                                │
+╚─────────────────────────────────────────────────────────────────────────────*/
+#include "libc/mem/alg.h"
+#include "libc/mem/mem.h"
+#include "libc/runtime/runtime.h"
+#include "libc/str/str.h"
+
+// Credit: Andrew Schein. 2009. Open-source C++ implementation of Radix
+//         Sort for double-precision floating points. (2009).
+
+#define HIST_SIZE               (size_t)2048
+#define get_byte_0(v)           ((v)&0x7FF)
+#define get_byte_1(v)           (((v) >> 11) & 0x7FF)
+#define get_byte_2_flip_sign(v) (((unsigned)(v) >> 22) ^ 0x200)
+
+bool radix_sort_int32(int32_t *A, size_t n) {
+  int32_t *T, *reader, *writer;
+  size_t i, pos, sum0, sum1, sum2, tsum, *b0, *b1, *b2;
+
+  if (n < HIST_SIZE) {
+    _intsort(A, n);
+    return true;
+  }
+
+  if (!(T = (int32_t *)malloc(n * sizeof(int32_t)))) {
+    return false;
+  }
+
+  if (!(b0 = (size_t *)calloc(HIST_SIZE * 3, sizeof(size_t)))) {
+    free(T);
+    return false;
+  }
+
+  b1 = b0 + HIST_SIZE;
+  b2 = b1 + HIST_SIZE;
+
+  for (i = 0; i < n; i++) {
+    b0[get_byte_0(A[i])]++;
+    b1[get_byte_1(A[i])]++;
+    b2[get_byte_2_flip_sign(A[i])]++;
+  }
+
+  sum0 = sum1 = sum2 = tsum = 0;
+
+  for (i = 0; i < HIST_SIZE; i++) {
+    tsum = b0[i] + sum0;
+    b0[i] = sum0 - 1;
+    sum0 = tsum;
+
+    tsum = b1[i] + sum1;
+    b1[i] = sum1 - 1;
+    sum1 = tsum;
+
+    tsum = b2[i] + sum2;
+    b2[i] = sum2 - 1;
+    sum2 = tsum;
+  }
+
+  writer = T;
+  reader = A;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_0(reader[i]);
+    writer[++b0[pos]] = reader[i];
+  }
+
+  writer = A;
+  reader = T;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_1(reader[i]);
+    writer[++b1[pos]] = reader[i];
+  }
+
+  writer = T;
+  reader = A;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_2_flip_sign(reader[i]);
+    writer[++b2[pos]] = reader[i];
+  }
+
+  memcpy(A, T, n * sizeof(int));
+
+  free(b0);
+  free(T);
+  return true;
+}

libc/mem/radix_sort_int64.c

@@ -0,0 +1,144 @@
+/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
+│vi: set net ft=c ts=2 sts=2 sw=2 fenc=utf-8                                :vi│
+╞══════════════════════════════════════════════════════════════════════════════╡
+│ Copyright 2023 Justine Alexandra Roberts Tunney                              │
+│                                                                              │
+│ Permission to use, copy, modify, and/or distribute this software for         │
+│ any purpose with or without fee is hereby granted, provided that the         │
+│ above copyright notice and this permission notice appear in all copies.      │
+│                                                                              │
+│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL                │
+│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED                │
+│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE             │
+│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL         │
+│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR        │
+│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER               │
+│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
+│ PERFORMANCE OF THIS SOFTWARE.                                                │
+╚─────────────────────────────────────────────────────────────────────────────*/
+#include "libc/mem/alg.h"
+#include "libc/mem/mem.h"
+#include "libc/runtime/runtime.h"
+#include "libc/str/str.h"
+
+// Credit: Andrew Schein. 2009. Open-source C++ implementation of Radix
+//         Sort for double-precision floating points. (2009).
+
+#define HIST_SIZE               (size_t)2048
+#define get_byte_0(v)           ((v)&0x7FF)
+#define get_byte_1(v)           (((v) >> 11) & 0x7FF)
+#define get_byte_2(v)           (((v) >> 22) & 0x7FF)
+#define get_byte_3(v)           (((v) >> 33) & 0x7FF)
+#define get_byte_4(v)           (((v) >> 44) & 0x7FF)
+#define get_byte_5(v)           (((v) >> 55) & 0x7FF)
+#define get_byte_2_flip_sign(v) (((unsigned)(v) >> 22) ^ 0x200)
+#define get_byte_5_flip_sign(v) ((((v) >> 55) & 0x7FF) ^ 0x400)
+
+bool radix_sort_int64(int64_t *A, size_t n) {
+  int64_t *T, *reader, *writer;
+  size_t *b0, *b1, *b2, *b3, *b4, *b5;
+  size_t i, pos, sum0, sum1, sum2, sum3, sum4, sum5, tsum;
+
+  if (n < HIST_SIZE) {
+    _longsort(A, n);
+    return true;
+  }
+
+  if (!(T = (int64_t *)malloc(n * sizeof(int64_t)))) {
+    return false;
+  }
+
+  if (!(b0 = (size_t *)calloc(HIST_SIZE * 6, sizeof(size_t)))) {
+    free(T);
+    return false;
+  }
+
+  b1 = b0 + HIST_SIZE;
+  b2 = b1 + HIST_SIZE;
+  b3 = b2 + HIST_SIZE;
+  b4 = b3 + HIST_SIZE;
+  b5 = b4 + HIST_SIZE;
+
+  for (i = 0; i < n; i++) {
+    b0[get_byte_0(A[i])]++;
+    b1[get_byte_1(A[i])]++;
+    b2[get_byte_2(A[i])]++;
+    b3[get_byte_3(A[i])]++;
+    b4[get_byte_4(A[i])]++;
+    b5[get_byte_5_flip_sign(A[i])]++;
+  }
+
+  sum0 = sum1 = sum2 = sum3 = sum4 = sum5 = tsum = 0;
+
+  for (i = 0; i < HIST_SIZE; i++) {
+    tsum = b0[i] + sum0;
+    b0[i] = sum0 - 1;
+    sum0 = tsum;
+
+    tsum = b1[i] + sum1;
+    b1[i] = sum1 - 1;
+    sum1 = tsum;
+
+    tsum = b2[i] + sum2;
+    b2[i] = sum2 - 1;
+    sum2 = tsum;
+
+    tsum = b3[i] + sum3;
+    b3[i] = sum3 - 1;
+    sum3 = tsum;
+
+    tsum = b4[i] + sum4;
+    b4[i] = sum4 - 1;
+    sum4 = tsum;
+
+    tsum = b5[i] + sum5;
+    b5[i] = sum5 - 1;
+    sum5 = tsum;
+  }
+
+  writer = T;
+  reader = A;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_0(reader[i]);
+    writer[++b0[pos]] = reader[i];
+  }
+
+  writer = A;
+  reader = T;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_1(reader[i]);
+    writer[++b1[pos]] = reader[i];
+  }
+
+  writer = T;
+  reader = A;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_2(reader[i]);
+    writer[++b2[pos]] = reader[i];
+  }
+
+  writer = A;
+  reader = T;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_3(reader[i]);
+    writer[++b3[pos]] = reader[i];
+  }
+
+  writer = T;
+  reader = A;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_4(reader[i]);
+    writer[++b4[pos]] = reader[i];
+  }
+
+  writer = A;
+  reader = T;
+  for (i = 0; i < n; i++) {
+    pos = get_byte_5_flip_sign(reader[i]);
+    writer[++b5[pos]] = reader[i];
+  }
+
+  free(b0);
+  free(T);
+  return true;
+}

test/libc/str/longsort_test.c

@@ -21,9 +21,11 @@
 #include "libc/mem/mem.h"
 #include "libc/runtime/runtime.h"
 #include "libc/stdio/rand.h"
+#include "libc/stdio/stdio.h"
 #include "libc/str/str.h"
 #include "libc/testlib/ezbench.h"
 #include "libc/testlib/testlib.h"
+#include "third_party/vqsort/vqsort.h"
 
 int CompareLong(const void *a, const void *b) {
   const long *x = a;
@@ -44,13 +46,88 @@ TEST(_longsort, test) {
   ASSERT_EQ(0, memcmp(b, a, n * sizeof(long)));
 }
 
+TEST(vqsort_int64_avx2, test) {
+  if (!X86_HAVE(AVX2)) return;
+  size_t n = 5000;
+  long *a = gc(calloc(n, sizeof(long)));
+  long *b = gc(calloc(n, sizeof(long)));
+  rngset(a, n * sizeof(long), 0, 0);
+  memcpy(b, a, n * sizeof(long));
+  qsort(a, n, sizeof(long), CompareLong);
+  vqsort_int64_avx2(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(long)));
+}
+
+TEST(vqsort_int64_sse4, test) {
+  if (!X86_HAVE(SSE4_2)) return;
+  size_t n = 5000;
+  long *a = gc(calloc(n, sizeof(long)));
+  long *b = gc(calloc(n, sizeof(long)));
+  rngset(a, n * sizeof(long), 0, 0);
+  memcpy(b, a, n * sizeof(long));
+  qsort(a, n, sizeof(long), CompareLong);
+  vqsort_int64_sse4(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(long)));
+}
+
+TEST(vqsort_int64_ssse3, test) {
+  if (!X86_HAVE(SSSE3)) return;
+  size_t n = 5000;
+  long *a = gc(calloc(n, sizeof(long)));
+  long *b = gc(calloc(n, sizeof(long)));
+  rngset(a, n * sizeof(long), 0, 0);
+  memcpy(b, a, n * sizeof(long));
+  qsort(a, n, sizeof(long), CompareLong);
+  vqsort_int64_ssse3(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(long)));
+}
+
+TEST(vqsort_int64_sse2, test) {
+  size_t n = 5000;
+  long *a = gc(calloc(n, sizeof(long)));
+  long *b = gc(calloc(n, sizeof(long)));
+  rngset(a, n * sizeof(long), 0, 0);
+  memcpy(b, a, n * sizeof(long));
+  qsort(a, n, sizeof(long), CompareLong);
+  vqsort_int64_sse2(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(long)));
+}
+
+TEST(radix_sort_int64, test) {
+  size_t n = 5000;
+  long *a = gc(calloc(n, sizeof(long)));
+  long *b = gc(calloc(n, sizeof(long)));
+  rngset(a, n * sizeof(long), 0, 0);
+  memcpy(b, a, n * sizeof(long));
+  qsort(a, n, sizeof(long), CompareLong);
+  radix_sort_int64(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(long)));
+}
+
 BENCH(_longsort, bench) {
-  size_t n = 1000;
+  printf("\n");
+  size_t n = 5000;
   long *p1 = gc(malloc(n * sizeof(long)));
   long *p2 = gc(malloc(n * sizeof(long)));
   rngset(p1, n * sizeof(long), 0, 0);
   EZBENCH2("_longsort", memcpy(p2, p1, n * sizeof(long)), _longsort(p2, n));
-  EZBENCH2("qsort", memcpy(p2, p1, n * sizeof(long)),
+  if (X86_HAVE(AVX2)) {
+    EZBENCH2("vqsort_int64_avx2", memcpy(p2, p1, n * sizeof(long)),
+             vqsort_int64_avx2(p2, n));
+  }
+  if (X86_HAVE(SSE4_2)) {
+    EZBENCH2("vqsort_int64_sse4", memcpy(p2, p1, n * sizeof(long)),
+             vqsort_int64_sse4(p2, n));
+  }
+  if (X86_HAVE(SSSE3)) {
+    EZBENCH2("vqsort_int64_ssse3", memcpy(p2, p1, n * sizeof(long)),
+             vqsort_int64_ssse3(p2, n));
+  }
+  EZBENCH2("vqsort_int64_sse2", memcpy(p2, p1, n * sizeof(long)),
+           vqsort_int64_sse2(p2, n));
+  EZBENCH2("radix_sort_int64", memcpy(p2, p1, n * sizeof(long)),
+           radix_sort_int64(p2, n));
+  EZBENCH2("qsort(long)", memcpy(p2, p1, n * sizeof(long)),
            qsort(p2, n, sizeof(long), CompareLong));
 }
 
@@ -73,12 +150,88 @@ TEST(_intsort, test) {
   ASSERT_EQ(0, memcmp(b, a, n * sizeof(int)));
 }
 
+TEST(vqsort_int32_avx2, test) {
+  if (!X86_HAVE(AVX2)) return;
+  size_t n = 5000;
+  int *a = gc(calloc(n, sizeof(int)));
+  int *b = gc(calloc(n, sizeof(int)));
+  rngset(a, n * sizeof(int), 0, 0);
+  memcpy(b, a, n * sizeof(int));
+  qsort(a, n, sizeof(int), CompareInt);
+  vqsort_int32_avx2(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(int)));
+}
+
+TEST(vqsort_int32_sse4, test) {
+  if (!X86_HAVE(SSE4_2)) return;
+  size_t n = 5000;
+  int *a = gc(calloc(n, sizeof(int)));
+  int *b = gc(calloc(n, sizeof(int)));
+  rngset(a, n * sizeof(int), 0, 0);
+  memcpy(b, a, n * sizeof(int));
+  qsort(a, n, sizeof(int), CompareInt);
+  vqsort_int32_sse4(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(int)));
+}
+
+TEST(vqsort_int32_ssse3, test) {
+  if (!X86_HAVE(SSSE3)) return;
+  size_t n = 5000;
+  int *a = gc(calloc(n, sizeof(int)));
+  int *b = gc(calloc(n, sizeof(int)));
+  rngset(a, n * sizeof(int), 0, 0);
+  memcpy(b, a, n * sizeof(int));
+  qsort(a, n, sizeof(int), CompareInt);
+  vqsort_int32_ssse3(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(int)));
+}
+
+TEST(vqsort_int32_sse2, test) {
+  size_t n = 5000;
+  int *a = gc(calloc(n, sizeof(int)));
+  int *b = gc(calloc(n, sizeof(int)));
+  rngset(a, n * sizeof(int), 0, 0);
+  memcpy(b, a, n * sizeof(int));
+  qsort(a, n, sizeof(int), CompareInt);
+  vqsort_int32_sse2(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(int)));
+}
+
+TEST(radix_sort_int32, test) {
+  size_t n = 5000;
+  int *a = gc(calloc(n, sizeof(int)));
+  int *b = gc(calloc(n, sizeof(int)));
+  rngset(a, n * sizeof(int), 0, 0);
+  memcpy(b, a, n * sizeof(int));
+  qsort(a, n, sizeof(int), CompareInt);
+  radix_sort_int32(b, n);
+  ASSERT_EQ(0, memcmp(b, a, n * sizeof(int)));
+}
+
 BENCH(_intsort, bench) {
-  size_t n = 1000;
+  printf("\n");
+  size_t n = 10000;
   int *p1 = gc(malloc(n * sizeof(int)));
   int *p2 = gc(malloc(n * sizeof(int)));
   rngset(p1, n * sizeof(int), 0, 0);
   EZBENCH2("_intsort", memcpy(p2, p1, n * sizeof(int)), _intsort(p2, n));
-  EZBENCH2("qsort", memcpy(p2, p1, n * sizeof(int)),
+  if (X86_HAVE(AVX2)) {
+    EZBENCH2("vqsort_int32_avx2", memcpy(p2, p1, n * sizeof(int)),
+             vqsort_int32_avx2(p2, n));
+  }
+  if (X86_HAVE(SSE4_2)) {
+    EZBENCH2("vqsort_int32_sse4", memcpy(p2, p1, n * sizeof(int)),
+             vqsort_int32_sse4(p2, n));
+  }
+  if (X86_HAVE(SSSE3)) {
+    EZBENCH2("vqsort_int32_ssse3", memcpy(p2, p1, n * sizeof(int)),
+             vqsort_int32_ssse3(p2, n));
+  }
+  EZBENCH2("vqsort_int32_sse2", memcpy(p2, p1, n * sizeof(int)),
+           vqsort_int32_sse2(p2, n));
+  EZBENCH2("djbsort", memcpy(p2, p1, n * sizeof(int)), djbsort(p2, n));
+  EZBENCH2("radix_sort_int32", memcpy(p2, p1, n * sizeof(int)),
+           radix_sort_int32(p2, n));
+  EZBENCH2("qsort(int)", memcpy(p2, p1, n * sizeof(int)),
            qsort(p2, n, sizeof(int), CompareInt));
 }

test/libc/str/test.mk

@@ -51,7 +51,8 @@ TEST_LIBC_STR_DIRECTDEPS =						\
 	THIRD_PARTY_REGEX						\
 	THIRD_PARTY_ZLIB						\
 	THIRD_PARTY_LIBCXX						\
-	THIRD_PARTY_SMALLZ4
+	THIRD_PARTY_SMALLZ4						\
+	THIRD_PARTY_VQSORT
 
 TEST_LIBC_STR_DEPS :=							\
 	$(call uniq,$(foreach x,$(TEST_LIBC_STR_DIRECTDEPS),$($(x))))

third_party/compiler_rt/popcountdi2.c

@@ -22,6 +22,10 @@ STATIC_YOINK("huge_compiler_rt_license");
 COMPILER_RT_ABI si_int
 __popcountdi2(di_int a)
 {
+#ifdef __POPCNT__
+    asm("popcnt\t%1,%0" : "=r"(a) : "r"(a) : "cc");
+    return a;
+#else
     du_int x2 = (du_int)a;
     x2 = x2 - ((x2 >> 1) & 0x5555555555555555uLL);
     /* Every 2 bits holds the sum of every pair of bits (32) */
@@ -36,4 +40,5 @@ __popcountdi2(di_int a)
     /* The lower 16 bits hold two 32 bit sums (6 significant bits). */
     /*   Upper 16 bits are garbage */
     return (x + (x >> 8)) & 0x0000007F;  /* (7 significant bits) */
+#endif
 }

third_party/third_party.mk

@@ -34,6 +34,7 @@ o/$(MODE)/third_party:				\
 	o/$(MODE)/third_party/tidy		\
 	o/$(MODE)/third_party/tr		\
 	o/$(MODE)/third_party/unzip		\
+	o/$(MODE)/third_party/vqsort		\
 	o/$(MODE)/third_party/xed		\
 	o/$(MODE)/third_party/zip		\
 	o/$(MODE)/third_party/zlib

third_party/vqsort/README.cosmo

@@ -0,0 +1,23 @@
+DESCRIPTION
+
+  vqsort implements vectorized quicksort using avx2. this is the fastest
+  way to sort integers. this goes as fast as djbsort for 32-bit integers
+  except it supports 64-bit integers too, which go just as fast: about a
+  gigabyte of memory sorted per second. It's 3x faster than simple radix
+  sort. It's 5x faster than simple quicksort. It's 10x faster than qsort
+
+LICENSE
+
+  Apache 2.o
+
+ORIGIN
+
+  https://github.com/google/highway/
+  commit 50331e0523bbf5f6c94b94263a91680f118e0986
+  Author: Jan Wassenberg <janwas@google.com>
+  Date:   Wed Apr 26 11:20:33 2023 -0700
+  Faster vqsort for small arrays (7x speedup! for N=100)
+
+LOCAL CHANGES
+
+  Precompiled beacuse upstream codebase is slow, gigantic, and hairy.

third_party/vqsort/vqsort.h

@@ -0,0 +1,20 @@
+#ifndef COSMOPOLITAN_THIRD_PARTY_VQSORT_H_
+#define COSMOPOLITAN_THIRD_PARTY_VQSORT_H_
+#if !(__ASSEMBLER__ + __LINKER__ + 0)
+COSMOPOLITAN_C_START_
+
+void vqsort_int64(int64_t *, size_t);
+void vqsort_int64_avx2(int64_t *, size_t);
+void vqsort_int64_sse4(int64_t *, size_t);
+void vqsort_int64_ssse3(int64_t *, size_t);
+void vqsort_int64_sse2(int64_t *, size_t);
+
+void vqsort_int32(int32_t *, size_t);
+void vqsort_int32_avx2(int32_t *, size_t);
+void vqsort_int32_sse4(int32_t *, size_t);
+void vqsort_int32_ssse3(int32_t *, size_t);
+void vqsort_int32_sse2(int32_t *, size_t);
+
+COSMOPOLITAN_C_END_
+#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
+#endif /* COSMOPOLITAN_THIRD_PARTY_VQSORT_H_ */

third_party/vqsort/vqsort.mk

@@ -0,0 +1,52 @@
+#-*-mode:makefile-gmake;indent-tabs-mode:t;tab-width:8;coding:utf-8-*-┐
+#───vi: set et ft=make ts=8 tw=8 fenc=utf-8 :vi───────────────────────┘
+
+PKGS += THIRD_PARTY_VQSORT
+
+THIRD_PARTY_VQSORT_ARTIFACTS += THIRD_PARTY_VQSORT_A
+THIRD_PARTY_VQSORT = $(THIRD_PARTY_VQSORT_A_DEPS) $(THIRD_PARTY_VQSORT_A)
+THIRD_PARTY_VQSORT_A = o/$(MODE)/third_party/vqsort/vqsort.a
+THIRD_PARTY_VQSORT_A_FILES := $(wildcard third_party/vqsort/*)
+THIRD_PARTY_VQSORT_A_HDRS = $(filter %.h,$(THIRD_PARTY_VQSORT_A_FILES))
+THIRD_PARTY_VQSORT_A_SRCS_C = $(filter %.c,$(THIRD_PARTY_VQSORT_A_FILES))
+THIRD_PARTY_VQSORT_A_SRCS_S = $(filter %.S,$(THIRD_PARTY_VQSORT_A_FILES))
+THIRD_PARTY_VQSORT_A_SRCS = $(THIRD_PARTY_VQSORT_A_SRCS_C) $(THIRD_PARTY_VQSORT_A_SRCS_S)
+THIRD_PARTY_VQSORT_A_OBJS_C = $(THIRD_PARTY_VQSORT_A_SRCS_C:%.c=o/$(MODE)/%.o)
+THIRD_PARTY_VQSORT_A_OBJS_S = $(THIRD_PARTY_VQSORT_A_SRCS_S:%.S=o/$(MODE)/%.o)
+THIRD_PARTY_VQSORT_A_OBJS = $(THIRD_PARTY_VQSORT_A_OBJS_C) $(THIRD_PARTY_VQSORT_A_OBJS_S)
+
+THIRD_PARTY_VQSORT_A_CHECKS =			\
+	$(THIRD_PARTY_VQSORT_A).pkg		\
+	$(THIRD_PARTY_VQSORT_A_HDRS:%=o/$(MODE)/%.ok)
+
+THIRD_PARTY_VQSORT_A_DIRECTDEPS =		\
+	LIBC_INTRIN				\
+	LIBC_MEM				\
+	LIBC_NEXGEN32E				\
+	LIBC_RUNTIME				\
+	LIBC_STDIO				\
+	LIBC_STR				\
+	LIBC_STUBS				\
+	THIRD_PARTY_COMPILER_RT
+
+THIRD_PARTY_VQSORT_A_DEPS :=			\
+	$(call uniq,$(foreach x,$(THIRD_PARTY_VQSORT_A_DIRECTDEPS),$($(x))))
+
+$(THIRD_PARTY_VQSORT_A):			\
+		third_party/vqsort/		\
+		$(THIRD_PARTY_VQSORT_A).pkg	\
+		$(THIRD_PARTY_VQSORT_A_OBJS)
+
+$(THIRD_PARTY_VQSORT_A).pkg:			\
+		$(THIRD_PARTY_VQSORT_A_OBJS)	\
+		$(foreach x,$(THIRD_PARTY_VQSORT_A_DIRECTDEPS),$($(x)_A).pkg)
+
+THIRD_PARTY_VQSORT_LIBS = $(foreach x,$(THIRD_PARTY_VQSORT_ARTIFACTS),$($(x)))
+THIRD_PARTY_VQSORT_SRCS = $(foreach x,$(THIRD_PARTY_VQSORT_ARTIFACTS),$($(x)_SRCS))
+THIRD_PARTY_VQSORT_HDRS = $(foreach x,$(THIRD_PARTY_VQSORT_ARTIFACTS),$($(x)_HDRS))
+THIRD_PARTY_VQSORT_CHECKS = $(foreach x,$(THIRD_PARTY_VQSORT_ARTIFACTS),$($(x)_CHECKS))
+THIRD_PARTY_VQSORT_OBJS = $(foreach x,$(THIRD_PARTY_VQSORT_ARTIFACTS),$($(x)_OBJS))
+$(THIRD_PARTY_VQSORT_OBJS): $(BUILD_FILES) third_party/vqsort/vqsort.mk
+
+.PHONY: o/$(MODE)/third_party/vqsort
+o/$(MODE)/third_party/vqsort: $(THIRD_PARTY_VQSORT_CHECKS)

third_party/vqsort/vqsort_int32.c

@@ -0,0 +1,29 @@
+/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
+│vi: set net ft=c ts=2 sts=2 sw=2 fenc=utf-8                                :vi│
+╞══════════════════════════════════════════════════════════════════════════════╡
+│ Copyright 2023 Justine Alexandra Roberts Tunney                              │
+│                                                                              │
+│ Permission to use, copy, modify, and/or distribute this software for         │
+│ any purpose with or without fee is hereby granted, provided that the         │
+│ above copyright notice and this permission notice appear in all copies.      │
+│                                                                              │
+│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL                │
+│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED                │
+│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE             │
+│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL         │
+│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR        │
+│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER               │
+│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
+│ PERFORMANCE OF THIS SOFTWARE.                                                │
+╚─────────────────────────────────────────────────────────────────────────────*/
+#include "libc/mem/alg.h"
+#include "libc/nexgen32e/x86feature.h"
+#include "third_party/vqsort/vqsort.h"
+
+void vqsort_int32(int32_t *A, size_t n) {
+  if (X86_HAVE(AVX2)) {
+    vqsort_int32_avx2(A, n);
+  } else {
+    radix_sort_int32(A, n);
+  }
+}

third_party/vqsort/vqsort_int64.c

@@ -0,0 +1,29 @@
+/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
+│vi: set net ft=c ts=2 sts=2 sw=2 fenc=utf-8                                :vi│
+╞══════════════════════════════════════════════════════════════════════════════╡
+│ Copyright 2023 Justine Alexandra Roberts Tunney                              │
+│                                                                              │
+│ Permission to use, copy, modify, and/or distribute this software for         │
+│ any purpose with or without fee is hereby granted, provided that the         │
+│ above copyright notice and this permission notice appear in all copies.      │
+│                                                                              │
+│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL                │
+│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED                │
+│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE             │
+│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL         │
+│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR        │
+│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER               │
+│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
+│ PERFORMANCE OF THIS SOFTWARE.                                                │
+╚─────────────────────────────────────────────────────────────────────────────*/
+#include "libc/mem/alg.h"
+#include "libc/nexgen32e/x86feature.h"
+#include "third_party/vqsort/vqsort.h"
+
+void vqsort_int64(int64_t *A, size_t n) {
+  if (X86_HAVE(AVX2)) {
+    vqsort_int64_avx2(A, n);
+  } else {
+    radix_sort_int64(A, n);
+  }
+}