From 1931a78291b0e3aa28887ba17d74fc3c9914215d Mon Sep 17 00:00:00 2001
From: goldsteinn <35538541+goldsteinn@users.noreply.github.com>
Date: Tue, 10 Sep 2024 07:05:20 -0700
Subject: [PATCH] clarify bitset_extract_setbits scalar code (BLSI -> BLSR)
 (#659)

The "BLSI" logic is essentially a complicated way to get to BLSR, so
refactor to be more explicit.
---
 src/bitset_util.c | 31 ++++++++-----------------------
 1 file changed, 8 insertions(+), 23 deletions(-)

diff --git a/src/bitset_util.c b/src/bitset_util.c
index 0ae7d9258..6bc12b44b 100644
--- a/src/bitset_util.c
+++ b/src/bitset_util.c
@@ -613,16 +613,13 @@ size_t bitset_extract_setbits_avx512(const uint64_t *words, size_t length,
     for (; (i < length) && (out < safeout); ++i) {
         uint64_t w = words[i];
         while ((w != 0) && (out < safeout)) {
-            uint64_t t =
-                w & (~w + 1);  // on x64, should compile to BLSI (careful: the
-                               // Intel compiler seems to fail)
             int r =
                 roaring_trailing_zeroes(w);  // on x64, should compile to TZCNT
             uint32_t val = r + base;
             memcpy(out, &val,
                    sizeof(uint32_t));  // should be compiled as a MOV on x64
             out++;
-            w ^= t;
+            w &= (w - 1);
         }
         base += 64;
     }
@@ -667,15 +664,12 @@ size_t bitset_extract_setbits_avx512_uint16(const uint64_t *array,
     for (; (i < length) && (out < safeout); ++i) {
         uint64_t w = array[i];
         while ((w != 0) && (out < safeout)) {
-            uint64_t t =
-                w & (~w + 1);  // on x64, should compile to BLSI (careful: the
-                               // Intel compiler seems to fail)
             int r =
                 roaring_trailing_zeroes(w);  // on x64, should compile to TZCNT
             uint32_t val = r + base;
             memcpy(out, &val, sizeof(uint16_t));
             out++;
-            w ^= t;
+            w &= (w - 1);
         }
         base += 64;
     }
@@ -725,16 +719,13 @@ size_t bitset_extract_setbits_avx2(const uint64_t *words, size_t length,
     for (; (i < length) && (out < safeout); ++i) {
         uint64_t w = words[i];
         while ((w != 0) && (out < safeout)) {
-            uint64_t t =
-                w & (~w + 1);  // on x64, should compile to BLSI (careful: the
-                               // Intel compiler seems to fail)
             int r =
                 roaring_trailing_zeroes(w);  // on x64, should compile to TZCNT
             uint32_t val = r + base;
             memcpy(out, &val,
                    sizeof(uint32_t));  // should be compiled as a MOV on x64
             out++;
-            w ^= t;
+            w &= (w - 1);
         }
         base += 64;
     }
@@ -749,16 +740,13 @@ size_t bitset_extract_setbits(const uint64_t *words, size_t length,
     for (size_t i = 0; i < length; ++i) {
         uint64_t w = words[i];
         while (w != 0) {
-            uint64_t t =
-                w & (~w + 1);  // on x64, should compile to BLSI (careful: the
-                               // Intel compiler seems to fail)
             int r =
                 roaring_trailing_zeroes(w);  // on x64, should compile to TZCNT
             uint32_t val = r + base;
             memcpy(out + outpos, &val,
                    sizeof(uint32_t));  // should be compiled as a MOV on x64
             outpos++;
-            w ^= t;
+            w &= (w - 1);
         }
         base += 64;
     }
@@ -772,10 +760,9 @@ size_t bitset_extract_intersection_setbits_uint16(
     for (size_t i = 0; i < length; ++i) {
         uint64_t w = words1[i] & words2[i];
         while (w != 0) {
-            uint64_t t = w & (~w + 1);
             int r = roaring_trailing_zeroes(w);
             out[outpos++] = (uint16_t)(r + base);
-            w ^= t;
+            w &= (w - 1);
         }
         base += 64;
     }
@@ -836,11 +823,10 @@ size_t bitset_extract_setbits_sse_uint16(const uint64_t *words, size_t length,
     for (; (i < length) && (out < safeout); ++i) {
         uint64_t w = words[i];
         while ((w != 0) && (out < safeout)) {
-            uint64_t t = w & (~w + 1);
             int r = roaring_trailing_zeroes(w);
             *out = (uint16_t)(r + base);
             out++;
-            w ^= t;
+            w &= (w - 1);
         }
         base += 64;
     }
@@ -864,10 +850,9 @@ size_t bitset_extract_setbits_uint16(const uint64_t *words, size_t length,
     for (size_t i = 0; i < length; ++i) {
         uint64_t w = words[i];
         while (w != 0) {
-            uint64_t t = w & (~w + 1);
             int r = roaring_trailing_zeroes(w);
             out[outpos++] = (uint16_t)(r + base);
-            w ^= t;
+            w &= (w - 1);
         }
         base += 64;
     }
@@ -1158,4 +1143,4 @@ void bitset_flip_list(uint64_t *words, const uint16_t *list, uint64_t length) {
 #endif
 #if defined(__GNUC__) && !defined(__clang__)
 #pragma GCC diagnostic pop
-#endif
\ No newline at end of file
+#endif