From f0ec26992a5f8f4dc1af97a2a0614ef5ba613b87 Mon Sep 17 00:00:00 2001
From: John Keiser <john@johnkeiser.com>
Date: Tue, 1 Sep 2020 08:37:46 -0700
Subject: [PATCH] Remove bit_or (bad perf on Windows)

---
 src/arm64/simd.h                           | 10 ----------
 src/haswell/dom_parser_implementation.cpp  |  6 +++++-
 src/haswell/simd.h                         |  2 +-
 src/westmere/dom_parser_implementation.cpp |  8 +++++++-
 src/westmere/simd.h                        | 22 ++++++----------------
 5 files changed, 19 insertions(+), 29 deletions(-)

diff --git a/src/arm64/simd.h b/src/arm64/simd.h
index 6ece602a..f59cd1ae 100644
--- a/src/arm64/simd.h
+++ b/src/arm64/simd.h
@@ -462,16 +462,6 @@ simdjson_really_inline int8x16_t make_int8x16_t(int8_t x1,  int8_t x2,  int8_t x
       return vgetq_lane_u64(vreinterpretq_u64_u8(sum0), 0);
     }
 
-    simdjson_really_inline simd8x64<T> bit_or(const T m) const {
-      const simd8<T> mask = simd8<T>::splat(m);
-      return simd8x64<T>(
-        this->chunks[0] | mask,
-        this->chunks[1] | mask,
-        this->chunks[2] | mask,
-        this->chunks[3] | mask
-      );
-    }
-
     simdjson_really_inline uint64_t eq(const T m) const {
       const simd8<T> mask = simd8<T>::splat(m);
       return  simd8x64<bool>(
diff --git a/src/haswell/dom_parser_implementation.cpp b/src/haswell/dom_parser_implementation.cpp
index 623969c9..b2104967 100644
--- a/src/haswell/dom_parser_implementation.cpp
+++ b/src/haswell/dom_parser_implementation.cpp
@@ -70,7 +70,11 @@ simdjson_really_inline json_character_block json_character_block::classify(const
     _mm256_shuffle_epi8(whitespace_table, in.chunks[0]),
     _mm256_shuffle_epi8(whitespace_table, in.chunks[1])
   });
-  const simd8x64<uint8_t> curlified = in.bit_or(0x20); // Turn [ and ] into { and }
+  // Turn [ and ] into { and }
+  const simd8x64<uint8_t> curlified{
+    in.chunks[0] | 0x20,
+    in.chunks[1] | 0x20
+  };
   const uint64_t op = curlified.eq({
     _mm256_shuffle_epi8(op_table, in.chunks[0]),
     _mm256_shuffle_epi8(op_table, in.chunks[1])
diff --git a/src/haswell/simd.h b/src/haswell/simd.h
index 93491408..276bd3a1 100644
--- a/src/haswell/simd.h
+++ b/src/haswell/simd.h
@@ -337,7 +337,7 @@ namespace simd {
       ).to_bitmask();
     }
 
-    simdjson_really_inline uint64_t eq(const simd8x64<uint8_t> other) const {
+    simdjson_really_inline uint64_t eq(const simd8x64<uint8_t> &other) const {
       return  simd8x64<bool>(
         this->chunks[0] == other.chunks[0],
         this->chunks[1] == other.chunks[1]
diff --git a/src/westmere/dom_parser_implementation.cpp b/src/westmere/dom_parser_implementation.cpp
index 5a174b00..5518ba52 100644
--- a/src/westmere/dom_parser_implementation.cpp
+++ b/src/westmere/dom_parser_implementation.cpp
@@ -65,7 +65,13 @@ simdjson_really_inline json_character_block json_character_block::classify(const
     _mm_shuffle_epi8(whitespace_table, in.chunks[2]),
     _mm_shuffle_epi8(whitespace_table, in.chunks[3])
   });
-  const simd8x64<uint8_t> curlified = in.bit_or(0x20); // Turn [ and ] into { and }
+  // Turn [ and ] into { and }
+  const simd8x64<uint8_t> curlified{
+    in.chunks[0] | 0x20,
+    in.chunks[1] | 0x20,
+    in.chunks[2] | 0x20,
+    in.chunks[3] | 0x20
+  };
   const uint64_t op = curlified.eq({
     _mm_shuffle_epi8(op_table, in.chunks[0]),
     _mm_shuffle_epi8(op_table, in.chunks[1]),
diff --git a/src/westmere/simd.h b/src/westmere/simd.h
index 5b92d980..c321586b 100644
--- a/src/westmere/simd.h
+++ b/src/westmere/simd.h
@@ -288,23 +288,13 @@ namespace simd {
     }
 
     simdjson_really_inline uint64_t to_bitmask() const {
-      uint64_t r0 = uint32_t(this->chunks[0].to_bitmask());
-      uint64_t r1 =                       this->chunks[1].to_bitmask();
-      uint64_t r2 =                       this->chunks[2].to_bitmask();
-      uint64_t r3 =                       this->chunks[3].to_bitmask();
+      uint64_t r0 = uint32_t(this->chunks[0].to_bitmask() );
+      uint64_t r1 =          this->chunks[1].to_bitmask() ;
+      uint64_t r2 =          this->chunks[2].to_bitmask() ;
+      uint64_t r3 =          this->chunks[3].to_bitmask() ;
       return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48);
     }
-
-    simdjson_really_inline simd8x64<T> bit_or(const T m) const {
-      const simd8<T> mask = simd8<T>::splat(m);
-      return simd8x64<T>(
-        this->chunks[0] | mask,
-        this->chunks[1] | mask,
-        this->chunks[2] | mask,
-        this->chunks[3] | mask
-      );
-    }
-
+    
     simdjson_really_inline uint64_t eq(const T m) const {
       const simd8<T> mask = simd8<T>::splat(m);
       return  simd8x64<bool>(
@@ -315,7 +305,7 @@ namespace simd {
       ).to_bitmask();
     }
 
-    simdjson_really_inline uint64_t eq(const simd8x64<uint8_t> other) const {
+    simdjson_really_inline uint64_t eq(const simd8x64<uint8_t> &other) const {
       return  simd8x64<bool>(
         this->chunks[0] == other.chunks[0],
         this->chunks[1] == other.chunks[1],