daily update on 20240206 by nitload youmu, token parser change to use regex generator

内核源代码/StgSharp.C/StgSharpC.h

@@ -36,6 +36,7 @@ extern "C" {
 #endif
 
 #define byte char
+#define cs_char uint16_t
 #define ssc_null_assert(ptr) assert(ptr != NULL)
 
 #define ALIGN(simdVec) _mm_loadu_ps(&simdVec)

内核源代码/StgSharp.C/StgSharpC.vcxproj

@@ -171,6 +171,7 @@
     <ClCompile Include="stb_image.c" />
     <ClCompile Include="StgSharpC.c" />
     <ClCompile Include="hash.c" />
+    <ClCompile Include="string.c" />
     <ClCompile Include="vector.c" />
     <ClCompile Include="vector_dot.c" />
   </ItemGroup>

内核源代码/StgSharp.C/hash.c

@@ -1,9 +1,9 @@
 #include "ssc_intrinsic.h"
 #include "StgSharpC.h"
 
-static inline uint32_t fmix32(uint32_t h);
+static inline uint32_t fmix32(uint32_t hash);
 
-INTERNAL int SSCDECL city_hash_simplify(byte const *str, int const length)
+INTERNAL int SSCDECL city_hash_simplify_sse(byte const *str, int const length)
 {
         uint32_t hash = length, size = length << 1;
         byte *end = str + size, *ptr = str;
@@ -37,15 +37,10 @@ INTERNAL int SSCDECL city_hash_simplify(byte const *str, int const length)
                         hash = _mm_crc32_u8(hash,*ptr);
                 }
         }
+        hash ^= hash >> 16;
+        hash *= 0x85ebca6b;
+        hash ^= hash >> 13;
+        hash *= 0xc2b2ae35;
+        hash ^= hash >> 16;
         return fmix32(hash);
 }
-
-INTERNAL inline uint32_t fmix32(uint32_t h)
-{
-        h ^= h >> 16;
-        h *= 0x85ebca6b;
-        h ^= h >> 13;
-        h *= 0xc2b2ae35;
-        h ^= h >> 16;
-        return h;
-}

内核源代码/StgSharp.C/intrinsic.c

@@ -82,7 +82,8 @@ void load_intrinsic_sse(ssc_intrinsic *context)
         context->dot_41 = dot_41_sse;
         context->dot_42 = dot_42_sse;
         context->dot_43 = dot_43_sse;
-        context->city_hash_simplify = city_hash_simplify;
+        context->city_hash_simplify = city_hash_simplify_sse;
+        context->index_pair = index_pair_sse;
 }
 
 void load_intrinsic_avx(ssc_intrinsic *context)
@@ -106,7 +107,8 @@ void load_intrinsic_avx(ssc_intrinsic *context)
         context->dot_41 = dot_41_avx;
         context->dot_42 = dot_42_avx;
         context->dot_43 = dot_43_avx;
-        context->city_hash_simplify = city_hash_simplify;
+        context->city_hash_simplify = city_hash_simplify_sse;
+        context->index_pair = index_pair_sse;
 }
 
 void load_intrinsic_512(ssc_intrinsic *context)
@@ -130,5 +132,6 @@ void load_intrinsic_512(ssc_intrinsic *context)
         context->dot_41 = dot_41_avx;
         context->dot_42 = dot_42_avx;
         context->dot_43 = dot_43_avx;
-        context->city_hash_simplify = city_hash_simplify;
+        context->city_hash_simplify = city_hash_simplify_sse;
+        context->index_pair = index_pair_sse;
 }

内核源代码/StgSharp.C/ssc_intrinsic.h

@@ -103,9 +103,10 @@ INTERNAL void SSCDECL dot_43_512(__4_columnset *transpose, __m128 *vector, __m12
 
 #pragma endregion
 
-#pragma region quick_hash
+#pragma region string
 
-INTERNAL int city_hash_simplify(byte const *str, int length);
+INTERNAL int SSCDECL city_hash_simplify_sse(byte const *str, int length);
+INTERNAL int SSCDECL index_pair_sse(byte const *str, uint32_t target, int length);
 
 #pragma endregion
 
@@ -121,6 +122,7 @@ typedef struct ssc_intrinsic {
         DOTPROC dot_41;
         DOTPROC dot_42;
         DOTPROC dot_43;
+        int (*index_pair)(byte *str, byte *target, int length, int start);
         VECTORNORMALIZEPROC normalize_3;
         MATRIXTRANSPOSEPROC transpose_23;
         MATRIXTRANSPOSEPROC transpose_24;

内核源代码/StgSharp.C/string.c

@@ -0,0 +1,31 @@
+#include "StgSharpC.h"
+#include "ssc_intrinsic.h"
+
+int index_pair_sse(byte const *str, uint32_t target, int length)
+{
+        int index = 0, result = 0, size = length << 1, end = size - 14;
+        __m128i source = { .m128i_i32 = {
+                                   target,
+                                   target,
+                                   target,
+                                   target,
+                           } };
+        for (; (index < end) && (!result); index += 14) { // move 7*2 bytes instead of 8*2
+                __m128i vec = _mm_loadu_epi32(str + index);
+                __m128i com = _mm_cmpeq_epi32(vec, source);
+                result = _mm_movemask_epi8(com);
+        }
+        if (result) {
+                for (; result; result >> 2) {
+                        index++;
+                }
+                return index;
+        }
+        for (; (index < size - 2) && (!result); index += 2) {
+                result = *(int *)(str + index) == target;
+        }
+        if (result) {
+                return index;
+        }
+        return -1;
+}