brush up AVX optimization in popcount

tomoaki0705 · tomoaki0705 · commit 820fdbff2309 · 2017-01-26T18:37:56.000+09:00
- make sure SIMD optimization works even when AVX is not available
diff --git a/modules/core/src/stat.cpp b/modules/core/src/stat.cpp
@@ -4261,7 +4261,9 @@ int normHamming(const uchar* a, int n)
         _r0 = _mm256_add_epi32(_r0, _mm256_shuffle_epi32(_r0, 2));
         result = _mm256_extract_epi32_(_mm256_add_epi32(_r0, _mm256_permute2x128_si256(_r0, _r0, 1)), 0);
     }
-#elif CV_POPCNT
+#endif // CV_AVX2
+
+#if CV_POPCNT
     if(checkHardwareSupport(CV_CPU_POPCNT))
     {
 #  if defined CV_POPCNT_U64
@@ -4275,17 +4277,20 @@ int normHamming(const uchar* a, int n)
             result += CV_POPCNT_U32(*(uint*)(a + i));
         }
     }
-#elif CV_SIMD128
+#endif // CV_POPCNT
+
+#if CV_SIMD128
     if(hasSIMD128())
     {
         v_uint32x4 t = v_setzero_u32();
         for(; i <= n - v_uint8x16::nlanes; i += v_uint8x16::nlanes)
         {
             t += v_popcount(v_load(a + i));
         }
-        result = v_reduce_sum(t);
+        result += v_reduce_sum(t);
     }
-#endif
+#endif // CV_SIMD128
+
     for(; i <= n - 4; i += 4)
     {
         result += popCountTable[a[i]] + popCountTable[a[i+1]] +
@@ -4327,7 +4332,9 @@ int normHamming(const uchar* a, const uchar* b, int n)
         _r0 = _mm256_add_epi32(_r0, _mm256_shuffle_epi32(_r0, 2));
         result = _mm256_extract_epi32_(_mm256_add_epi32(_r0, _mm256_permute2x128_si256(_r0, _r0, 1)), 0);
     }
-#elif CV_POPCNT
+#endif // CV_AVX2
+
+#if CV_POPCNT
     if(checkHardwareSupport(CV_CPU_POPCNT))
     {
 #  if defined CV_POPCNT_U64
@@ -4341,17 +4348,20 @@ int normHamming(const uchar* a, const uchar* b, int n)
             result += CV_POPCNT_U32(*(uint*)(a + i) ^ *(uint*)(b + i));
         }
     }
-#elif CV_SIMD128
+#endif // CV_POPCNT
+
+#if CV_SIMD128
     if(hasSIMD128())
     {
         v_uint32x4 t = v_setzero_u32();
         for(; i <= n - v_uint8x16::nlanes; i += v_uint8x16::nlanes)
         {
             t += v_popcount(v_load(a + i) ^ v_load(b + i));
         }
-        result = v_reduce_sum(t);
+        result += v_reduce_sum(t);
     }
-#endif
+#endif // CV_SIMD128
+
     for(; i <= n - 4; i += 4)
     {
         result += popCountTable[a[i] ^ b[i]] + popCountTable[a[i+1] ^ b[i+1]] +

Original file line number	Diff line number	Diff line change
`@@ -4261,7 +4261,9 @@ int normHamming(const uchar* a, int n)`
`4261`	`4261`	`_r0 = _mm256_add_epi32(_r0, _mm256_shuffle_epi32(_r0, 2));`
`4262`	`4262`	`result = _mm256_extract_epi32_(_mm256_add_epi32(_r0, _mm256_permute2x128_si256(_r0, _r0, 1)), 0);`
`4263`	`4263`	`}`
`4264`		`-#elif CV_POPCNT`
	`4264`	`+#endif // CV_AVX2`
	`4265`	`+`
	`4266`	`+#if CV_POPCNT`
`4265`	`4267`	`if(checkHardwareSupport(CV_CPU_POPCNT))`
`4266`	`4268`	`{`
`4267`	`4269`	`# if defined CV_POPCNT_U64`
`@@ -4275,17 +4277,20 @@ int normHamming(const uchar* a, int n)`
`4275`	`4277`	`result += CV_POPCNT_U32((uint)(a + i));`
`4276`	`4278`	`}`
`4277`	`4279`	`}`
`4278`		`-#elif CV_SIMD128`
	`4280`	`+#endif // CV_POPCNT`
	`4281`	`+`
	`4282`	`+#if CV_SIMD128`
`4279`	`4283`	`if(hasSIMD128())`
`4280`	`4284`	`{`
`4281`	`4285`	`v_uint32x4 t = v_setzero_u32();`
`4282`	`4286`	`for(; i <= n - v_uint8x16::nlanes; i += v_uint8x16::nlanes)`
`4283`	`4287`	`{`
`4284`	`4288`	`t += v_popcount(v_load(a + i));`
`4285`	`4289`	`}`
`4286`		`- result = v_reduce_sum(t);`
	`4290`	`+ result += v_reduce_sum(t);`
`4287`	`4291`	`}`
`4288`		`-#endif`
	`4292`	`+#endif // CV_SIMD128`
	`4293`	`+`
`4289`	`4294`	`for(; i <= n - 4; i += 4)`
`4290`	`4295`	`{`
`4291`	`4296`	`result += popCountTable[a[i]] + popCountTable[a[i+1]] +`
`@@ -4327,7 +4332,9 @@ int normHamming(const uchar* a, const uchar* b, int n)`
`4327`	`4332`	`_r0 = _mm256_add_epi32(_r0, _mm256_shuffle_epi32(_r0, 2));`
`4328`	`4333`	`result = _mm256_extract_epi32_(_mm256_add_epi32(_r0, _mm256_permute2x128_si256(_r0, _r0, 1)), 0);`
`4329`	`4334`	`}`
`4330`		`-#elif CV_POPCNT`
	`4335`	`+#endif // CV_AVX2`
	`4336`	`+`
	`4337`	`+#if CV_POPCNT`
`4331`	`4338`	`if(checkHardwareSupport(CV_CPU_POPCNT))`
`4332`	`4339`	`{`
`4333`	`4340`	`# if defined CV_POPCNT_U64`
`@@ -4341,17 +4348,20 @@ int normHamming(const uchar* a, const uchar* b, int n)`
`4341`	`4348`	`result += CV_POPCNT_U32((uint)(a + i) ^ (uint)(b + i));`
`4342`	`4349`	`}`
`4343`	`4350`	`}`
`4344`		`-#elif CV_SIMD128`
	`4351`	`+#endif // CV_POPCNT`
	`4352`	`+`
	`4353`	`+#if CV_SIMD128`
`4345`	`4354`	`if(hasSIMD128())`
`4346`	`4355`	`{`
`4347`	`4356`	`v_uint32x4 t = v_setzero_u32();`
`4348`	`4357`	`for(; i <= n - v_uint8x16::nlanes; i += v_uint8x16::nlanes)`
`4349`	`4358`	`{`
`4350`	`4359`	`t += v_popcount(v_load(a + i) ^ v_load(b + i));`
`4351`	`4360`	`}`
`4352`		`- result = v_reduce_sum(t);`
	`4361`	`+ result += v_reduce_sum(t);`
`4353`	`4362`	`}`
`4354`		`-#endif`
	`4363`	`+#endif // CV_SIMD128`
	`4364`	`+`
`4355`	`4365`	`for(; i <= n - 4; i += 4)`
`4356`	`4366`	`{`
`4357`	`4367`	`result += popCountTable[a[i] ^ b[i]] + popCountTable[a[i+1] ^ b[i+1]] +`