const ocl::Device & dev = ocl::Device::getDefault();
bool doubleSupport = dev.doubleFPConfig() > 0;
- if ((!doubleSupport && depth == CV_64F))
+ if (!doubleSupport && depth == CV_64F)
return false;
bool haveC = matC.kind() != cv::_InputArray::NONE;
Size sizeD(sizeB.width, sizeA.height);
- CV_Assert( matB.type() == type && (!haveC || matC.type() == type) );
+ CV_Assert( !haveC || matC.type() == type );
CV_Assert( sizeA.width == sizeB.height && (!haveC || sizeC == sizeD) );
int max_wg_size = (int)dev.maxWorkGroupSize();
if (haveC)
ctrans ? transpose(matC, D) : matC.copyTo(D);
- else
- D.setTo(Scalar::all(0));
int vectorWidths[] = { 4, 4, 2, 2, 1, 4, cn, -1 };
-
int kercn = ocl::checkOptimalVectorWidth(vectorWidths, B, D);
String opts = format("-D T=%s -D T1=%s -D WT=%s -D cn=%d -D kercn=%d -D LOCAL_SIZE=%d %s %s %s",
#if cn==2
#if kercn==2
-#define MUL(i, a, b)\
+#define MUL(a, b)\
{\
sum.x += fma(a.x, b.x, - a.y * b.y);\
sum.y += fma(a.x, b.y, a.y * b.x);\
}
#else
-#define MUL(i, a, b)\
+#define MUL(a, b)\
{\
sum.x += fma(a.x, b.x, - a.y * b.y);\
sum.y += fma(a.x, b.y, a.y * b.x);\
}
#endif
#else
-#define MUL(i, a, b) sum = fma(a, b, sum);
+#define MUL(a, b) sum = fma(a, b, sum);
#endif
if (x < D_cols && y < D_rows)
{
for (int i = 0; i < n; ++i)
- MUL(i, A[i], B[i*STEP_B]);
+ MUL(A[i], B[i*STEP_B]);
#else
__local T a_local[LOCAL_SIZE*LOCAL_SIZE];
if (x < D_cols && y < D_rows)
{
- for (int i = 0; i < LOCAL_SIZE
#if NO_MULT
- && p * LOCAL_SIZE + i < n
+ int ie = min(LOCAL_SIZE, n - p * LOCAL_SIZE);
+ for (int i = 0; i < ie; ++i)
+#else
+ for (int i = 0; i < LOCAL_SIZE; ++i)
#endif
- ; ++i)
- MUL(i, a_local[mad24(lidy, LOCAL_SIZE, i)], b_local[mad24(i, LOCAL_SIZE, lidx)]);
+ MUL(a_local[mad24(lidy, LOCAL_SIZE, i)], b_local[mad24(i, LOCAL_SIZE, lidx)]);
}
-
barrier(CLK_LOCAL_MEM_FENCE);
}