aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl')
-rw-r--r--src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl4
1 files changed, 4 insertions, 0 deletions
diff --git a/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl b/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl
index 8a8458798e..345469063a 100644
--- a/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl
+++ b/src/core/CL/cl_kernels/nhwc/dwc_native_fp_nhwc.cl
@@ -158,7 +158,11 @@ __kernel void dwc_native_fp_nhwc(
{
LOOP_UNROLLING(int, xk, 0, 1, _IWEI_WIDTH,
{
+#if GPU_ARCH == GPU_ARCH_MIDGARD
+ c[m0].v += a[xk + m0].v * b[xk].v;
+#else // GPU_ARCH == GPU_ARCH_MIDGARD
c[m0].v = fma(a[xk + m0].v, b[xk].v, c[m0].v);
+#endif // GPU_ARCH == GPU_ARCH_MIDGARD
})
})
}