COMPMID-661: softmax-fp32 optimisation (#14)
Change-Id: I2007af1ed9dcf68065cf412aa50f73a2025b31a6
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/94605
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
diff --git a/src/core/CL/cl_kernels/fixed_point.h b/src/core/CL/cl_kernels/fixed_point.h
index 5476a6e..b329118 100644
--- a/src/core/CL/cl_kernels/fixed_point.h
+++ b/src/core/CL/cl_kernels/fixed_point.h
@@ -359,7 +359,12 @@
return select((type)stype##_MAX, select(sum << dec_m, sum >> -dec_m, dec_m < (type)0), clz(sum) > dec_m); /* Saturate result if needed */ \
}
+EXPQ_IMPL(qs8, qs8x2, 2)
+EXPQ_IMPL(qs8, qs8x4, 4)
+EXPQ_IMPL(qs8, qs8x8, 8)
EXPQ_IMPL(qs8, qs8x16, 16)
+EXPQ_IMPL(qs16, qs16x2, 2)
+EXPQ_IMPL(qs16, qs16x4, 4)
EXPQ_IMPL(qs16, qs16x8, 8)
EXPQ_IMPL(qs16, qs16x16, 16)