generated/transpose_i4.c \
generated/transpose_i8.c \
generated/transpose_i16.c \
+generated/transpose_r10.c \
+generated/transpose_r16.c \
generated/transpose_c4.c \
generated/transpose_c8.c \
generated/transpose_c10.c \
generated/reshape_i4.c \
generated/reshape_i8.c \
generated/reshape_i16.c \
+generated/reshape_r10.c \
+generated/reshape_r16.c \
generated/reshape_c4.c \
generated/reshape_c8.c \
generated/reshape_c10.c \
intrinsics/dprod_r8.f90 \
intrinsics/f2c_specifics.F90
-# Turn on vectorization for matmul.
-$(patsubst %.c,%.lo,$(notdir $(i_matmul_c))): AM_CFLAGS += -ftree-vectorize
+# Turn on vectorization and loop unrolling for matmul.
+$(patsubst %.c,%.lo,$(notdir $(i_matmul_c))): AM_CFLAGS += -ftree-vectorize -funroll-loops
+# Logical matmul doesn't vectorize.
+$(patsubst %.c,%.lo,$(notdir $(i_matmull_c))): AM_CFLAGS += -funroll-loops
BUILT_SOURCES=$(gfor_built_src) $(gfor_built_specific_src) \
$(gfor_built_specific2_src)