Compile files for AVX, AVX2 or SSE only when needed

Signed-off-by: Stefan Weil <sw@weilnetz.de>

src/api/Makefile.am

@@ -53,9 +53,6 @@ libtesseract_la_LIBADD = \
     ../dict/libtesseract_dict.la \
     ../arch/libtesseract_arch.la \
     ../arch/libtesseract_native.la \
-    ../arch/libtesseract_avx.la \
-    ../arch/libtesseract_avx2.la \
-    ../arch/libtesseract_sse.la \
     ../lstm/libtesseract_lstm.la \
     ../ccstruct/libtesseract_ccstruct.la \
     ../cutil/libtesseract_cutil.la \
@@ -63,6 +60,16 @@ libtesseract_la_LIBADD = \
     ../ccutil/libtesseract_ccutil.la \
     ../opencl/libtesseract_opencl.la
 
+if AVX_OPT
+libtesseract_la_LIBADD += ../arch/libtesseract_avx.la
+endif
+if AVX2_OPT
+libtesseract_la_LIBADD += ../arch/libtesseract_avx2.la
+endif
+if SSE41_OPT
+libtesseract_la_LIBADD += ../arch/libtesseract_sse.la
+endif
+
 libtesseract_la_LDFLAGS += -version-info $(GENERIC_LIBRARY_VERSION) $(NOUNDEFINED)
 
 bin_PROGRAMS = tesseract

src/arch/Makefile.am

@@ -15,8 +15,15 @@ noinst_HEADERS += intsimdmatrix.h
 noinst_HEADERS += simddetect.h
 
 noinst_LTLIBRARIES = libtesseract_native.la
-noinst_LTLIBRARIES += libtesseract_avx.la libtesseract_avx2.la
+if AVX_OPT
+noinst_LTLIBRARIES += libtesseract_avx.la
+endif
+if AVX2_OPT
+noinst_LTLIBRARIES += libtesseract_avx2.la
+endif
+if SSE41_OPT
 noinst_LTLIBRARIES += libtesseract_sse.la
+endif
 noinst_LTLIBRARIES += libtesseract_arch.la
 
 libtesseract_arch_la_CPPFLAGS = $(AM_CPPFLAGS)
@@ -41,8 +48,14 @@ libtesseract_native_la_SOURCES = dotproduct.cpp
 
 libtesseract_arch_la_SOURCES = intsimdmatrix.cpp simddetect.cpp
 
+if AVX_OPT
 libtesseract_avx_la_SOURCES = dotproductavx.cpp
+endif
 
+if AVX2_OPT
 libtesseract_avx2_la_SOURCES = intsimdmatrixavx2.cpp
+endif
 
+if SSE41_OPT
 libtesseract_sse_la_SOURCES = dotproductsse.cpp intsimdmatrixsse.cpp
+endif

src/arch/dotproductavx.cpp

@@ -16,8 +16,10 @@
 // limitations under the License.
 ///////////////////////////////////////////////////////////////////////
 
-#if defined(__AVX__)
+#if !defined(__AVX__)
-// Implementation for avx capable archs.
+#error Implementation only for AVX capable architectures
+#endif
+
 #include <immintrin.h>
 #include <cstdint>
 #include "dotproductavx.h"
@@ -96,5 +98,3 @@ double DotProductAVX(const double* u, const double* v, int n) {
 }
 
 }  // namespace tesseract.
-
-#endif  // __AVX__

src/arch/dotproductsse.cpp

@@ -16,7 +16,9 @@
 // limitations under the License.
 ///////////////////////////////////////////////////////////////////////
 
-#if defined(__SSE4_1__)
+#if !defined(__SSE4_1__)
+#error Implementation only for SSE 4.1 capable architectures
+#endif
 
 #include <emmintrin.h>
 #include <smmintrin.h>
@@ -117,5 +119,3 @@ int32_t IntDotProductSSE(const int8_t* u, const int8_t* v, int n) {
 }
 
 }  // namespace tesseract.
-
-#endif  // __SSE4_1__

src/arch/intsimdmatrix.cpp

@@ -25,12 +25,11 @@ namespace tesseract {
 
 const IntSimdMatrix* IntSimdMatrix::intSimdMatrix = nullptr;
 
-// Computes a reshaped copy of the weight matrix w. If there are no
+// Computes a reshaped copy of the weight matrix w.
-// partial_funcs_, it does nothing.
+void IntSimdMatrix::Init(const GENERIC_2D_ARRAY<int8_t>& w,
-void IntSimdMatrix::Init(const GENERIC_2D_ARRAY<int8_t>& w, std::vector<int8_t>& shaped_w) const {
+                         std::vector<int8_t>& shaped_w) const {
-  if (partial_funcs_.empty()) return;
+  const int num_out = w.dim1();
-  int num_out = w.dim1();
+  const int num_in = w.dim2() - 1;
-  int num_in = w.dim2() - 1;
   // The rounded-up sizes of the reshaped weight matrix, excluding biases.
   int rounded_num_in = Roundup(num_in, num_inputs_per_group_);
   int rounded_num_out = RoundOutputs(num_out);

src/arch/intsimdmatrixavx2.cpp

@@ -16,9 +16,12 @@
 // limitations under the License.
 ///////////////////////////////////////////////////////////////////////
 
+#if !defined(__AVX2__)
+#error Implementation only for AVX2 capable architectures
+#endif
+
 #include "intsimdmatrix.h"
 
-#ifdef __AVX2__
 #include <immintrin.h>
 #include <cstdint>
 #include <algorithm>
@@ -265,16 +268,9 @@ static void PartialMatrixDotVector8(const int8_t* wi, const double* scales,
   }
   ExtractResults(result0, shift_id, wi, scales, num_out, v);
 }
-#else
-namespace tesseract {
-#endif  // __AVX2__
 
-#ifdef __AVX2__
 const IntSimdMatrix IntSimdMatrix::IntSimdMatrixAVX2 =
   IntSimdMatrix(kNumOutputsPerRegister, kMaxOutputRegisters, kNumInputsPerRegister, kNumInputsPerGroup, kNumInputGroups, {PartialMatrixDotVector64, PartialMatrixDotVector32,
    PartialMatrixDotVector16, PartialMatrixDotVector8});
-#else
-const IntSimdMatrix IntSimdMatrix::IntSimdMatrixAVX2 = IntSimdMatrix(1, 1, 1, 1, 1, {});
-#endif  // __AVX2__
 
 }  // namespace tesseract.

src/arch/intsimdmatrixsse.cpp

@@ -15,6 +15,10 @@
 // limitations under the License.
 ///////////////////////////////////////////////////////////////////////
 
+#if !defined(__SSE4_1__)
+#error Implementation only for SSE 4.1 capable architectures
+#endif
+
 #include "intsimdmatrix.h"
 
 #include <cstdint>
@@ -22,7 +26,6 @@
 
 namespace tesseract {
 
-#ifdef __SSE4_1__
 // Computes part of matrix.vector v = Wu. Computes 1 result.
 static void PartialMatrixDotVector1(const int8_t* wi, const double* scales,
                                     const int8_t* u, int num_in, int num_out,
@@ -31,14 +34,8 @@ static void PartialMatrixDotVector1(const int8_t* wi, const double* scales,
   // Add in the bias and correct for integer values.
   *v = (total / INT8_MAX + wi[num_in]) * *scales;
 }
-#endif  // __SSE4_1__
 
-#ifdef __SSE4_1__
 const IntSimdMatrix IntSimdMatrix::IntSimdMatrixSSE =
   IntSimdMatrix(1, 1, 1, 1, 1, {PartialMatrixDotVector1});
-#else
-const IntSimdMatrix IntSimdMatrix::IntSimdMatrixSSE =
-  IntSimdMatrix(1, 1, 1, 1, 1, {});
-#endif  // __SSE4_1__
 
 }  // namespace tesseract.

unittest/Makefile.am

@@ -175,6 +175,13 @@ intfeaturemap_test_LDADD = $(GTEST_LIBS) $(TESS_LIBS)
 
 intsimdmatrix_test_SOURCES = intsimdmatrix_test.cc
 intsimdmatrix_test_LDADD = $(GTEST_LIBS) $(TESS_LIBS)
+intsimdmatrix_test_CPPFLAGS = $(AM_CPPFLAGS)
+if AVX2_OPT
+intsimdmatrix_test_CPPFLAGS += -DAVX2
+endif
+if SSE41_OPT
+intsimdmatrix_test_CPPFLAGS += -DSSE4_1
+endif
 
 lang_model_test_SOURCES = lang_model_test.cc
 lang_model_test_LDADD = $(ABSEIL_LIBS) $(GTEST_LIBS) $(TRAINING_LIBS) $(TESS_LIBS) $(ICU_I18N_LIBS) $(ICU_UC_LIBS)

unittest/intsimdmatrix_test.cc

@@ -92,24 +92,32 @@ TEST_F(IntSimdMatrixTest, C) {
 
 // Tests that the SSE implementation gets the same result as the vanilla.
 TEST_F(IntSimdMatrixTest, SSE) {
+#if defined(SSE4_1)
   if (SIMDDetect::IsSSEAvailable()) {
     tprintf("SSE found! Continuing...");
   } else {
-    tprintf("No SSE found! Not Tested!");
+    tprintf("No SSE found! Not tested!");
     return;
   }
   ExpectEqualResults(IntSimdMatrix::IntSimdMatrixSSE);
+#else
+  tprintf("SSE unsupported! Not tested!");
+#endif
 }
 
 // Tests that the AVX2 implementation gets the same result as the vanilla.
 TEST_F(IntSimdMatrixTest, AVX2) {
+#if defined(AVX2)
   if (SIMDDetect::IsAVX2Available()) {
     tprintf("AVX2 found! Continuing...");
   } else {
-    tprintf("No AVX2 found! Not Tested!");
+    tprintf("No AVX2 found! Not tested!");
     return;
   }
   ExpectEqualResults(IntSimdMatrix::IntSimdMatrixAVX2);
+#else
+  tprintf("AVX2 unsupported! Not tested!");
+#endif
 }
 
 }  // namespace