Fix TFloat builds for Apple M1

Signed-off-by: Stefan Weil <[email protected]>
GerHobbelt · Jul 13, 2021 · a09531a · a09531a
1 parent 01ae69e
commit a09531a
Show file tree

Hide file tree

Showing 5 changed files with 43 additions and 7 deletions.
diff --git a/Makefile.am b/Makefile.am
@@ -146,8 +146,8 @@ noinst_LTLIBRARIES += libtesseract_native.la
 libtesseract_native_la_CXXFLAGS = -O3 -ffast-math
 if MARCH_NATIVE_OPT
 libtesseract_native_la_CXXFLAGS += -march=native -mtune=native
-libtesseract_native_la_CXXFLAGS += -I$(top_srcdir)/src/ccutil
 endif
+libtesseract_native_la_CXXFLAGS += -I$(top_srcdir)/src/ccutil
 libtesseract_native_la_SOURCES = src/arch/dotproduct.cpp
 
 if HAVE_AVX

diff --git a/src/arch/dotproduct.h b/src/arch/dotproduct.h
@@ -37,6 +37,7 @@ TFloat DotProductFMA(const TFloat *u, const TFloat *v, int n);
 // Uses Intel SSE intrinsics to access the SIMD instruction set.
 TFloat DotProductSSE(const TFloat *u, const TFloat *v, int n);
 
+TFloat DotProductAccelerate(const TFloat *u, const TFloat *v, int n);
 } // namespace tesseract.
 
 #endif // TESSERACT_ARCH_DOTPRODUCT_H_
diff --git a/src/arch/intsimdmatrixneon.cpp b/src/arch/intsimdmatrixneon.cpp
@@ -19,6 +19,7 @@
 #if defined(__ARM_NEON)
 
 #  include "intsimdmatrix.h"
+#  include "tfloat.h"
 
 #  include <algorithm>
 #  include <cstdint>
@@ -27,6 +28,12 @@
 
 namespace tesseract {
 
+#if defined(FAST_FLOAT)
+
+const IntSimdMatrix *IntSimdMatrix::intSimdMatrixNEON = nullptr;
+
+#else
+
 // Number of outputs held in each register. (Actually, we use a
 // pair of 4x32 registers, so 8 x 32 bit ints).
 constexpr int kNumOutputsPerRegister = 8;
@@ -186,7 +193,7 @@ static void matrixDotVector(int dim1, int dim2, const int8_t *wi, const double *
                             num_out & (kNumOutputsPerRegister - 1));
 }
 
-static const IntSimdMatrix intSimdMatrix = {
+static const IntSimdMatrix simdMatrix = {
     // Function.
     matrixDotVector,
     // Number of 32 bit outputs held in each register.
@@ -199,7 +206,9 @@ static const IntSimdMatrix intSimdMatrix = {
     kNumInputsPerGroup
 };
 
-const IntSimdMatrix *IntSimdMatrix::intSimdMatrixNEON = &intSimdMatrix;
+const IntSimdMatrix *IntSimdMatrix::intSimdMatrixNEON = &simdMatrix;
+
+#endif // FAST_FLOAT
 
 } // namespace tesseract.
 

diff --git a/src/arch/simddetect.cpp b/src/arch/simddetect.cpp
@@ -101,10 +101,14 @@ bool SIMDDetect::sse_available_;
 #endif
 
 #if defined(HAVE_FRAMEWORK_ACCELERATE)
-static double DotProductAccelerate(const double* u, const double* v, int n) {
-  double total = 0.0;
+TFloat DotProductAccelerate(const TFloat* u, const TFloat* v, int n) {
+  TFloat total = 0;
   const int stride = 1;
+#if defined(FAST_FLOAT)
+  vDSP_dotpr(u, stride, v, stride, &total, n);
+#else
   vDSP_dotprD(u, stride, v, stride, &total, n);
+#endif
   return total;
 }
 #endif
@@ -138,7 +142,7 @@ SIMDDetect::SIMDDetect() {
   SetDotProduct(DotProductGeneric);
   const char* dotproduct_env = getenv("DOTPRODUCT");
   if (dotproduct_env != nullptr) {
-    dotproduct = env;
+    dotproduct = dotproduct_env;
     Update();
     if (strcmp(dotproduct_env, "native") == 0) {
       SetDotProduct(DotProductNative);

diff --git a/unittest/dotproduct_test.cc b/unittest/dotproduct_test.cc
@@ -30,7 +30,7 @@ class DotProductTest : public ::testing::Test {
     std::locale::global(std::locale(""));
   }
   void RunTest(TFloat (*f)(const TFloat *u, const TFloat *v, int n));
-  static const size_t multiplications = 500000000;
+  static const size_t multiplications = 5000000000U;
   static const size_t n = 40;
   //static const size_t n = 1000000;
   TFloat u[n];
@@ -118,4 +118,26 @@ TEST_F(DotProductTest, FMA) {
 #endif
 }
 
+#if defined(HAVE_FRAMEWORK_ACCELERATE)
+TEST_F(DotProductTest, Accelerate) {
+  RunTest(DotProductAccelerate);
+}
+#endif
+
+#if 0
+// Tests that the NEON implementation gets the same result as the vanilla.
+TEST_F(DotProductTest, NEON) {
+#if defined(HAVE_NEON)
+  if (!SIMDDetect::IsNEONAvailable()) {
+    GTEST_LOG_(INFO) << "No NEON found! Not tested!";
+    GTEST_SKIP();
+  }
+  RunTest(DotProductNEON);
+#else
+  GTEST_LOG_(INFO) << "NEON unsupported! Not tested!";
+  GTEST_SKIP();
+#endif
+}
+#endif
+
 } // namespace tesseract