Issue AcademySoftwareFoundation#2116 : Fixes Metal backend's generate…

…d shaders with float/int constant Array Performance (AcademySoftwareFoundation#2117) * Issue AcademySoftwareFoundation#2116 : Improves Metal Backend Perf. moves the constant float/int declaration to constant space so it doesnt get initialized per thread. This improved color correction performance on M4 Max 3-4 times better. Signed-off-by: Morteza <smostajabodaveh@apple.com> * Tiny refactoring to improve code maintainability Signed-off-by: Morteza <smostajabodaveh@apple.com> --------- Signed-off-by: Morteza <smostajabodaveh@apple.com>
wetadigital · Feb 10, 2025 · d807b38 · d807b38
1 parent c5c85b0
commit d807b38
Show file tree

Hide file tree

Showing 2 changed files with 53 additions and 46 deletions.
diff --git a/src/OpenColorIO/GpuShaderUtils.cpp b/src/OpenColorIO/GpuShaderUtils.cpp
@@ -513,6 +513,18 @@ void GpuShaderText::declareFloatArrayConst(const std::string & name, int size, c
     }
 
     auto nl = newLine();
+
+    auto emitArrayValues = [&]()
+    {
+        for (int i = 0; i < size; ++i)
+        {
+            nl << getFloatString(v[i], m_lang);
+            if (i + 1 != size)
+            {
+                nl << ", ";
+            }
+        }
+    };
 
     switch (m_lang)
     {
@@ -524,34 +536,30 @@ void GpuShaderText::declareFloatArrayConst(const std::string & name, int size, c
         {
             nl << floatKeywordConst() << " " << name << "[" << size << "] = ";
             nl << floatKeyword() << "[" << size << "](";
-            for (int i = 0; i < size; ++i)
-            {
-                nl << getFloatString(v[i], m_lang);
-                if (i + 1 != size)
-                {
-                    nl << ", ";
-                }
-            }
+            emitArrayValues();
             nl << ");";
             break;
         }
         case LANGUAGE_OSL_1:
         case GPU_LANGUAGE_CG:
         case GPU_LANGUAGE_HLSL_SM_5_0:
+        {
+            nl << floatKeywordConst();
+            nl << " " << name << "[" << size << "] = {";
+            emitArrayValues();
+            nl << "};";
+            break;
+        }
+
         case GPU_LANGUAGE_MSL_2_0:
         {
-            nl << floatKeywordConst() << " " << name << "[" << size << "] = {";
-            for (int i = 0; i < size; ++i)
-            {
-                nl << getFloatString(v[i], m_lang);
-                if (i + 1 != size)
-                {
-                    nl << ", ";
-                }
-            }
+            nl << "constant constexpr static float";
+            nl << " " << name << "[" << size << "] = {";
+            emitArrayValues();
             nl << "};";
             break;
         }
+
     }
 }
 
@@ -567,6 +575,18 @@ void GpuShaderText::declareIntArrayConst(const std::string & name, int size, con
     }
 
     auto nl = newLine();
+
+    auto emitArrayValues = [&]()
+    {
+        for (int i = 0; i < size; ++i)
+        {
+            nl << v[i];
+            if (i + 1 != size)
+            {
+                nl << ", ";
+            }
+        }
+    };
 
     switch (m_lang)
     {
@@ -578,44 +598,31 @@ void GpuShaderText::declareIntArrayConst(const std::string & name, int size, con
         {
             nl << intKeywordConst() << " " << name << "[" << size << "] = "
                << intKeyword() << "[" << size << "](";
-            for (int i = 0; i < size; ++i)
-            {
-                nl << v[i];
-                if (i + 1 != size)
-                {
-                    nl << ", ";
-                }
-            }
+            emitArrayValues();
             nl << ");";
             break;
         }
         case GPU_LANGUAGE_HLSL_SM_5_0:
+        {
+            nl << intKeywordConst();
+            nl << " " << name << "[" << size << "] = {";
+            emitArrayValues();
+            nl << "};";
+            break;
+        }
         case GPU_LANGUAGE_MSL_2_0:
         {
-            nl << intKeywordConst() << " " << name << "[" << size << "] = {";
-            for (int i = 0; i < size; ++i)
-            {
-                nl << v[i];
-                if (i + 1 != size)
-                {
-                    nl << ", ";
-                }
-            }
+            nl << "constant constexpr static int";
+            nl << " " << name << "[" << size << "] = {";
+            emitArrayValues();
             nl << "};";
             break;
         }
         case LANGUAGE_OSL_1:
         case GPU_LANGUAGE_CG:
         {
             nl << intKeyword() << " " << name << "[" << size << "] = {";
-            for (int i = 0; i < size; ++i)
-            {
-                nl << v[i];
-                if (i + 1 != size)
-                {
-                    nl << ", ";
-                }
-            }
+            emitArrayValues();
             nl << "};";
             break;
         }

diff --git a/tests/cpu/GpuShader_tests.cpp b/tests/cpu/GpuShader_tests.cpp
@@ -1032,10 +1032,10 @@ ocioOCIOMain(
 // Declaration of all helper methods
 
 
-const int ocio_grading_rgbcurve_knotsOffsets_0[8] = {0, 5, -1, 0, -1, 0, -1, 0};
-const float ocio_grading_rgbcurve_knots_0[5] = {0., 0.333333343, 0.5, 0.666666508, 1.};
-const int ocio_grading_rgbcurve_coefsOffsets_0[8] = {0, 12, -1, 0, -1, 0, -1, 0};
-const float ocio_grading_rgbcurve_coefs_0[12] = {0.0982520878, 0.393008381, 0.347727984, 0.08693178, 0.934498608, 1., 1.13100278, 1.246912, 0., 0.322416425, 0.5, 0.698159397};
+constant constexpr static int ocio_grading_rgbcurve_knotsOffsets_0[8] = {0, 5, -1, 0, -1, 0, -1, 0};
+constant constexpr static float ocio_grading_rgbcurve_knots_0[5] = {0., 0.333333343, 0.5, 0.666666508, 1.};
+constant constexpr static int ocio_grading_rgbcurve_coefsOffsets_0[8] = {0, 12, -1, 0, -1, 0, -1, 0};
+constant constexpr static float ocio_grading_rgbcurve_coefs_0[12] = {0.0982520878, 0.393008381, 0.347727984, 0.08693178, 0.934498608, 1., 1.13100278, 1.246912, 0., 0.322416425, 0.5, 0.698159397};
 
 float ocio_grading_rgbcurve_evalBSplineCurve_0(int curveIdx, float x)
 {