ComputeMinMax(): speed-up computation in the non-mask case for all non-complex types (but Byte and UInt16 that were already optimized)

timings wih a SSE2 optimized build:

- before:
testByte(): 0.306
testInt8(): 5.779
testUInt16(): 0.726
testInt16(): 2.519
testUInt32(): 6.933
testInt32(): 7.537
testUInt64(): 8.140
testInt64(): 7.807
testFloat16(): 12.953
testFloat32(): 6.105
testFloat64(): 6.525

- after
testByte(): 0.308
testInt8(): 0.440
testUInt16(): 0.710
testInt16(): 0.897
testUInt32(): 1.986
testInt32(): 1.826
testUInt64(): 6.004
testInt64(): 5.641
testFloat16(): 2.120
testFloat32(): 1.811
testFloat64(): 3.571

Author: rouault

gcore/gdalrasterband.cpp

@@ -8260,6 +8260,18 @@ CPLErr GDALRasterBand::ComputeRasterMinMax(int bApproxOK, double *adfMinMax)
         }
     }
 
+    if (!bApproxOK &&
+        (eDataType == GDT_Int8 || eDataType == GDT_Int16 ||
+         eDataType == GDT_UInt32 || eDataType == GDT_Int32 ||
+         eDataType == GDT_UInt64 || eDataType == GDT_Int64 ||
+         eDataType == GDT_Float16 || eDataType == GDT_Float32 ||
+         eDataType == GDT_Float64) &&
+        !poMaskBand)
+    {
+        return ComputeRasterMinMaxLocation(&adfMinMax[0], &adfMinMax[1],
+                                           nullptr, nullptr, nullptr, nullptr);
+    }
+
     bool bSignedByte = false;
     if (eDataType == GDT_Byte)
     {

perftests/computeminmax.py

@@ -8,8 +8,14 @@
 tab_ds = {}
 for dt in (
     gdal.GDT_Byte,
+    gdal.GDT_Int8,
     gdal.GDT_UInt16,
     gdal.GDT_Int16,
+    gdal.GDT_UInt32,
+    gdal.GDT_Int32,
+    gdal.GDT_UInt64,
+    gdal.GDT_Int64,
+    gdal.GDT_Float16,
     gdal.GDT_Float32,
     gdal.GDT_Float64,
 ):
@@ -27,6 +33,10 @@ def test(dt):
     "testByte(): %.3f"
     % timeit.timeit("test(gdal.GDT_Byte)", setup=setup, number=NITERS)
 )
+print(
+    "testInt8(): %.3f"
+    % timeit.timeit("test(gdal.GDT_Int8)", setup=setup, number=NITERS)
+)
 print(
     "testUInt16(): %.3f"
     % timeit.timeit("test(gdal.GDT_UInt16)", setup=setup, number=NITERS)
@@ -35,6 +45,26 @@ def test(dt):
     "testInt16(): %.3f"
     % timeit.timeit("test(gdal.GDT_Int16)", setup=setup, number=NITERS)
 )
+print(
+    "testUInt32(): %.3f"
+    % timeit.timeit("test(gdal.GDT_UInt32)", setup=setup, number=NITERS)
+)
+print(
+    "testInt32(): %.3f"
+    % timeit.timeit("test(gdal.GDT_Int32)", setup=setup, number=NITERS)
+)
+print(
+    "testUInt64(): %.3f"
+    % timeit.timeit("test(gdal.GDT_UInt64)", setup=setup, number=NITERS)
+)
+print(
+    "testInt64(): %.3f"
+    % timeit.timeit("test(gdal.GDT_Int64)", setup=setup, number=NITERS)
+)
+print(
+    "testFloat16(): %.3f"
+    % timeit.timeit("test(gdal.GDT_Float16)", setup=setup, number=NITERS)
+)
 print(
     "testFloat32(): %.3f"
     % timeit.timeit("test(gdal.GDT_Float32)", setup=setup, number=NITERS)