Add comparison results of DFT performance on macbook pro. (#668)

Author: shibatch

Jenkinsfile

@@ -151,19 +151,19 @@ pipeline {
             	     }
                 }
 
-                stage('aarch64 linux clang-18') {
+                stage('aarch64 linux clang-19') {
             	     agent { label 'aarch64 && ubuntu24 && apple' }
                      options { skipDefaultCheckout() }
             	     steps {
                          cleanWs()
                          checkout scm
 	    	     	 sh '''
-                	 echo "aarch64 clang-18 on" `hostname`
-			 export CC=clang-18
-			 export CXX=clang++-18
+                	 echo "aarch64 clang-19 on" `hostname`
+			 export CC=clang-19
+			 export CXX=clang++-19
  			 mkdir build
 			 cd build
-			 cmake .. -GNinja -DCMAKE_INSTALL_PREFIX=../../install -DSLEEF_SHOW_CONFIG=1 -DSLEEF_BUILD_DFT=TRUE -DSLEEF_ENFORCE_DFT=TRUE -DSLEEF_BUILD_QUAD=TRUE -DSLEEF_BUILD_INLINE_HEADERS=TRUE -DSLEEF_ENFORCE_SVE=TRUE -DEMULATOR=qemu-aarch64-static -DSLEEF_ENFORCE_TESTER4=True -DSLEEF_ENABLE_TESTER=False
+			 cmake .. -GNinja -DCMAKE_INSTALL_PREFIX=../../install -DSLEEF_SHOW_CONFIG=1 -DSLEEF_BUILD_DFT=TRUE -DSLEEF_ENFORCE_DFT=TRUE -DSLEEF_BUILD_QUAD=TRUE -DSLEEF_BUILD_INLINE_HEADERS=TRUE -DSLEEF_ENFORCE_SVE=TRUE -DEMULATOR=qemu-aarch64-static -DSLEEF_ENFORCE_TESTER4=True -DSLEEF_ENABLE_TESTER=False -DSLEEF_ENABLE_LTO=True -DCMAKE_EXE_LINKER_FLAGS="-fuse-ld=lld-19"
 			 cmake -E time oomstaller ninja -j `nproc`
 		         export CTEST_OUTPUT_ON_FAILURE=TRUE
 		         ctest -j `nproc`

docs/5-performance/README.md

@@ -96,16 +96,39 @@ FFTW_MEASURE mode, respectively.
 
 <p style="text-align:center; margin-bottom:2cm;">
   <a class="nothing" href="../img/dftzen4dp.png">
-    <img src="../img/dftzen4dp.png" alt="Performance graph for DP DFT"/>
+    <img src="../img/dftzen4dp.png" alt="Performance graph for DP DFT on Ryzen 9 7950X"/>
   </a>
   <br />
   Fig. 6.5: Performance of transform in double precision
 </p>
 
 <p style="text-align:center; margin-bottom:2cm;">
   <a class="nothing" href="../img/dftzen4sp.png">
-    <img src="../img/dftzen4sp.png" alt="Performance graph for SP DFT"/>
+    <img src="../img/dftzen4sp.png" alt="Performance graph for SP DFT on Ryzen 9 7950X"/>
   </a>
   <br />
   Fig. 6.6: Performance of transform in single precision
 </p>
+
+Below is the result of comparison on M1 MacBook Pro with the following
+settings.
+
+* OS : Ubuntu 24.04, Linux 6.8.0-1011-asahi-arm
+* Compiler : Ubuntu clang version 19.1.1 (1ubuntu1~24.04.2)
+* FFTW version 3.3.10-1ubuntu3
+
+<p style="text-align:center; margin-bottom:2cm;">
+  <a class="nothing" href="../img/dftm1dp.png">
+    <img src="../img/dftm1dp.png" alt="Performance graph for DP DFT on M1"/>
+  </a>
+  <br />
+  Fig. 6.7: Performance of transform in double precision
+</p>
+
+<p style="text-align:center; margin-bottom:2cm;">
+  <a class="nothing" href="../img/dftm1sp.png">
+    <img src="../img/dftm1sp.png" alt="Performance graph for SP DFTon M1"/>
+  </a>
+  <br />
+  Fig. 6.8: Performance of transform in single precision
+</p>

docs/README.md

@@ -77,9 +77,8 @@ and multiple cores can be utilized for efficient computation. It has an API
 similar to that of FFTW for easy migration. The subroutines can utilize long
 vectors up to 2048 bits. The helper files for abstracting SIMD intrinsics are
 shared with SLEEF libm, and thus it is easy to port the DFT subroutines to other
-architectures. [Preliminary results of
-benchmark](https://github.com/shibatch/sleef/wiki/DFT-Performance) are now
-available.
+architectures. [Benchmark results of the DFT subroutines](5-performance) are
+now available.
 
 <h2 id="environment">Supported environments</h2>