Hi,<br><br>We have a performance problem on SH4 with gcc4.x.<br><br><span style="font-family: courier new,monospace;">SciMark2 Numeric Benchmark, see <a href="http://math.nist.gov/scimark">http://math.nist.gov/scimark</a></span>
<br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">================================================================</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">
           GCC: 3.4.6   4.2.1   4.3.0 (20070907)</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">     Composite:  6.05    5.01    4.82</span><br style="font-family: courier new,monospace;">
<span style="font-family: courier new,monospace;">           FFT:  4.90    4.15    4.21</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">           SOR: 10.10    8.36
    7.64</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">    MonteCarlo:  3.68    3.06    3.04</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">
Sparse matmult:  5.45    4.45    4.03</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">            LU:  6.10    5.03    5.18</span><br style="font-family: courier new,monospace;">
<span style="font-family: courier new,monospace;">================================================================</span><br style="font-family: courier new,monospace;"><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">
BYTEmark* Native Mode Benchmark ver. 2 (10/95)</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">================================================================</span>
<br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">             GCC:      3.4.6      4.2.1  4.3.0 (20070907)</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">
    NUMERIC SORT:     35.459       32.2      29.327</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">     STRING SORT:     0.5943    0.57604      0.8603</span><br style="font-family: courier new,monospace;">
<span style="font-family: courier new,monospace;">        BITFIELD: 1.0585e+07  9.269e+06  9.4138e+06</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">    FP EMULATION:     
4.4944     4.6012       5.364</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">         FOURIER:     272.28     241.34      259.12</span><br style="font-family: courier new,monospace;">
<span style="font-family: courier new,monospace;">      ASSIGNMENT:    0.35997    0.38373     0.39683</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">            IDEA:     
124.11     95.057      100.07</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">         HUFFMAN:     45.593     52.083      56.391</span><br style="font-family: courier new,monospace;">
<span style="font-family: courier new,monospace;">      NEURAL NET:    0.36153    0.30922     0.31348</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">LU DECOMPOSITION:     
11.331     9.4938       8.255</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">================================================================</span><br><br>The "real world application" has 20%-200% performance regression with GCC 
4.x.<br><br>This effectively prevents us from moving to RTEMS 4.7 from 4.6.<br><br>I've reported this issue to gcc bugzilla: <a href="http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33431">http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33431
</a><br><br>But SH4 backend maintainer Kazumoto Kojima, was unable to reproduce it under linux-sh:<br>================================================================<br><span style="font-family: courier new,monospace;">                        
gcc-3.4.6    gcc-4.2.1    gcc-4.3.0(20070910)</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">Composite Score:            16.76        16.86        16.99</span><br style="font-family: courier new,monospace;">
<span style="font-family: courier new,monospace;">FFT              Mflops:    12.92        13.36        13.36</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">SOR              Mflops:    
27.88        26.76        28.01</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">MonteCarlo:      Mflops:     9.96         9.73         9.67</span><br style="font-family: courier new,monospace;">
<span style="font-family: courier new,monospace;">Sparse matmult   Mflops:    14.95        16.06        14.84</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">LU               Mflops:    
18.08        18.39        19.05</span><br>================================================================<br><br>Maybe, somebody is also using RTEMS on SH4 and can confirm my or Kojima results?<br><br>----<br>Nickolay