[Toon-members] TooN/test normalize_test.cc

[Edward Rosten]

CVSROOT:        /cvsroot/toon
Module name:    TooN
Changes by:     Edward Rosten <edrosten>        09/03/25 20:18:17

Added files:
        test           : normalize_test.cc 

Log message:
        Added test function to test normalization in place (norm_v_1), 
normalization
        as assignment fomr a unit vector computed out of place (norm_v_2) and 
assignment
        from a unit vector computed out of place on a slice (norm_v_3). 
Compilation options are:
        -O3  -ffast-math -mfpmath=sse -msse
        to make the asm code more readable. The results are:
        
            _Z8norm_v_1v:                      _Z8norm_v_2v:                    
  _Z8norm_v_3v:
        .LFB1511:                              .LFB1512:                        
  .LFB1513:
            movl    v, %eax                     movl    v, %eax                 
    movl    v, %eax
            pushl   %ebp                        pushl   %ebp                    
    pushl   %ebp
        .LCFI5:                                .LCFI7:                          
  .LCFI3:
            movl    %esp, %ebp                  movl    %esp, %ebp              
    movl    %esp, %ebp
        .LCFI6:                                .LCFI8:                          
  .LCFI4:
            movsd   (%eax), %xmm3               movsd   (%eax), %xmm2           
    movsd   (%eax), %xmm3
            movsd   8(%eax), %xmm4              movsd   8(%eax), %xmm3          
    movsd   8(%eax), %xmm4
            movsd   16(%eax), %xmm5             movsd   16(%eax), %xmm4         
    movsd   16(%eax), %xmm5
            movapd  %xmm3, %xmm1                movapd  %xmm2, %xmm1            
    movsd   24(%eax), %xmm6
            movapd  %xmm4, %xmm2                movapd  %xmm3, %xmm5            
    movapd  %xmm3, %xmm1
            mulsd   %xmm3, %xmm1                mulsd   %xmm2, %xmm1            
    movapd  %xmm4, %xmm2
            mulsd   %xmm4, %xmm2                mulsd   %xmm3, %xmm5            
    mulsd   %xmm3, %xmm1
            movapd  %xmm5, %xmm0                movapd  %xmm4, %xmm0            
    mulsd   %xmm4, %xmm2
            mulsd   %xmm5, %xmm0                mulsd   %xmm4, %xmm0            
    movapd  %xmm5, %xmm0
            addsd   %xmm2, %xmm0                addsd   %xmm5, %xmm0            
    mulsd   %xmm5, %xmm0
            movsd   24(%eax), %xmm2             movsd   24(%eax), %xmm5         
    addsd   %xmm2, %xmm0
            addsd   %xmm0, %xmm1                addsd   %xmm0, %xmm1            
    addsd   %xmm0, %xmm1
            movapd  %xmm2, %xmm0                movapd  %xmm5, %xmm0            
    movapd  %xmm6, %xmm0
            mulsd   %xmm2, %xmm0                mulsd   %xmm5, %xmm0            
    mulsd   %xmm6, %xmm0
            addsd   %xmm1, %xmm0                addsd   %xmm1, %xmm0            
    addsd   %xmm1, %xmm0
            movsd   .LC0, %xmm1                 movsd   .LC0, %xmm1             
    movsd   .LC0, %xmm1
            sqrtsd  %xmm0, %xmm0                sqrtsd  %xmm0, %xmm0            
    sqrtsd  %xmm0, %xmm0
            divsd   %xmm0, %xmm1                divsd   %xmm0, %xmm1            
    divsd   %xmm0, %xmm1
            mulsd   %xmm1, %xmm3                mulsd   %xmm1, %xmm2            
    mulsd   %xmm1, %xmm3
            mulsd   %xmm1, %xmm4                mulsd   %xmm1, %xmm3            
    mulsd   %xmm1, %xmm4
            movsd   %xmm3, (%eax)               mulsd   %xmm1, %xmm4            
    movsd   %xmm3, (%eax)
            movsd   %xmm4, 8(%eax)              movsd   %xmm2, (%eax)           
    movsd   %xmm4, 8(%eax)
            mulsd   %xmm1, %xmm5                movsd   %xmm3, 8(%eax)          
    mulsd   %xmm1, %xmm5
            mulsd   %xmm1, %xmm2                movsd   %xmm4, 16(%eax)         
    mulsd   %xmm1, %xmm6
            movsd   %xmm5, 16(%eax)             mulsd   %xmm5, %xmm1            
    movsd   %xmm5, 16(%eax)
            movsd   %xmm2, 24(%eax)             movsd   %xmm1, 24(%eax)         
    movsd   %xmm6, 24(%eax)
            popl    %ebp                        popl    %ebp                    
    popl    %ebp
            ret                                 ret                             
    ret
        
        Note that the three functions are identical up to a minor reording in 
norm_v_3.
        There is therefore no need to have a normalize in place function, which 
is
        useful since working on slices is inconvenient since they are non-const
        temporaries.

CVSWeb URLs:
http://cvs.savannah.gnu.org/viewcvs/TooN/test/normalize_test.cc?cvsroot=toon&rev=1.1

Patches:
Index: normalize_test.cc
===================================================================
RCS file: normalize_test.cc
diff -N normalize_test.cc
--- /dev/null   1 Jan 1970 00:00:00 -0000
+++ normalize_test.cc   25 Mar 2009 20:18:16 -0000      1.1
@@ -0,0 +1,36 @@
+#include <TooN/TooN.h>
+#include <cmath>
+
+using namespace TooN;
+using namespace std;
+
+extern Vector<4>& v;
+
+inline void norm_in_place(Vector<4>& v)
+{
+       v/=sqrt(v*v);   
+}
+
+
+inline Vector<4> unit(const Vector<4>& v)
+{
+       return v/sqrt(v*v);     
+}
+
+
+void norm_v_1()
+{
+       norm_in_place(v);
+}
+
+void norm_v_2()
+{
+       v = unit(v);
+}
+
+
+void norm_v_3()
+{
+       v.slice<0,4>() = unit(v.slice<0,4>());
+}
+