Session 7 (finale)

Readme.md

@@ -14,4 +14,4 @@ Sessions :
 4. Unit testing. normalize example. Packaging & distributions concerns.
 5. Practice with paddedtensor example.
 6. Numerical problems (floats)
-7. *performance concerns*
+7. Performance 101, overview of hardware details

content.md

@@ -546,6 +546,7 @@ Optimizing this part of the code has the most effect.
 Use tools for the code language !
 Using a C profiler in python will tell you which interpreter internal function costs more, not functions from the python code.
 Compiled languages with C compatibility : perf + processing (flamegraphs, ...) is very effective.
+Python : cprofile.
 
 ## Optimize
 Many strategies are very dependent on the code.
@@ -568,9 +569,9 @@ Readability can be reduced in optimized code
 Hardware related : generally not possible in interpreted languages (lack of fine control).
 
 Computer has many sub-components that can affect performance :
+- cpu caches (data & instruction)
 - memory connection to cpu
 - memory internal structure
-- cpu caches (data & instruction)
 - TLB / MMU : virtual memory (mostly for security)
 - computing units number, organisation, timings : simple integer ALU (&,|,+...), complex integer (*,/,mod,...), floating point, vectorized
 - superscalar paradigm : "asynchronous" fetch/decode/execute