Testing the OpenAD installation

Testing the Installation

There is a small regression test suite with a setup script contained in
<path_to_OpenAD>/xaifBooster/testRoundTrip/
We compare the numerical output against a reference file. The recent test results are found in the table at the bottom. Depending on the machine the tests run on, the compiler you use and the chosen transformation (tangent-linear, joint or split adjoint) the numbers may vary slightly. To execute all tests first
cd <path_to_OpenAD>
and set up the OpenAD environment for

shell/ksh/bash etc users with:
source ./setenv.sh
csh/tcsh etc. users with:
source ./setenv.csh

modify the Makefile to use the Fortran90 compiler of your choice (default is ifort) and the invoke the script
./testAll.bash
For the first example it asks a few questions determining the transformation mode as in the following example. For regression testing you may choose

for tangent-linear code: BasicBlockPreaccumulation or MemOpsTradeoffPreaccumulation
for adjoint code: BasicBlockPreaccumulationReverse

We marked the user inputs, the transformation and compile/link steps. The environment setup defines the absolute paths; your output may look slightly different.

xaifBooster/testRoundTrip> ./testAll.bash
use reverse mode y/[n]n
using plain drivers
rm -f head.B head.prh* driver driver.o 
rm -f -rf tmpOutput 
rm -f head.f driver.f params.conf
rm -f -rf  tmpOutput verifiedOutput 
** running mini1 *************************************************
ln -sf <path_to_OpenaAD>/Open64/osprey1.0/targ_ia32_ia64_linux/crayf90/sgi/mfef90 .
./mfef90 -F -N132 head.f
ln -sf <path_to_OpenAD>/OpenADFortTk/obj/i686-Linux/obj/whirl2xaif .
./whirl2xaif -o head.xaif head.B
ln -sf <path_to_OpenAD>/xaif/schema/xaif.xsd .
ln -sf <path_to_OpenAD>/xaif/schema/xaif_base.xsd .
ln -sf <path_to_OpenAD>/xaif/schema/xaif_inlinable_intrinsics.xsd .
ln -sf <path_to_OpenAD>/xaif/schema/xaif_derivative_propagator.xsd .
ln -sf <path_to_OpenAD>/xaif/schema/xaif_output.xsd .
./pickAlgorithm.bash
pick one of the following algorithms
1 : BasicBlockPreaccumulation
2 : BasicBlockPreaccumulationReverse
3 : BasicBlockPreaccumulationTape
4 : BasicBlockPreaccumulationTapeAdjoint
5 : ControlFlowReversal
6 : Linearization
7 : MemOpsTradeoffPreaccumulation
pick a number: 1
./xaifBooster -i head.xaif -c <path_to_OpenAD>/xaif-1.0/schema/examples/inlinable_intrinsics.xaif -o head.xb.xaif
ln -sf <path_to_OpenAD>/OpenADFortTk/obj/i686-Linux/obj/xaif2whirl .
./xaif2whirl --structured head.B head.xb.xaif
ln -sf <path_to_OpenAD>/Open64/osprey1.0/targ_ia32_ia64_linux/whirl2f/whirl2f .
ln -sf <path_to_OpenAD>/Open64/osprey1.0/targ_ia32_ia64_linux/whirl2f/whirl2f_be .
./whirl2f -openad head.xb.x2w.B
./whirl2f translates head.xb.x2w.B into head.xb.x2w.w2f.f, based on source head.xb.x2w.f
ln -sf <path_to_OpenAD>/OpenADFortTk/postprocess/pp.pl .
perl pp.pl -f head.xb.x2w.w2f.f
f95 -c w2f__types.f
Obsolescent: w2f__types.f, line 1: Fixed source form
f95 -free -c active_module.f
f95 -free -c driver.f
f95 -c head.xb.x2w.w2f.pp.f
Obsolescent: head.xb.x2w.w2f.pp.f, line 6: Fixed source form
f95 w2f__types.o active_module.o driver.o head.xb.x2w.w2f.pp.o -o driver
ln -s simple.params.conf params.conf
mkdir -p tmpOutput
./driver
no diffs
xaifBooster/testRoundTrip

test results:

name	issues (small numerical differences are always expected)
	tangent-linear (using BasicBlockPreaccumulation)	adjoint (reverse, using BasicBlockPreaccumulationReverse)
	tangent-linear (using BasicBlockPreaccumulation)	joint	split
argumentArrayIndex
argumentArrayIndex2
argumentArrayIndex3
argumentArrayIndex4
argumentConversion_1
argumentConversion_2
argumentConversion_3
argumentConversion_4
argumentConversion_4a	This is an example for f77 style pseudo pointer arithmetic where we interpret ax(1,1) and apx(1,1) as the beginning of a vector of length 2 rather than a scalar. This doesn't work with the f90 style conversion routines.	This is an example for f77 style pseudo pointer arithmetic where we interpret ax(1,1) and apx(1,1) as the beginning of a vector of length 2 rather than a scalar. This doesn't work with the f90 style conversion routines.	This is an example for f77 style pseudo pointer arithmetic where we interpret ax(1,1) and apx(1,1) as the beginning of a vector of length 2 rather than a scalar. This doesn't work with the f90 style conversion routines.
argumentConversion_4b	The interface for foo in the definition of head does not pick up on the activity of the formal parameters bug report	The interface for foo in the definition of head does not pick up on the activity of the formal parameters bug report	The interface for foo in the definition of head does not pick up on the activity of the formal parameters bug report
argumentConversion_4c
argumentConversion_5
bat
boxmodel
box_timestep
bratu
cast1
ConsFold
ConsFoldSimp0
constant0
constant1
constPrec
daerfj
dfdcfj
dfdcfj_MemOps
directDependency1
directDependency2
directDependency3
duud1
duud2
duudChains1
duudChains2
externalCalls1
externalCalls2
externalCalls3
float_intrinsic
gcmBit1
globalActive_1
globalActive_2
globalActive_3
globalActive_4
globalActive_5
globalActive_6	R1: the analysis doesn't recognize the module variables as active, bug report	R1: the analysis doesn't recognize the module variables as active, bug report	R1: the analysis doesn't recognize the module variables as active, bug report
globalActive_7
globalActive_8
If_Loop_1
If_Loop_2
If_Loop_3
If_Loop_4
If_Loop_5
IfStatement_1
IfStatement_2
lion
Loop_1
Loop_2
Loop_3
Loop_simple_1
Loop_simple_10
Loop_simple_10a
Loop_simple_11
Loop_simple_12		the simple loop assumptions are violated, a more generic solution involves the recomputation by inversion etc.; bug report	the simple loop assumptions are violated, a more generic solution involves the recomputation by inversion etc.; bug report
Loop_simple_13
Loop_simple_14
Loop_simple_15
Loop_simple_1a
Loop_simple_1b
Loop_simple_1c
Loop_simple_1d
Loop_simple_1e
Loop_simple_1f
Loop_simple_2
Loop_simple_2a
Loop_simple_3
Loop_simple_4
Loop_simple_4a		We try to restore j but cannot see that head as top level routine is called with constant for the actual parameter. This is expected behavior.	We try to restore j but cannot see that head as top level routine is called with constant for the actual parameter. This is expected behavior.
Loop_simple_4b		the simple loop assumptions are violated, a more generic solution involves the recomputation by inversion etc.; bug report	the simple loop assumptions are violated, a more generic solution involves the recomputation by inversion etc.; bug report
Loop_simple_4c
Loop_simple_5
Loop_simple_6
Loop_simple_7
Loop_simple_7a
Loop_simple_8
Loop_simple_9
Loop_simple_9a		the simple loop assumptions are violated, a more generic solution involves the recomputation by inversion etc.; bug report	the simple loop assumptions are violated, a more generic solution involves the recomputation by inversion etc.; bug report
Loop_simple_9b		the simple loop assumptions are violated, a more generic solution involves the recomputation by inversion etc.; bug report	the simple loop assumptions are violated, a more generic solution involves the recomputation by inversion etc.; bug report
mini1
nested_calls_1
nested_calls_2
nested_calls_3
nested_calls_4
nested_dfdcfj
nestedFunc0		postprocessor fails on contained subroutines bug report	postprocessor fails on contained subroutines bug report
nestedFunc1		postprocessor fails on contained subroutines bug report	postprocessor fails on contained subroutines bug report
NestedIf
NestedIfLoop
NestedLoopIf
NestedLoops
passiveStatement
passiveSubroutine
readWrite1	format failure for read of active variable, bug report	format failure for read of active variable, bug report	format failure for read of active variable, bug report
readWrite2	format failure for read of active variable, bug report	format failure for read of active variable, bug report	format failure for read of active variable, bug report
RoehFlux
RoehFlux_MemOps
Select_1
Select_2		control flow unparsing problem; bug report	control flow unparsing problem; bug report
Select_3
SeqOfIfs
SeqOfLoops
sideEffect_1
sideEffect_2
sideEffect_3
sideEffect_4
signIntrinsic
simple0
simple1
simple2
simple3
speelpenning
tanh_f_arg
todd
todd1
u_def_struct_1	fix up structure access bug report	fix up structure access bug report	fix up structure access bug report
u_def_struct_2	fix up structure access bug report	fix up structure access bug report	fix up structure access bug report