Claude/optimize packmol turbo wl4 eu

AUTHORS

@@ -9,3 +9,8 @@ dynamics and docking. Journal of Computational Chemistry, 24(7):819-825,
 2003. 
 
 Home-Page: http://m3g.iqm.unicamp.br/packmol
+
+Contributor:
+- Michele Bonus, Heinrich Heine University Düsseldorf
+  Email: michele.bonus@hhu.de
+  GitHub: https://github.com/MicheleBonus/

CHANGELOG.md

@@ -13,6 +13,8 @@ packmol changelog
 Version 21.2.2-DEV
 --------------
 - ![INFO][badge-info] Add changelog verification CI run.
+- ![INFO][badge-info] Update README build/profile instructions and document recent cell-list and pair-kernel performance work.
+- ![INFO][badge-info] Add contributor credentials for Michele Bonus (Heinrich Heine University Düsseldorf).
 
 Version 21.2.1
 --------------

CMakeLists.txt

@@ -1,15 +1,48 @@
 cmake_minimum_required(VERSION 3.10)
 project(packmol Fortran)
 
+set(PACKMOL_PROFILE "baseline" CACHE STRING "Build profile: baseline, perf-native, debug, sanitize")
+set_property(CACHE PACKMOL_PROFILE PROPERTY STRINGS baseline perf-native debug sanitize)
+option(PACKMOL_UNSAFE_MATH "Enable unsafe/fast math optimizations" OFF)
 
 if(NOT CMAKE_BUILD_TYPE)
     set(CMAKE_BUILD_TYPE Release)
-endif(NOT CMAKE_BUILD_TYPE)
+endif()
 
 if(CMAKE_Fortran_COMPILER_ID MATCHES GNU)
-    add_compile_options(
-        -Wall "$<$<CONFIG:Debug>:-Werror>"
-    )
+    add_compile_options(-Wall)
+endif()
+
+if(PACKMOL_PROFILE STREQUAL "baseline")
+    add_compile_options(-O2)
+elseif(PACKMOL_PROFILE STREQUAL "perf-native")
+    add_compile_options(-O3)
+    if(CMAKE_Fortran_COMPILER_ID MATCHES GNU)
+        add_compile_options(-march=native -mtune=native -funroll-loops)
+    elseif(CMAKE_Fortran_COMPILER_ID MATCHES IntelLLVM OR CMAKE_Fortran_COMPILER_ID MATCHES Intel)
+        add_compile_options(-xHost -ipo)
+    endif()
+elseif(PACKMOL_PROFILE STREQUAL "debug")
+    add_compile_options(-g)
+    if(CMAKE_Fortran_COMPILER_ID MATCHES GNU)
+        add_compile_options(-fcheck=bounds -fbacktrace -ffpe-trap=zero,overflow,underflow)
+    endif()
+elseif(PACKMOL_PROFILE STREQUAL "sanitize")
+    add_compile_options(-O1 -g)
+    if(CMAKE_Fortran_COMPILER_ID MATCHES GNU)
+        add_compile_options(-fno-omit-frame-pointer -fsanitize=address,undefined -fcheck=all)
+        add_link_options(-fsanitize=address,undefined)
+    endif()
+else()
+    message(FATAL_ERROR "Unknown PACKMOL_PROFILE='${PACKMOL_PROFILE}'. Use one of: baseline, perf-native, debug, sanitize")
+endif()
+
+if(PACKMOL_UNSAFE_MATH)
+    if(CMAKE_Fortran_COMPILER_ID MATCHES GNU)
+        add_compile_options(-ffast-math)
+    elseif(CMAKE_Fortran_COMPILER_ID MATCHES IntelLLVM OR CMAKE_Fortran_COMPILER_ID MATCHES Intel)
+        add_compile_options(-fp-model fast=2)
+    endif()
 endif()
 
 # Build the executable

Makefile

@@ -18,7 +18,26 @@ FORTRAN=/usr/bin/gfortran
 #
 # Change the flags of the compilation if you want:
 #
-FLAGS= -O3 -march=native -funroll-loops
+PROFILE ?= baseline
+UNSAFE_MATH ?= 0
+
+ifneq (,$(filter perf perf-native,$(MAKECMDGOALS)))
+PROFILE := perf-native
+endif
+ifneq (,$(filter devel,$(MAKECMDGOALS)))
+PROFILE := devel
+endif
+ifneq (,$(filter sanitize,$(MAKECMDGOALS)))
+PROFILE := sanitize
+endif
+ifneq (,$(filter static,$(MAKECMDGOALS)))
+PROFILE := static
+endif
+ifneq (,$(filter baseline,$(MAKECMDGOALS)))
+PROFILE := baseline
+endif
+
+FLAGS=
 SRCDIR= src
 MAINDIR= app 
 ###################################################################
@@ -30,20 +49,54 @@ MAINDIR= app
 # Flags for compiling development version
 #
 GENCANFLAGS := $(FLAGS)
-# Flags for the routines that signal with --fast-math
-IEEE_SIGNAL_FLAGS := $(FLAGS)
-ifeq ($(MAKECMDGOALS),devel)
+FC_VERSION := $(shell $(FORTRAN) --version 2>/dev/null | head -n 1)
+
+ifeq ($(PROFILE),baseline)
+FLAGS = -O2
+GENCANFLAGS = $(FLAGS)
+endif
+
+ifeq ($(PROFILE),perf-native)
+FLAGS = -O3
+GENCANFLAGS = $(FLAGS)
+ifneq (,$(findstring GNU Fortran,$(FC_VERSION)))
+FLAGS += -march=native -mtune=native -funroll-loops
+GENCANFLAGS += -march=native -mtune=native -funroll-loops
+else ifneq (,$(findstring ifx,$(FC_VERSION)))
+FLAGS += -xHost -ipo
+GENCANFLAGS += -xHost -ipo
+else ifneq (,$(findstring ifort,$(FC_VERSION)))
+FLAGS += -xHost -ipo
+GENCANFLAGS += -xHost -ipo
+endif
+endif
+
+ifeq ($(PROFILE),devel)
 FLAGS = -Wall -fcheck=bounds -g -fbacktrace -ffpe-trap=zero,overflow,underflow
-GENCANFLAGS = -fcheck=bounds -g -fbacktrace -ffpe-trap=zero,overflow,underflow 
+GENCANFLAGS = -fcheck=bounds -g -fbacktrace -ffpe-trap=zero,overflow,underflow
 endif
-ifeq ($(MAKECMDGOALS),perf)
-FLAGS = -g -pg
-GENCANFLAGS = -g -pg
+
+ifeq ($(PROFILE),sanitize)
+FLAGS = -O1 -g -fno-omit-frame-pointer
+GENCANFLAGS = $(FLAGS)
+ifneq (,$(findstring GNU Fortran,$(FC_VERSION)))
+FLAGS += -fsanitize=address,undefined -fcheck=all
+GENCANFLAGS += -fsanitize=address,undefined -fcheck=all
 endif
-ifeq ($(MAKECMDGOALS),static)
-FLAGS = -O3 --fast-math -static
-GENCANFLAGS = -O3 --fast-math -static
 endif
+
+ifeq ($(PROFILE),static)
+FLAGS = -O2 -static
+GENCANFLAGS = $(FLAGS)
+endif
+
+ifeq ($(UNSAFE_MATH),1)
+FLAGS += -ffast-math
+GENCANFLAGS += -ffast-math
+endif
+IEEE_SIGNAL_FLAGS = $(FLAGS)
+
+
 #
 # Files required
 #
@@ -103,17 +156,22 @@ all : $(oall)
 #
 # Compiling with flags for development
 #
-static : devel
-perf : devel
-devel : $(oall)
-	@echo " ------------------------------------------------------ " 
-	@echo " Compiling packmol with $(FORTRAN) " 
-	@echo " Flags: $(FLAGS)"
-	@echo " ------------------------------------------------------ "
-	@$(FORTRAN) -o packmol $(oall) $(FLAGS)
-	@echo " ------------------------------------------------------ " 
-	@echo " Packmol succesfully built. " 
-	@echo " ------------------------------------------------------ " 
+baseline: PROFILE=baseline
+baseline: all
+
+perf-native: PROFILE=perf-native
+perf-native: all
+
+devel: PROFILE=devel
+devel: all
+
+sanitize: PROFILE=sanitize
+sanitize: all
+
+static: PROFILE=static
+static: all
+
+perf : perf-native
 #
 # Modules
 #

README.md

@@ -83,6 +83,101 @@ uvx packmol < inp.pack
    `FPM_FC=compiler`, for example for `ifort`, use in bash, `export FPM_FC=ifort`.
 
 
+
+### Build profiles (Makefile and CMake)
+
+Packmol now supports explicit build profiles so you can choose between conservative optimized builds, aggressive benchmarking builds, and debugging/sanitized builds.
+
+#### Baseline release profile (default)
+
+This profile prioritizes stability and portability with conservative optimization.
+
+- **Makefile**: `make` (or `make baseline`)
+- **CMake**: `cmake -S . -B build -DPACKMOL_PROFILE=baseline`
+
+Then build with:
+
+```bash
+cmake --build build --config Release
+```
+
+#### Aggressive benchmark profile (`perf-native`)
+
+This profile is opt-in and applies compiler-specific native tuning (for example, `-march=native` for GNU or `-xHost` for Intel compilers). Use this for local benchmarking only.
+
+- **Makefile**: `make perf-native`
+- **CMake**: `cmake -S . -B build -DPACKMOL_PROFILE=perf-native`
+
+Then build with:
+
+```bash
+cmake --build build --config Release
+```
+
+#### Debug / sanitized profiles
+
+Use these profiles for troubleshooting and memory/UB diagnostics.
+
+- **Makefile debug checks**: `make devel`
+- **Makefile sanitizers (GNU)**: `make sanitize`
+- **CMake debug checks**: `cmake -S . -B build -DPACKMOL_PROFILE=debug`
+- **CMake sanitizers (GNU)**: `cmake -S . -B build -DPACKMOL_PROFILE=sanitize`
+
+Then build with:
+
+```bash
+cmake --build build --config Debug
+```
+
+### Performance-oriented kernel improvements
+
+Recent performance work in the pairwise cell-list kernels includes:
+
+- Prebuilt neighbor-cell offsets used in `computef` / `computeg` traversal.
+- Reduced branch pressure in pair kernels by separating hot paths.
+- Better data locality in hot loops through structure-of-arrays style access.
+- Conservative per-cell broad-phase reject checks before expensive pair evaluation.
+- Consistent `x*x` math micro-optimizations replacing hot-loop `**2` exponentiation.
+- Build-profile driven tuning with numerics validation support.
+
+These updates are designed to speed up large systems while preserving numerical behavior.
+
+## Contributors
+
+- Michele Bonus, Heinrich Heine University Düsseldorf  
+  Email: `michele.bonus@hhu.de`  
+  GitHub: https://github.com/MicheleBonus/
+
+#### Unsafe math is explicit opt-in
+
+Unsafe/fast-math is **not** enabled globally.
+
+- **Makefile**: add `UNSAFE_MATH=1`, e.g. `make perf-native UNSAFE_MATH=1`
+- **CMake**: set `-DPACKMOL_UNSAFE_MATH=ON`
+
+### Reproducibility and numerics check across profiles
+
+To compare objective and gradient behavior across profiles using existing inputs under `testing/`, run:
+
+```bash
+./testing/check_numerics_profiles.sh
+```
+
+What it does:
+
+1. Builds and runs `baseline` and `perf-native` binaries.
+2. Uses `testing/input_files/benzene2.inp` with `chkgrad` enabled (small system for fast finite-difference gradient checking).
+3. Compares values extracted from each run's `chkgrad.log`:
+   - Objective function value (`Function Value = ...`)
+   - Gradient check summary (`Maximum difference = ... Error= ...`)
+4. Fails if differences exceed tolerances.
+
+You can override tolerances with environment variables:
+
+```bash
+OBJ_TOL=1e-8 GRAD_TOL=1e-3 ./testing/check_numerics_profiles.sh
+```
+
 ## References
 
 Please always cite one of the following references in publications for which Packmol was useful:
@@ -93,4 +188,3 @@ JM Martinez, L Martinez, Packing optimization for the automated generation of co
 (https://doi.org/10.1002/jcc.10216)
 
 
-

app/packmol.f90

@@ -593,9 +593,14 @@ program packmol
                    keyword(iline,1).eq.'below') then
                   nratom(icart) = nratom(icart) + 1
                   iratcount = iratcount + 1
+                  iirest = -1
                   do irest = 1, nrest
                     if(irestline(irest).eq.iline) iirest = irest
                   end do
+                  if(iirest.eq.-1) then
+                    write(*,*) ' ERROR: Could not map atom-level restriction to definition at line ', iline
+                    stop exit_code_input_error
+                  end if
                   iratom(icart,iratcount) = iirest
                 end if
               end if
@@ -608,9 +613,14 @@ program packmol
                   keyword(iline,1).eq.'below') then
             nratom(icart) = nratom(icart) + 1    
             iratcount = iratcount + 1
+            iirest = -1
             do irest = 1, nrest
               if(irestline(irest).eq.iline) iirest = irest
             end do
+            if(iirest.eq.-1) then
+              write(*,*) ' ERROR: Could not map restriction to definition at line ', iline
+              stop exit_code_input_error
+            end if
             iratom(icart,iratcount) = iirest
           end if
         end do
@@ -710,6 +720,9 @@ program packmol
 
   call initial(n,x)
 
+  ! Pre-compute fixed_short_marker once (fixedatom and use_short_radius are static)
+  call init_fixed_short_marker()
+
   ! Computing the energy at the initial point
 
   radscale = 1.d0
@@ -978,4 +991,3 @@ program packmol
   end if
 
 end program packmol
-

src/cell_indexing.f90

@@ -1,7 +1,30 @@
 module cell_indexing
 
-implicit none
-public :: setcell, icell_to_cell, index_cell
+implicit none
+public :: setcell, icell_to_cell, index_cell, n_forward_offsets, forward_offsets
+
+! Ordered list of relative cell offsets used for forward-only traversal.
+!  1: (0, 0, 0) -> self cell (handled with latomnext to avoid double-counting)
+!  2-4: forward face neighbors
+!  5-10: forward edge neighbors
+! 11-14: forward vertex neighbors
+integer, parameter :: n_forward_offsets = 14
+integer, parameter :: forward_offsets(3,n_forward_offsets) = reshape([ &
+   0, 0, 0, &
+   1, 0, 0, &
+   0, 1, 0, &
+   0, 0, 1, &
+   1,-1, 0, &
+   1, 0,-1, &
+   0, 1,-1, &
+   0, 1, 1, &
+   1, 1, 0, &
+   1, 0, 1, &
+   1,-1,-1, &
+   1,-1, 1, &
+   1, 1,-1, &
+   1, 1, 1  &
+], [3,n_forward_offsets])
 
 contains 
 !

src/checkpoint.f90

@@ -12,7 +12,7 @@
 subroutine checkpoint(n,x)
 
    use sizes
-   use compute_data
+   use compute_data, only : comptype, init1
    use input
    use usegencan
    use ahestetic