KOKKOS: close feature gaps, lift instance caps, and fix host/device coherence faults

cmake/common/process/sparta_build_options.cmake

@@ -199,6 +199,15 @@ endif()
     set(SPARTA_DEFAULT_CXX_COMPILE_FLAGS -DSPARTA_KOKKOS_EXACT
                                          ${SPARTA_DEFAULT_CXX_COMPILE_FLAGS})
   endif()
+  # SPARTA_KOKKOS_FIXED_LISTS restores the fixed-size KKCopy arrays for the
+  # per-type tally compute lists instead of runtime-sized device buffers.  The
+  # buffers lift the instance caps; the arrays keep every compute inside the
+  # kernel functor.  Which performs better is hardware-dependent -- functor
+  # size trades against occupancy and data locality -- so both are buildable.
+  if(SPARTA_KOKKOS_FIXED_LISTS)
+    set(SPARTA_DEFAULT_CXX_COMPILE_FLAGS -DSPARTA_KOKKOS_FIXED_LISTS
+                                         ${SPARTA_DEFAULT_CXX_COMPILE_FLAGS})
+  endif()
   # PKG_KOKKOS depends on BUILD_KOKKOS
   set(BUILD_KOKKOS ON)
 endif()

cmake/common/set/sparta_cmake_defaults.cmake

@@ -49,6 +49,10 @@ if(SPARTA_ENABLE_TESTING)
       "surf_react_heatflux"
       "chem_rates"
       "custom"
+      "tally_computes"
+      "ambi_3body"
+      "regions"
+      "qk"
       "explicit2implicit"
       "mfp_mct"
       "optmove"
@@ -89,9 +93,6 @@ if(SPARTA_ENABLE_TESTING)
   # the non-KOKKOS configurations.
   if(SPARTA_KOKKOS_EXACT)
     list(APPEND SPARTA_DISABLED_TESTS
-        # external field fix not KOKKOS-enabled
-        "in.bfield"
-        "in.bfield.grid"
         # VTK dump styles have no KOKKOS variant
         "in.vtk"
         "in.vtk.3d"

doc/Section_accelerate.txt

@@ -530,6 +530,35 @@ non-Kokkos fix or compute, or performing I/O for "stat"_stats.html or
 "dump"_dump.html output will cause data to be copied back to the CPU
 incurring a performance penalty.
 
+NOTE: Most non-Kokkos styles degrade this way, costing performance but
+still producing correct results.  A few, however, are rejected outright
+and will stop the run, rather than falling back to the host.  Those
+restrictions are documented on the page for the style that imposes them.
+
+NOTE: A run may define any number of instances of the styles that the
+device kernels capture -- "surf_collide"_surf_collide.html,
+"surf_react"_surf_react.html, and the per-event tally computes such as
+"compute boundary"_compute_boundary.html, "compute
+surf"_compute_surf.html, "compute isurf/grid"_compute_isurf_grid.html,
+"compute react/boundary"_compute_react_boundary.html, "compute
+react/surf"_compute_react_surf.html and "compute
+react/isurf/grid"_compute_react_isurf_grid.html.  Each is held in a
+runtime-sized device buffer, so there is no compile-time cap.  The one
+remaining exception is that at most two "compute
+surf"_compute_surf.html instances may tally for a single "fix
+emit/surf"_fix_emit_surf.html; exceeding that stops the run with an
+explanatory message.
+
+NOTE: Building with -DSPARTA_KOKKOS_FIXED_LISTS instead holds those
+styles in fixed-size arrays inside the kernel functor, which restores
+the former caps: at most two instances of each
+"surf_react"_surf_react.html style, at most four surf react models in
+total, at most two active instances of each per-event tally compute
+listed above, and at most four gas-phase tally computes.  The two
+layouts produce identical results; which is faster is a
+hardware-dependent question, so both are kept so they can be compared
+by rebuilding.
+
 [Run with the KOKKOS package by editing an input script:]
 
 Alternatively the effect of the "-sf" or "-pk" switches can be

doc/Section_commands.txt

@@ -326,7 +326,7 @@ section"_#cmd_4 lists many of the same commands, grouped by category.
 "create_box"_create_box.html,
 "create_grid"_create_grid.html,
 "create_isurf"_create_isurf.html,
-"create_particles"_create_particles.html,
+"create_particles (k)"_create_particles.html,
 "custom"_custom.html,
 "dimension"_dimension.html,
 "dump"_dump.html,
@@ -357,9 +357,9 @@ section"_#cmd_4 lists many of the same commands, grouped by category.
 "read_isurf"_read_isurf.html,
 "read_particles"_read_particles.html,
 "read_restart"_read_restart.html,
-"read_surf"_read_surf.html,
+"read_surf (k)"_read_surf.html,
 "region"_region.html,
-"remove_surf"_remove_surf.html,
+"remove_surf (k)"_remove_surf.html,
 "reset_timestep"_reset_timestep.html,
 "restart"_restart.html,
 "run"_run.html,
@@ -410,14 +410,14 @@ This is indicated by additional letters in parenthesis: k = KOKKOS.
 "dt/reset (k)"_fix_dt_reset.html,
 "emit/face (k)"_fix_emit_face.html,
 "emit/face/file"_fix_emit_face_file.html,
-"emit/surf"_fix_emit_surf.html,
-"field/grid"_fix_field_grid.html,
-"field/particle"_fix_field_particle.html,
+"emit/surf (k)"_fix_emit_surf.html,
+"field/grid (k)"_fix_field_grid.html,
+"field/particle (k)"_fix_field_particle.html,
 "grid/check (k)"_fix_grid_check.html,
 "halt"_fix_halt.html,
 "move/surf (k)"_fix_move_surf.html,
 "print"_fix_print.html,
-"surf/temp"_fix_surf_temp.html,
+"surf/temp (k)"_fix_surf_temp.html,
 "temp/global/rescale (k)"_fix_temp_global_rescale.html,
 "temp/rescale (k)"_fix_temp_rescale.html,
 "vibmode (k)"_fix_vibmode.html :tb(c=6,ea=c)
@@ -440,24 +440,24 @@ letters in parenthesis: k = KOKKOS.
 "eflux/grid (k)"_compute_eflux_grid.html,
 "fft/grid (k)"_compute_fft_grid.html,
 "gas/collision/grid (k)"_compute_gas_collision_grid.html,
-"gas/collision/tally"_compute_gas_collision_tally.html,
+"gas/collision/tally (k)"_compute_gas_collision_tally.html,
 "gas/reaction/grid (k)"_compute_gas_reaction_grid.html,
-"gas/reaction/tally"_compute_gas_reaction_tally.html,
+"gas/reaction/tally (k)"_compute_gas_reaction_tally.html,
 "grid (k)"_compute_grid.html,
 "isurf/grid (k)"_compute_isurf_grid.html,
 "ke/particle (k)"_compute_ke_particle.html,
 "lambda/grid (k)"_compute_lambda_grid.html,
 "pflux/grid (k)"_compute_pflux_grid.html,
 "property/grid (k)"_compute_property_grid.html,
 "property/surf (k)"_compute_property_surf.html,
-"react/boundary"_compute_react_boundary.html,
+"react/boundary (k)"_compute_react_boundary.html,
 "react/surf (k)"_compute_react_surf.html,
 "react/isurf/grid (k)"_compute_react_isurf_grid.html,
-"reduce"_compute_reduce.html,
+"reduce (k)"_compute_reduce.html,
 "sonine/grid (k)"_compute_sonine_grid.html,
 "surf (k)"_compute_surf.html,
-"surf/collision/tally"_compute_surf_collision_tally.html,
-"surf/reaction/tally"_compute_surf_reaction_tally.html,
+"surf/collision/tally (k)"_compute_surf_collision_tally.html,
+"surf/reaction/tally (k)"_compute_surf_reaction_tally.html,
 "temp (k)"_compute_temp.html,
 "thermal/grid (k)"_compute_thermal_grid.html,
 "tvib/grid (k)"_compute_tvib_grid.html :tb(c=6,ea=c)

doc/compute_gas_collision_grid.txt

@@ -55,6 +55,29 @@ for an overview of SPARTA output options.
 
 :line
 
+:line
+
+Styles with a {kk} suffix are functionally the same as the
+corresponding style without the suffix.  They have been optimized to
+run faster, depending on your available hardware, as discussed in the
+"Accelerating SPARTA"_Section_accelerate.html section of the manual.
+The accelerated styles take the same arguments and should produce the
+same results, except for different random number, round-off and
+precision issues.
+
+These accelerated styles are part of the KOKKOS package. They are only
+enabled if SPARTA was built with that package.  See the "Making
+SPARTA"_Section_start.html#start_3 section for more info.
+
+You can specify the accelerated styles explicitly in your input script
+by including their suffix, or you can use the "-suffix command-line
+switch"_Section_start.html#start_7 when you invoke SPARTA, or you can
+use the "suffix"_suffix.html command in your input script.
+
+See the "Accelerating SPARTA"_Section_accelerate.html section of the
+manual for more instructions on how to use the accelerated styles
+effectively.
+
 [Restrictions:] none
 
 [Related commands:]

doc/compute_gas_collision_tally.txt

@@ -96,6 +96,29 @@ velocity keywords are in velocity units.
 
 :line
 
+:line
+
+Styles with a {kk} suffix are functionally the same as the
+corresponding style without the suffix.  They have been optimized to
+run faster, depending on your available hardware, as discussed in the
+"Accelerating SPARTA"_Section_accelerate.html section of the manual.
+The accelerated styles take the same arguments and should produce the
+same results, except for different random number, round-off and
+precision issues.
+
+These accelerated styles are part of the KOKKOS package. They are only
+enabled if SPARTA was built with that package.  See the "Making
+SPARTA"_Section_start.html#start_3 section for more info.
+
+You can specify the accelerated styles explicitly in your input script
+by including their suffix, or you can use the "-suffix command-line
+switch"_Section_start.html#start_7 when you invoke SPARTA, or you can
+use the "suffix"_suffix.html command in your input script.
+
+See the "Accelerating SPARTA"_Section_accelerate.html section of the
+manual for more instructions on how to use the accelerated styles
+effectively.
+
 [Restrictions:] none
 
 [Related commands:]

doc/compute_gas_reaction_grid.txt

@@ -95,6 +95,29 @@ options.
 
 :line
 
+:line
+
+Styles with a {kk} suffix are functionally the same as the
+corresponding style without the suffix.  They have been optimized to
+run faster, depending on your available hardware, as discussed in the
+"Accelerating SPARTA"_Section_accelerate.html section of the manual.
+The accelerated styles take the same arguments and should produce the
+same results, except for different random number, round-off and
+precision issues.
+
+These accelerated styles are part of the KOKKOS package. They are only
+enabled if SPARTA was built with that package.  See the "Making
+SPARTA"_Section_start.html#start_3 section for more info.
+
+You can specify the accelerated styles explicitly in your input script
+by including their suffix, or you can use the "-suffix command-line
+switch"_Section_start.html#start_7 when you invoke SPARTA, or you can
+use the "suffix"_suffix.html command in your input script.
+
+See the "Accelerating SPARTA"_Section_accelerate.html section of the
+manual for more instructions on how to use the accelerated styles
+effectively.
+
 [Restrictions:] none
 
 [Related commands:]

doc/compute_gas_reaction_tally.txt

@@ -130,6 +130,29 @@ velocity components are in velocity units.
 
 :line
 
+:line
+
+Styles with a {kk} suffix are functionally the same as the
+corresponding style without the suffix.  They have been optimized to
+run faster, depending on your available hardware, as discussed in the
+"Accelerating SPARTA"_Section_accelerate.html section of the manual.
+The accelerated styles take the same arguments and should produce the
+same results, except for different random number, round-off and
+precision issues.
+
+These accelerated styles are part of the KOKKOS package. They are only
+enabled if SPARTA was built with that package.  See the "Making
+SPARTA"_Section_start.html#start_3 section for more info.
+
+You can specify the accelerated styles explicitly in your input script
+by including their suffix, or you can use the "-suffix command-line
+switch"_Section_start.html#start_7 when you invoke SPARTA, or you can
+use the "suffix"_suffix.html command in your input script.
+
+See the "Accelerating SPARTA"_Section_accelerate.html section of the
+manual for more instructions on how to use the accelerated styles
+effectively.
+
 [Restrictions:] none
 
 [Related commands:]

doc/compute_react_boundary.txt

@@ -7,6 +7,7 @@
 :line
 
 compute react/boundary command :h3
+compute react/boundary/kk command :h3
 
 [Syntax:]
 
@@ -88,6 +89,29 @@ on each face.
 
 :line
 
+:line
+
+Styles with a {kk} suffix are functionally the same as the
+corresponding style without the suffix.  They have been optimized to
+run faster, depending on your available hardware, as discussed in the
+"Accelerating SPARTA"_Section_accelerate.html section of the manual.
+The accelerated styles take the same arguments and should produce the
+same results, except for different random number, round-off and
+precision issues.
+
+These accelerated styles are part of the KOKKOS package. They are only
+enabled if SPARTA was built with that package.  See the "Making
+SPARTA"_Section_start.html#start_3 section for more info.
+
+You can specify the accelerated styles explicitly in your input script
+by including their suffix, or you can use the "-suffix command-line
+switch"_Section_start.html#start_7 when you invoke SPARTA, or you can
+use the "suffix"_suffix.html command in your input script.
+
+See the "Accelerating SPARTA"_Section_accelerate.html section of the
+manual for more instructions on how to use the accelerated styles
+effectively.
+
 [Restrictions:] none
 
 [Related commands:]

doc/compute_reduce.txt

@@ -7,6 +7,7 @@
 :line
 
 compute reduce command :h3
+compute reduce/kk command :h3
 
 [Syntax:]
 
@@ -263,6 +264,38 @@ overview of SPARTA output options.
 The scalar or vector values will be in whatever "units"_units.html the
 quantities being reduced are in.
 
+:line
+
+Styles with a {kk} suffix are functionally the same as the
+corresponding style without the suffix.  They have been optimized to
+run faster, depending on your available hardware, as discussed in the
+"Accelerating SPARTA"_Section_accelerate.html section of the manual.
+The accelerated styles take the same arguments and should produce the
+same results, except for different random number, round-off and
+precision issues.
+
+These accelerated styles are part of the KOKKOS package. They are only
+enabled if SPARTA was built with that package.  See the "Making
+SPARTA"_Section_start.html#start_3 section for more info.
+
+For the {kk} style, an input is reduced on the device when its source is
+device-resident: the explicit per-particle attributes ({x}, {vx}, {ke},
+{erot}, {evib}, etc), per-particle and per-grid custom attributes, and
+the per-particle or per-grid output of a compute or fix that itself has
+a {kk} variant.  Every other input -- per-surf inputs, {v_name}
+variables, and the output of a non-Kokkos compute or fix -- is reduced
+on the host, after the data it reads is copied back.  This is a
+performance distinction only; the result is the same either way.
+
+You can specify the accelerated styles explicitly in your input script
+by including their suffix, or you can use the "-suffix command-line
+switch"_Section_start.html#start_7 when you invoke SPARTA, or you can
+use the "suffix"_suffix.html command in your input script.
+
+See the "Accelerating SPARTA"_Section_accelerate.html section of the
+manual for more instructions on how to use the accelerated styles
+effectively.
+
 [Restrictions:] none
 
 [Related commands:]

doc/compute_surf_collision_tally.txt

@@ -112,6 +112,29 @@ are {vx/post}, {vy/post}, {vz/post}.
 
 :line
 
+:line
+
+Styles with a {kk} suffix are functionally the same as the
+corresponding style without the suffix.  They have been optimized to
+run faster, depending on your available hardware, as discussed in the
+"Accelerating SPARTA"_Section_accelerate.html section of the manual.
+The accelerated styles take the same arguments and should produce the
+same results, except for different random number, round-off and
+precision issues.
+
+These accelerated styles are part of the KOKKOS package. They are only
+enabled if SPARTA was built with that package.  See the "Making
+SPARTA"_Section_start.html#start_3 section for more info.
+
+You can specify the accelerated styles explicitly in your input script
+by including their suffix, or you can use the "-suffix command-line
+switch"_Section_start.html#start_7 when you invoke SPARTA, or you can
+use the "suffix"_suffix.html command in your input script.
+
+See the "Accelerating SPARTA"_Section_accelerate.html section of the
+manual for more instructions on how to use the accelerated styles
+effectively.
+
 [Restrictions:] none
 
 [Related commands:]