Implement blocking, loop unrolling

Reimplemented blocking of permutation ES. Additionally, the loops
responsible for updating n_as_extreme have been restructured to
facilitate unrolling by the compiler; this change greatly reduces
the runtime on CPUs with small cache memory.

Author: TylerSagendorf

DESCRIPTION

@@ -1,8 +1,8 @@
 Package: fast.ssgsea
 Type: Package
 Title: High-Performance Gene Set Enrichment Analysis (HP-GSEA)
-Version: 0.1.0.9030
-Date: 2026-03-10
+Version: 0.1.0.9031
+Date: 2026-03-15
 Authors@R: 
       person(given = "Tyler", family = "Sagendorf", 
              email = "tyler.sagendorf@pnnl.gov", 

README.Rmd

@@ -48,24 +48,27 @@ No Windows binary is available, so [Rtools](https://cran.r-project.org/bin/windo
 
 ### Linux
 
-Most Linux distributions come pre-packaged with tools to compile C and C++ code, so no extra work needs to be done. Users can install the development version of `fast.ssgsea` on Linux by running the code below.
+Most Linux distributions come pre-packaged with tools to compile C and C++ code, so no extra work is needed. Users can install the development version of `fast.ssgsea` on Linux by running the code below.
 
 ### Install
 
-The development version of `fast.ssgsea` can be installed with
+The development version of `fast.ssgsea` can be installed with any of the following
 
 ``` r
 # install.packages("pak")
 pak::pak("pnnl/fast.ssgsea")
 ```
 
-or
-
 ``` r
 # install.packages("devtools")
 devtools::install_github("pnnl/fast.ssgsea")
 ```
 
+``` r
+# install.packages("renv")
+renv::install("pnnl/fast.ssgsea")
+```
+
 ## Usage
 
 ### Simulate Data
@@ -131,7 +134,7 @@ print(sessionInfo(), locale = FALSE, tzone = FALSE)
 
 ## Performance
 
-Tests were performed on a desktop computer with an AMD Ryzen 5 7600X CPU running at 4.7 GHz, single threaded, to measure the runtime of fast-ssGSEA (`fast.ssgsea::fast_ssgsea`) and FGSEA-simple (`fgsea::fgseaSimple`). Different combinations of the number of gene sets, maximum gene set size, and the number of permutations were tested in a random order (3 replicates each) to minimize the influence of previous runs. The R scripts and data are available in the simulation/ folder.
+Tests were performed on a desktop computer with an AMD Ryzen 5 7600X CPU running at 4.7 GHz, single threaded, to measure the runtime of fast-ssGSEA (`fast.ssgsea::fast_ssgsea`) and FGSEA-simple (`fgsea::fgseaSimple`). Different combinations of the number of gene sets, maximum gene set size, and the number of permutations ($\pi$) were tested in a random order (3 replicates each) to minimize the influence of previous runs. The R scripts and data are available in the simulation/ folder.
 
 ### fast-ssGSEA
 

README.md

@@ -93,25 +93,29 @@ installed to compile C and C++ code. Then, the development version of
 ### Linux
 
 Most Linux distributions come pre-packaged with tools to compile C and
-C++ code, so no extra work needs to be done. Users can install the
-development version of `fast.ssgsea` on Linux by running the code below.
+C++ code, so no extra work is needed. Users can install the development
+version of `fast.ssgsea` on Linux by running the code below.
 
 ### Install
 
-The development version of `fast.ssgsea` can be installed with
+The development version of `fast.ssgsea` can be installed with any of
+the following
 
 ``` r
 # install.packages("pak")
 pak::pak("pnnl/fast.ssgsea")
 ```
 
-or
-
 ``` r
 # install.packages("devtools")
 devtools::install_github("pnnl/fast.ssgsea")
 ```
 
+``` r
+# install.packages("renv")
+renv::install("pnnl/fast.ssgsea")
+```
+
 ## Usage
 
 ### Simulate Data
@@ -173,7 +177,7 @@ system.time({
 ```
 
     ##    user  system elapsed 
-    ##   1.323   0.106   1.348
+    ##   1.209   0.090   1.224
 
 ``` r
 str(res)
@@ -207,25 +211,26 @@ print(sessionInfo(), locale = FALSE, tzone = FALSE)
     ## [1] stats     graphics  grDevices utils     datasets  methods   base     
     ## 
     ## other attached packages:
-    ## [1] dqrng_0.4.1            fast.ssgsea_0.1.0.9030
+    ## [1] dqrng_0.4.1            fast.ssgsea_0.1.0.9031
     ## 
     ## loaded via a namespace (and not attached):
-    ##  [1] digest_0.6.37     collapse_2.1.3    fastmap_1.2.0     xfun_0.56        
-    ##  [5] knitr_1.50        parallel_4.5.2    htmltools_0.5.8.1 rmarkdown_2.29   
-    ##  [9] cli_3.6.5         data.table_1.17.8 compiler_4.5.2    rstudioapi_0.17.1
-    ## [13] tools_4.5.2       evaluate_1.0.5    Rcpp_1.1.0        yaml_2.3.10      
-    ## [17] rlang_1.1.7
+    ##  [1] digest_0.6.39       collapse_2.1.6      fastmap_1.2.0      
+    ##  [4] xfun_0.56           parallel_4.5.2      knitr_1.51         
+    ##  [7] htmltools_0.5.8.1   rmarkdown_2.29      cli_3.6.5          
+    ## [10] data.table_1.18.2.1 compiler_4.5.2      rstudioapi_0.18.0  
+    ## [13] tools_4.5.2         evaluate_1.0.5      Rcpp_1.1.1         
+    ## [16] yaml_2.3.12         otel_0.2.0          rlang_1.1.7
 
 ## Performance
 
 Tests were performed on a desktop computer with an AMD Ryzen 5 7600X CPU
 running at 4.7 GHz, single threaded, to measure the runtime of
 fast-ssGSEA (`fast.ssgsea::fast_ssgsea`) and FGSEA-simple
 (`fgsea::fgseaSimple`). Different combinations of the number of gene
-sets, maximum gene set size, and the number of permutations were tested
-in a random order (3 replicates each) to minimize the influence of
-previous runs. The R scripts and data are available in the simulation/
-folder.
+sets, maximum gene set size, and the number of permutations ($\pi$) were
+tested in a random order (3 replicates each) to minimize the influence
+of previous runs. The R scripts and data are available in the
+simulation/ folder.
 
 ### fast-ssGSEA
 

man/figures/README-figure-1.png

@@ -141,17 +141,17 @@ time_res <- prepare_data("../data/fast-ssGSEA_timing_results.rds")
 
 # 10,000 permutations
 pa <- filter(time_res, nperm == levels(nperm)[1L]) %>%
-  plot_time(y_breaks = seq(0, 1, 0.25), y_max = 1, expand_upper = 5e-3) +
+  plot_time(y_breaks = seq(0, 1, 0.2), y_max = 1, expand_upper = 0.05) +
   labs(x = "")
 
 # 100,000 permutations
 pb <- filter(time_res, nperm == levels(nperm)[2L]) %>%
-  plot_time(y_breaks = seq(0, 2.5, 0.5), y_max = 2.5, expand_upper = 5e-3) +
+  plot_time(y_breaks = seq(0, 2, 0.4), y_max = 2, expand_upper = 0.05) +
   labs(y = "")
 
 # 1,000,000 permutations
 pc <- filter(time_res, nperm == levels(nperm)[3L]) %>%
-  plot_time(y_breaks = seq(0, 25, 5), y_max = 25, expand_upper = 5e-3) +
+  plot_time(y_breaks = seq(0, 12, 2), y_max = 12, expand_upper = 0.05) +
   labs(x = "", y = "")
 
 # Combine plots
@@ -186,17 +186,17 @@ time_res <- prepare_data("../data/FGSEA_timing_results_nproc_1.rds")
 
 # 10,000 permutations
 pa <- filter(time_res, nperm == levels(nperm)[1L]) %>%
-  plot_time(y_breaks = seq(0, 25, 5), y_max = 26, expand_upper = 5e-3) +
+  plot_time(y_breaks = seq(0, 24, 4), y_max = 26, expand_upper = 5e-3) +
   labs(x = "")
 
 # 100,000 permutations
 pb <- filter(time_res, nperm == levels(nperm)[2L]) %>%
-  plot_time(y_breaks = seq(0, 250, 50), y_max = 260, expand_upper = 5e-3) +
+  plot_time(y_breaks = seq(0, 240, 40), y_max = 260, expand_upper = 5e-3) +
   labs(y = "")
 
 # 1,000,000 permutations
 pc <- filter(time_res, nperm == levels(nperm)[3L]) %>%
-  plot_time(y_breaks = seq(0, 2500, 500), y_max = 2600, expand_upper = 5e-3) +
+  plot_time(y_breaks = seq(0, 2400, 400), y_max = 2600, expand_upper = 5e-3) +
   labs(x = "", y = "")
 
 # Combine plots