From 00207e62b289f1d47acdeb58535650b07139c852 Mon Sep 17 00:00:00 2001
From: Luna Nova <git@lunnova.dev>
Date: Tue, 4 Nov 2025 19:41:04 -0800
Subject: [PATCH 1/4] rocmPackages.rocblas: remove reference to tensile in
 librocblas

rocblas's closure-size was bloated by around 400MB by a ref to
tensile in an error message string. tensile is used at build time
to build and link kernels, but is not needed at runtime.

Before:
/nix/store/zggwlwji0jca2cvhgl4ick7l3fwlwsfb-rocblas-gfx1030-6.4.3          1.3G

After:
/nix/store/71rz9qm323rzmzgkx31llcsnhd0gkvl3-rocblas-gfx1030-6.4.3        904.5M
---
 pkgs/development/rocm-modules/6/rocblas/default.nix | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/pkgs/development/rocm-modules/6/rocblas/default.nix b/pkgs/development/rocm-modules/6/rocblas/default.nix
index 0c9c2ee5c424..ffaac106de35 100644
--- a/pkgs/development/rocm-modules/6/rocblas/default.nix
+++ b/pkgs/development/rocm-modules/6/rocblas/default.nix
@@ -24,6 +24,7 @@
   python3Packages,
   rocm-smi,
   pkg-config,
+  removeReferencesTo,
   buildTensile ? true,
   buildTests ? false,
   buildBenchmarks ? false,
@@ -54,6 +55,7 @@ stdenv.mkDerivation (finalAttrs: {
     clr
     gitMinimal
     pkg-config
+    removeReferencesTo
   ]
   ++ lib.optionals buildTensile [
     tensile
@@ -148,6 +150,13 @@ stdenv.mkDerivation (finalAttrs: {
       --replace-fail '0.10' '1.0'
   '';
 
+  postInstall =
+    # tensile isn't needed at runtime and pulls in ~400MB of python deps
+    ''
+      remove-references-to -t ${tensile} \
+        "$out/lib/librocblas.so."*
+    '';
+
   passthru = {
     amdgpu_targets = gpuTargets';
     tests.rocblas-tests = finalAttrs.finalPackage.override {

From f9864971a4c6b7fc1638db60cc02dcd928ff2209 Mon Sep 17 00:00:00 2001
From: Luna Nova <git@lunnova.dev>
Date: Tue, 4 Nov 2025 19:55:21 -0800
Subject: [PATCH 2/4] rocmPackages: don't set build{Tests,Benchmarks}=false in
 scope

---
 pkgs/development/rocm-modules/6/default.nix | 2 --
 1 file changed, 2 deletions(-)

diff --git a/pkgs/development/rocm-modules/6/default.nix b/pkgs/development/rocm-modules/6/default.nix
index 6654bafef12d..44454b36958a 100644
--- a/pkgs/development/rocm-modules/6/default.nix
+++ b/pkgs/development/rocm-modules/6/default.nix
@@ -26,8 +26,6 @@ let
     {
       inherit rocmClangStdenv;
       stdenv = rocmClangStdenv;
-      buildTests = false;
-      buildBenchmarks = false;
 
       rocmUpdateScript = self.callPackage ./update.nix { };
 

From 5015e4b4468e63846ac14e9938f866d6c34c58e4 Mon Sep 17 00:00:00 2001
From: Luna Nova <git@lunnova.dev>
Date: Tue, 4 Nov 2025 19:54:45 -0800
Subject: [PATCH 3/4] rocmPackages.rocblas: split benchmark and test outputs

---
 .../rocm-modules/6/rocblas/default.nix        | 30 +++++++++++++------
 1 file changed, 21 insertions(+), 9 deletions(-)

diff --git a/pkgs/development/rocm-modules/6/rocblas/default.nix b/pkgs/development/rocm-modules/6/rocblas/default.nix
index ffaac106de35..37c303eca46f 100644
--- a/pkgs/development/rocm-modules/6/rocblas/default.nix
+++ b/pkgs/development/rocm-modules/6/rocblas/default.nix
@@ -26,8 +26,8 @@
   pkg-config,
   removeReferencesTo,
   buildTensile ? true,
-  buildTests ? false,
-  buildBenchmarks ? false,
+  buildTests ? true,
+  buildBenchmarks ? true,
   tensileSepArch ? true,
   tensileLazyLib ? true,
   withHipBlasLt ? true,
@@ -48,6 +48,8 @@ stdenv.mkDerivation (finalAttrs: {
     hash = "sha256-FCzo/BOk4xLEFkdOdqcCXh4a9t3/OIIBEy8oz6oOMWg=";
   };
 
+  outputs = [ "out" ] ++ lib.optional buildBenchmarks "benchmark" ++ lib.optional buildTests "test";
+
   nativeBuildInputs = [
     cmake
     # no ninja, it buffers console output and nix times out long periods of no output
@@ -78,10 +80,8 @@ stdenv.mkDerivation (finalAttrs: {
     python3Packages.msgpack
     python3Packages.zstandard
   ]
-  ++ lib.optionals buildTests [
-    gtest
-  ]
   ++ lib.optionals (buildTests || buildBenchmarks) [
+    gtest
     gfortran
     rocm-smi
   ]
@@ -155,14 +155,26 @@ stdenv.mkDerivation (finalAttrs: {
     ''
       remove-references-to -t ${tensile} \
         "$out/lib/librocblas.so."*
+    ''
+    + lib.optionalString buildBenchmarks ''
+      moveToOutput "bin/*-tune" "$benchmark"
+      moveToOutput "bin/*-bench" "$benchmark"
+      moveToOutput "bin/*example*" "$benchmark"
+      cp "$out/bin/"*.{yaml,txt} "$benchmark/bin"
+    ''
+    + lib.optionalString buildTests ''
+      moveToOutput "bin/*test*" "$test"
+      cp "$out/bin/"*.{yaml,txt} "$test/bin"
+    ''
+    + ''
+      if [ -d $out/bin ]; then
+        rm $out/bin/*.{yaml,txt} || true
+        rmdir $out/bin
+      fi
     '';
 
   passthru = {
     amdgpu_targets = gpuTargets';
-    tests.rocblas-tests = finalAttrs.finalPackage.override {
-      buildBenchmarks = true;
-      buildTests = true;
-    };
     updateScript = rocmUpdateScript {
       name = finalAttrs.pname;
       inherit (finalAttrs.src) owner;

From 9a22a291a158dfbe3c24fc1b2a8bf5392d931b44 Mon Sep 17 00:00:00 2001
From: Luna Nova <git@lunnova.dev>
Date: Tue, 4 Nov 2025 21:42:35 -0800
Subject: [PATCH 4/4] rocmPackages.rocblas: fix FHS diff path in rocblas test

---
 pkgs/development/rocm-modules/6/rocblas/default.nix | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/pkgs/development/rocm-modules/6/rocblas/default.nix b/pkgs/development/rocm-modules/6/rocblas/default.nix
index 37c303eca46f..c6bce658f6cc 100644
--- a/pkgs/development/rocm-modules/6/rocblas/default.nix
+++ b/pkgs/development/rocm-modules/6/rocblas/default.nix
@@ -7,6 +7,7 @@
   cmake,
   rocm-cmake,
   clr,
+  diffutils,
   python3,
   tensile,
   boost,
@@ -148,6 +149,12 @@ stdenv.mkDerivation (finalAttrs: {
     substituteInPlace CMakeLists.txt \
       --replace-fail "4.43.0" "4.44.0" \
       --replace-fail '0.10' '1.0'
+  ''
+  # Fixes sh: line 1: /usr/bin/diff: No such file or directory
+  # /build/source/clients/gtest/../include/testing_logging.hpp:1117: Failure
+  + lib.optionals buildTests ''
+    substituteInPlace clients/include/testing_logging.hpp \
+      --replace-fail "/usr/bin/diff" "${lib.getExe' diffutils "diff"}"
   '';
 
   postInstall =