From 2d9b5647c30ff5f2f6fe073130b49e03e56093a6 Mon Sep 17 00:00:00 2001
From: Tom Rix <trix@redhat.com>
Date: Sat, 20 Jul 2024 06:16:21 -0600
Subject: [PATCH] Use fbgemm on 2.4

Signed-off-by: Tom Rix <trix@redhat.com>
---
 ...1-Add-cmake-option-USE_SYSTEM_FBGEMM.patch | 47 +++++++++++++++++++
 python-torch.spec                             | 31 ++++++++++--
 2 files changed, 73 insertions(+), 5 deletions(-)
 create mode 100644 next/0001-Add-cmake-option-USE_SYSTEM_FBGEMM.patch

diff --git a/next/0001-Add-cmake-option-USE_SYSTEM_FBGEMM.patch b/next/0001-Add-cmake-option-USE_SYSTEM_FBGEMM.patch
new file mode 100644
index 0000000..413c60d
--- /dev/null
+++ b/next/0001-Add-cmake-option-USE_SYSTEM_FBGEMM.patch
@@ -0,0 +1,47 @@
+From 091b7fe1ccbb5e4ff4ac6017d42bacb869f61a27 Mon Sep 17 00:00:00 2001
+From: Tom Rix <trix@redhat.com>
+Date: Sat, 20 Jul 2024 05:37:15 -0600
+Subject: [PATCH] Add cmake option USE_SYSTEM_FBGEMM
+
+Signed-off-by: Tom Rix <trix@redhat.com>
+---
+ CMakeLists.txt           | 1 +
+ cmake/Dependencies.cmake | 3 ++-
+ 2 files changed, 3 insertions(+), 1 deletion(-)
+
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index c4cd4b2c2a98..2068f7c6c4f2 100644
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -253,6 +253,7 @@ cmake_dependent_option(USE_STATIC_CUDNN "Use cuDNN static libraries" OFF
+                        "USE_CUDNN" OFF)
+ cmake_dependent_option(USE_CUSPARSELT "Use cuSPARSELt" ON "USE_CUDA" OFF)
+ option(USE_FBGEMM "Use FBGEMM (quantized 8-bit server operators)" ON)
++option(USE_SYSTEM_FBGEMM "Use system-wide FBGEMM" OFF)
+ option(USE_KINETO "Use Kineto profiling library" ON)
+ option(USE_CUPTI_SO "Use CUPTI as a shared library" ON)
+ option(USE_FAKELOWP "Use FakeLowp operators" OFF)
+diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
+index f1f2eb7cec31..192dac46f13b 100644
+--- a/cmake/Dependencies.cmake
++++ b/cmake/Dependencies.cmake
+@@ -706,6 +706,7 @@ endif()
+ 
+ # ---[ FBGEMM
+ if(USE_FBGEMM)
++  if (NOT USE_SYSTEM_FBGEMM)
+   set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party")
+   if(NOT DEFINED FBGEMM_SOURCE_DIR)
+     set(FBGEMM_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/fbgemm" CACHE STRING "FBGEMM source directory")
+@@ -746,7 +747,7 @@ if(USE_FBGEMM)
+       target_compile_options_if_supported(asmjit -Wno-unused-but-set-variable)
+     endif()
+   endif()
+-
++  endif()
+   if(USE_FBGEMM)
+     list(APPEND Caffe2_DEPENDENCY_LIBS fbgemm)
+   endif()
+-- 
+2.45.1
+
diff --git a/python-torch.spec b/python-torch.spec
index f27f099..3e74505 100644
--- a/python-torch.spec
+++ b/python-torch.spec
@@ -32,14 +32,14 @@
 %endif
 # hipblaslt is in development
 %bcond_with hipblaslt
-# magma is in development
-%bcond_with magma
 # Which families gpu build for
 %if 0%{?fedora} > 40
 %global rocm_gpu_list gfx8 gfx9 gfx10 gfx11 gfx90a gfx942 gfx1100
+%bcond_without magma
 %else
 # F40
 %global rocm_gpu_list gfx8 gfx9 gfx10 gfx11
+%bcond_with magma
 %endif
 %global rocm_default_gpu default
 %bcond_without rocm_loop
@@ -94,6 +94,16 @@
 %bcond_with pocketfft
 %endif
 
+%if 0%{?fedora} > 40
+%if %{with gitcommit}
+%bcond_without fbgemm
+%else
+%bcond_with fbgemm
+%endif
+%else
+%bcond_with fbgemm
+%endif
+
 # For testing cuda
 %ifarch x86_64
 %bcond_with cuda
@@ -230,6 +240,10 @@ Source80:       https://github.com/pytorch/kineto/archive/%{ki_commit}/kineto-%{
 %endif
 
 Patch0:        0001-no-third_party-foxi.patch
+%if %{with gitcommit}
+# https://github.com/pytorch/pytorch/pull/131282
+Patch1:        0001-Add-cmake-option-USE_SYSTEM_FBGEMM.patch
+%endif
 
 %if %{without gitcommit}
 Patch3:        0001-Stub-in-kineto-ActivityType.patch
@@ -265,7 +279,6 @@ Patch9:        0001-Enable-x86-CPU-vectorization-on-windows-submodule-sl.patch
 # For Python 3.13
 # https://github.com/pytorch/pytorch/pull/126033
 Patch10: 0001-Changes-to-compile-with-3.13-126033.patch
-
 %endif
 
 # ROCm patches
@@ -281,7 +294,6 @@ Patch102:      0001-silence-an-assert.patch
 %if %{without gitcommit}
 Patch103:      0001-can-not-use-with-c-files.patch
 %endif
-# Patch104:      0001-use-any-hip.patch
 Patch105:      0001-disable-use-of-aotriton.patch
 %endif
 Patch106:      0001-include-fmt-ranges.h-for-using-fmt-join.patch
@@ -292,6 +304,10 @@ ExclusiveArch:  x86_64 aarch64
 
 BuildRequires:  cmake
 BuildRequires:  eigen3-devel
+%if %{with fbgemm}
+BuildRequires:  asmjit-devel
+BuildRequires:  fbgemm-devel
+%endif
 BuildRequires:  flexiblas-devel
 BuildRequires:  fmt-devel
 %if %{with caffe2}
@@ -885,9 +901,14 @@ export INTERN_BUILD_MOBILE=OFF
 export USE_DISTRIBUTED=OFF
 export USE_CUDA=OFF
 export USE_FAKELOWP=OFF
+%if %{with fbgemm}
+export USE_FBGEMM=ON
+export USE_SYSTEM_FBGEMM=ON
+%else
 export USE_FBGEMM=OFF
+%endif
 export USE_FLASH_ATTENTION=OFF
-export USE_GOLD_LINKER=OFF
+export USE_GOLD_LINKER=ON
 export USE_GLOO=OFF
 export USE_ITT=OFF
 export USE_KINETO=OFF