initial rocBLAS logic files for iGPUs

- add initial rocBLAS logic files for rembrandt (gfx1035), raphael (gfx1036) and phoenix (gfx1103) iGPUs. - when testing with the https://github.com/LeiWang1999/rocblas-benchmark by using the std::make_tuple(8192, 8192, 8192, false, false, enable_tune), the speedup was about 4-5x. - gfx1035 without logic files Device 0: AMD Radeon Graphics m,n,k,a_t,b_t,enable_tune,fp32 time (msec),fp16-f32 time (msec), f16-f16 time (msec), int8-int32 time (msec) 8192,8192,8192,n,n,0,912.287,814.502,854.257,865.103 - gfx1035 with logic files Device 0: AMD Radeon Graphics m,n,k,a_t,b_t,enable_tune,fp32 time (msec),fp16-f32 time (msec), f16-f16 time (msec), int8-int32 time (msec) 8192,8192,8192,n,n,0,652.499,834.796,237.42,189.945 - gfx1103 without logic files Device 0: AMD Radeon 780M m,n,k,a_t,b_t,enable_tune,fp32 time (msec),fp16-f32 time (msec), f16-f16 time (msec), int8-int32 time (msec) 8192,8192,8192,n,n,0,916.684,820.721,823.48,1018.46 - gfx1103 with logic files ROCR_VISIBLE_DEVICES="1" ./rocblas_benchmark Device 0: AMD Radeon 780M m,n,k,a_t,b_t,enable_tune,fp32 time (msec),fp16-f32 time (msec), f16-f16 time (msec), int8-int32 time (msec) 8192,8192,8192,n,n,0,1346.02,634.836,193.613,119.29 Signed-off-by: Mika Laitio <[email protected]>
lamikr · Jul 17, 2024 · ee4b0fa · ee4b0fa
1 parent 44e5e79
commit ee4b0fa
Show file tree

Hide file tree

Showing 5 changed files with 2,909,834 additions and 9 deletions.
diff --git a/patches/rocm-6.1.2/rocBLAS/0001-add-mageia-9-support-to-install.sh.patch b/patches/rocm-6.1.2/rocBLAS/0001-add-mageia-9-support-to-install.sh.patch
@@ -1,7 +1,7 @@
-From b4e555f2d5c996b528cc13602f78012671383f3a Mon Sep 17 00:00:00 2001
+From c651124a45f972bdb57e15d55c80b34007c55240 Mon Sep 17 00:00:00 2001
 From: Mika Laitio <[email protected]>
 Date: Sat, 18 May 2024 18:17:42 -0700
-Subject: [PATCH 1/3] add mageia 9 support to install.sh
+Subject: [PATCH 1/5] add mageia 9 support to install.sh
 
 Signed-off-by: Mika Laitio <[email protected]>
 ---
@@ -54,5 +54,5 @@ index fc644b87..46c95775 100755
        elevate_if_not_root zypper -n --no-gpg-checks install rocblas-*.rpm
      ;;
 -- 
-2.45.2
+2.41.1
 
diff --git a/patches/rocm-6.1.2/rocBLAS/0002-add-gfx1035-gfx1036-and-gfx1103-to-gpulist.patch b/patches/rocm-6.1.2/rocBLAS/0002-add-gfx1035-gfx1036-and-gfx1103-to-gpulist.patch
@@ -1,7 +1,7 @@
-From 98c87b3db281d5048524ecb0c14c33c1fac0719c Mon Sep 17 00:00:00 2001
+From 5f26b83d1d6483decbd393ebadacfaf3cabcb2f8 Mon Sep 17 00:00:00 2001
 From: Mika Laitio <[email protected]>
 Date: Sat, 18 May 2024 18:18:33 -0700
-Subject: [PATCH 2/3] add gfx1035,gfx1036 and gfx1103 to gpulist
+Subject: [PATCH 2/5] add gfx1035,gfx1036 and gfx1103 to gpulist
 
 Signed-off-by: Mika Laitio <[email protected]>
 ---
@@ -106,5 +106,5 @@ index 1f0349fd..073bb244 100644
      }
 
 -- 
-2.45.2
+2.41.1
 
diff --git a/patches/rocm-6.1.2/rocBLAS/0003-OpenBLAS-and-BLIS-library-search-improvements.patch b/patches/rocm-6.1.2/rocBLAS/0003-OpenBLAS-and-BLIS-library-search-improvements.patch
@@ -1,7 +1,7 @@
-From 8e12ea9b4770c29d36ae205d55e19f085229ab1a Mon Sep 17 00:00:00 2001
+From b1b3f0d3e9ea5d4dac0e9df3930ec6e3f188f461 Mon Sep 17 00:00:00 2001
 From: Mika Laitio <[email protected]>
 Date: Sat, 18 May 2024 18:15:13 -0700
-Subject: [PATCH 3/3] OpenBLAS and BLIS library search improvements
+Subject: [PATCH 3/5] OpenBLAS and BLIS library search improvements
 
 - OpenBLAS and BLIS can now be found from
   rocm_sdk build by rocm sdk builder
@@ -50,5 +50,5 @@ index dc8040ea..704414b5 100755
    else() # WIN32
      set( BLAS_INCLUDE_DIR ${OPENBLAS_DIR}/include CACHE PATH "OpenBLAS library include path" )
 -- 
-2.45.2
+2.41.1