Merge pull request oneapi-src#2536 from mkrainiuk/devel

jimmytwei · web-flow · commit 02070b510688 · 2024-11-06T08:06:06.000-08:00
[oneMKL] add qmkl-sycl-impl option
diff --git a/Libraries/oneMKL/batched_linear_solver/README.md b/Libraries/oneMKL/batched_linear_solver/README.md
@@ -65,8 +65,8 @@ If everything is working correctly, the output should be similar to this:
 ```
 u172874@s001-n157:~/oneAPI-samples/Libraries/oneMKL/batched_linear_solver$ make
 ifx lu_solve_omp_offload.F90 -o lu_solve -i8 -free -qmkl
-ifx lu_solve_omp_offload.F90 -o lu_solve_omp_offload -i8 -free -qmkl -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -fsycl -L/glob/development-tools/versions/oneapi/2023.0.1/oneapi/mkl/2023.0.0/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -ldl
-ifx lu_solve_omp_offload_optimized.F90 -o lu_solve_omp_offload_optimized -i8 -free -qmkl -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -fsycl -L/glob/development-tools/versions/oneapi/2023.0.1/oneapi/mkl/2023.0.0/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -ldl
+ifx lu_solve_omp_offload.F90 -o lu_solve_omp_offload -i8 -free -qmkl -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -qmkl-sycl-impl=lapack
+ifx lu_solve_omp_offload_optimized.F90 -o lu_solve_omp_offload_optimized -i8 -free -qmkl -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -qmkl-sycl-impl=lapack
 ./lu_solve -n 64 -b 8 -r 1 -c 2
  Matrix dimensions:                    64
  Batch size:                     8
diff --git a/Libraries/oneMKL/batched_linear_solver/lu_solve_omp_offload.F90 b/Libraries/oneMKL/batched_linear_solver/lu_solve_omp_offload.F90
@@ -21,10 +21,9 @@
 !         lu_solve_omp_offload_ex1_timer.F90 -o lu_solve_ex1_timer
 !
 ! Compile for GPU:
-!     ifx -i8 -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -fsycl -free \
-!         lu_solve_omp_offload_ex1_timer.F90 -o lu_solve_ex1_omp_timer \
-!         -L${MKLROOT}/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 \
-!         -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -ldl
+!     ifx -i8 -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -free \
+!         -qmkl -qmkl-sycl-impl=lapack \
+!         lu_solve_omp_offload_ex1_timer.F90 -o lu_solve_ex1_omp_timer
 !
 ! Compile with -DSP to use single precision instead of double precision.
 !
diff --git a/Libraries/oneMKL/batched_linear_solver/lu_solve_omp_offload_optimized.F90 b/Libraries/oneMKL/batched_linear_solver/lu_solve_omp_offload_optimized.F90
@@ -21,10 +21,9 @@
 !         lu_solve_omp_offload_ex3_timer.F90 -o lu_solve_ex3_timer
 !
 ! Compile for GPU:
-!     ifx -i8 -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -fsycl -free \
-!         lu_solve_omp_offload_ex3_timer.F90 -o lu_solve_ex3_omp_timer \
-!         -L${MKLROOT}/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 \
-!         -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -ldl
+!     ifx -i8 -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -free \
+!         -qmkl -qmkl-sycl-impl=lapack \
+!         lu_solve_omp_offload_ex3_timer.F90 -o lu_solve_ex3_omp_timer
 !
 ! Compile with -DSP to use single precision instead of double precision.
 !
diff --git a/Libraries/oneMKL/batched_linear_solver/makefile b/Libraries/oneMKL/batched_linear_solver/makefile
@@ -10,7 +10,7 @@ run_all: lu_solve lu_solve_omp_offload lu_solve_omp_offload_optimized
 	./lu_solve_omp_offload_optimized -n 64 -b 8 -r 1 -c 2
 
 IFX_OPTS = -i8 -free -qmkl
-IFX_OPTS_OFFLOAD = -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -fsycl -L${MKLROOT}/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -ldl
+IFX_OPTS_OFFLOAD = -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -qmkl-sycl-impl=lapack
 
 lu_solve: lu_solve_omp_offload.F90
 	ifx $< -o $@ $(IFX_OPTS)
diff --git a/Libraries/oneMKL/binomial/GNUmakefile b/Libraries/oneMKL/binomial/GNUmakefile
@@ -2,11 +2,10 @@ all: binomial_sycl
 
 init_on_host ?= 0
 
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl=rng
 
 binomial_sycl: src/binomial_sycl.cpp src/binomial_main.cpp src/binomial.hpp
-	icpx -fsycl -O3 -DSMALL_OPT_N=0 -DVERBOSE=1 -DREPORT_COLD=1 -DREPORT_WARM=1 -DINIT_ON_HOST=$(init_on_host)  $(MKL_COPTS) -o $@ src/binomial_main.cpp src/binomial_sycl.cpp $(MKL_LIBS)
+	icpx -fsycl -O3 -DSMALL_OPT_N=0 -DVERBOSE=1 -DREPORT_COLD=1 -DREPORT_WARM=1 -DINIT_ON_HOST=$(init_on_host)  $(MKL_COPTS) -o $@ src/binomial_main.cpp src/binomial_sycl.cpp
 
 clean:
 	@rm -f binomial_sycl
diff --git a/Libraries/oneMKL/binomial/makefile b/Libraries/oneMKL/binomial/makefile
@@ -4,10 +4,10 @@ all: binomial_sycl.exe
 	INIT_ON_HOST=/DINIT_ON_HOST=1
 !endif
 
-DPCPP_OPTS=-O3 /I"$(MKLROOT)\include" /DMKL_ILP64 /DVERBOSE=1 /DSMALL_OPT_N=0 /DREPORT_COLD=1 /DREPORT_WARM=1 $(INIT_ON_HOST) -fsycl -qmkl -lOpenCL
+DPCPP_OPTS=-O3 /I"$(MKLROOT)\include" /DMKL_ILP64 /DVERBOSE=1 /DSMALL_OPT_N=0 /DREPORT_COLD=1 /DREPORT_WARM=1 $(INIT_ON_HOST) -fsycl /Qmkl /Qmkl-sycl-impl=rng OpenCL.lib
 
 binomial_sycl.exe: src\binomial_sycl.cpp src\binomial_main.cpp src\binomial.hpp
-	icpx $(DPCPP_OPTS) /DVERBOSE=1 /DSMALL_OPT_N=0 /DREPORT_COLD=1 /DREPORT_WARM=1 src\binomial_sycl.cpp src\binomial_main.cpp /obinomial_sycl.exe
+	icx $(DPCPP_OPTS) /DVERBOSE=1 /DSMALL_OPT_N=0 /DREPORT_COLD=1 /DREPORT_WARM=1 src\binomial_sycl.cpp src\binomial_main.cpp /obinomial_sycl.exe
 
 clean:
 	del /q binomial_sycl.exe
diff --git a/Libraries/oneMKL/black_scholes/GNUmakefile b/Libraries/oneMKL/black_scholes/GNUmakefile
@@ -7,11 +7,10 @@ all: black_scholes_sycl
 
 init_on_host ?= 0
 
-MKL_COPTS = -DMKL_ILP64 -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64 -qmkl -qmkl-sycl-impl=rng
 
 black_scholes_sycl: src/black_scholes_sycl.cpp
-	icpx -O3 -g -fsycl $(MKL_COPTS) $(MKL_LIBS) -DVERBOSE=1 -DSMALL_OPT_N=0 -DINIT_ON_HOST=$(init_on_host) -o $@  src/black_scholes_sycl.cpp
+	icpx -O3 -g -fsycl $(MKL_COPTS) -DVERBOSE=1 -DSMALL_OPT_N=0 -DINIT_ON_HOST=$(init_on_host) -o $@  src/black_scholes_sycl.cpp
 
 clean:
 	@rm -f black_scholes_sycl
diff --git a/Libraries/oneMKL/black_scholes/makefile b/Libraries/oneMKL/black_scholes/makefile
@@ -4,10 +4,10 @@ all: black_scholes_sycl.exe
         INIT_ON_HOST=/DINIT_ON_HOST=1
 !endif
 
-DPCPP_OPTS=-O3 /I"$(MKLROOT)\include" /DMKL_ILP64 /DVERBOSE=1 /DSMALL_OPT_N=0 $(INIT_ON_HOST) -fsycl -qmkl -lOpenCL
+DPCPP_OPTS=-O3 /I"$(MKLROOT)\include" /DMKL_ILP64 /DVERBOSE=1 /DSMALL_OPT_N=0 $(INIT_ON_HOST) -fsycl /Qmkl /Qmkl-sycl-impl=rng OpenCL.lib
 
 black_scholes_sycl.exe: src\black_scholes_sycl.cpp
-	icpx $(DPCPP_OPTS) src\black_scholes_sycl.cpp /oblack_scholes_sycl.exe
+	icx $(DPCPP_OPTS) src\black_scholes_sycl.cpp /oblack_scholes_sycl.exe
 
 clean:
 	del /q black_scholes_sycl.exe
diff --git a/Libraries/oneMKL/block_cholesky_decomposition/GNUmakefile b/Libraries/oneMKL/block_cholesky_decomposition/GNUmakefile
@@ -4,14 +4,13 @@ all: factor solve
 	./factor
 	./solve
 
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl="blas,lapack"
 
 factor: factor.cpp dpbltrf.cpp auxi.cpp
-	icpx $^ -o $@ -fsycl -fsycl-device-code-split=per_kernel $(MKL_COPTS) $(MKL_LIBS)
+	icpx $^ -o $@ -fsycl -fsycl-device-code-split=per_kernel $(MKL_COPTS)
 
 solve: solve.cpp dpbltrf.cpp dpbltrs.cpp auxi.cpp
-	icpx $^ -o $@ -fsycl -fsycl-device-code-split=per_kernel $(MKL_COPTS) $(MKL_LIBS)
+	icpx $^ -o $@ -fsycl -fsycl-device-code-split=per_kernel $(MKL_COPTS)
 
 clean:
 	-rm -f factor solve
diff --git a/Libraries/oneMKL/block_cholesky_decomposition/makefile b/Libraries/oneMKL/block_cholesky_decomposition/makefile
@@ -4,7 +4,7 @@ all: factor.exe solve.exe
 	.\factor.exe
 	.\solve.exe
 
-DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /DMKL_ILP64 /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
+DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /Qmkl-sycl-impl="blas,lapack" /DMKL_ILP64 /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
 
 factor.exe: factor.cpp dpbltrf.cpp auxi.cpp
 	icx-cl -fsycl factor.cpp dpbltrf.cpp auxi.cpp /Fefactor.exe $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/block_lu_decomposition/GNUmakefile b/Libraries/oneMKL/block_lu_decomposition/GNUmakefile
@@ -4,14 +4,13 @@ all: factor solve
 	./factor
 	./solve
 
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl="blas,lapack"
 
 factor: factor.cpp dgeblttrf.cpp auxi.cpp
-	icpx $^ -o $@ -fsycl -fsycl-device-code-split=per_kernel $(MKL_COPTS) $(MKL_LIBS)
+	icpx $^ -o $@ -fsycl -fsycl-device-code-split=per_kernel $(MKL_COPTS)
 
 solve: solve.cpp dgeblttrf.cpp dgeblttrs.cpp auxi.cpp
-	icpx $^ -o $@ -fsycl -fsycl-device-code-split=per_kernel $(MKL_COPTS) $(MKL_LIBS)
+	icpx $^ -o $@ -fsycl -fsycl-device-code-split=per_kernel $(MKL_COPTS)
 
 clean:
 	-rm -f factor solve genxir
diff --git a/Libraries/oneMKL/block_lu_decomposition/makefile b/Libraries/oneMKL/block_lu_decomposition/makefile
@@ -4,7 +4,7 @@ all: factor.exe solve.exe
 	.\factor.exe
 	.\solve.exe
 
-DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /DMKL_ILP64 /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
+DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /Qmkl-sycl-impl="blas,lapack" /DMKL_ILP64 /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
 
 factor.exe: factor.cpp dgeblttrf.cpp auxi.cpp
 	icx-cl -fsycl factor.cpp dgeblttrf.cpp auxi.cpp /Fefactor.exe $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/computed_tomography/GNUmakefile b/Libraries/oneMKL/computed_tomography/GNUmakefile
@@ -7,10 +7,9 @@ all: run
 run: computed_tomography
 	./computed_tomography 400 400 input.bmp radon.bmp restored.bmp
 
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl=dft
 
-DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel $(MKL_LIBS)
+DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel
 
 computed_tomography: computed_tomography.cpp
 	icpx $< -fsycl -o $@ $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/computed_tomography/makefile b/Libraries/oneMKL/computed_tomography/makefile
@@ -7,7 +7,7 @@ all: run
 run: computed_tomography.exe
 	.\computed_tomography.exe 400 400 input.bmp radon.bmp restored.bmp
 
-DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
+DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /Qmkl-sycl-impl=dft /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
 
 computed_tomography.exe: computed_tomography.cpp
 	icx-cl -fsycl computed_tomography.cpp /Fecomputed_tomography.exe $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/fourier_correlation/GNUmakefile b/Libraries/oneMKL/fourier_correlation/GNUmakefile
@@ -9,7 +9,7 @@ run_all: fcorr_1d_buff fcorr_1d_usm fcorr_2d_usm
 	./fcorr_1d_usm 4096
 	./fcorr_2d_usm
 
-DPCPP_OPTS = -DMKL_ILP64 -qmkl=parallel
+DPCPP_OPTS = -DMKL_ILP64 -qmkl -qmkl-sycl-impl="blas,dft,rng,vm"
 
 fcorr_1d_buff: fcorr_1d_buffers.cpp
 	icpx $< -fsycl -o $@ $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/fourier_correlation/makefile b/Libraries/oneMKL/fourier_correlation/makefile
@@ -9,7 +9,7 @@ run_all: fcorr_1d_buff.exe fcorr_1d_usm.exe fcorr_2d_usm.exe
 	.\fcorr_1d_usm 1024
 	.\fcorr_2d_usm
 
-DPCPP_OPTS=-DMKL_ILP64 -I"%MKLROOT%\include" /Qmkl OpenCL.lib /EHsc
+DPCPP_OPTS=-DMKL_ILP64 -I"%MKLROOT%\include" /Qmkl /Qmkl-sycl-impl="blas,dft,rng,vm" OpenCL.lib /EHsc
 
 fcorr_1d_buff.exe: fcorr_1d_buffers.cpp
 	icx-cl -fsycl fcorr_1d_buffers.cpp /Fefcorr_1d_buff.exe $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/guided_american_options_SYCLmigration/02_sycl_migrated_optimized/CMakeLists.txt b/Libraries/oneMKL/guided_american_options_SYCLmigration/02_sycl_migrated_optimized/CMakeLists.txt
@@ -6,7 +6,7 @@ if(USE_DEVICE_API)
 else()
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -DWITH_FUSED_BETA=1 ")
     add_executable (02_sycl_migrated_optimized_host longstaff_schwartz_svd_2.dp.cpp)
-    target_link_libraries(02_sycl_migrated_optimized_host sycl mkl_sycl mkl_core mkl_sequential mkl_intel_ilp64)
+    target_link_libraries(02_sycl_migrated_optimized_host sycl mkl_sycl_rng mkl_intel_ilp64 mkl_sequential mkl_core)
     add_custom_target (run_host_api cd ${CMAKE_SOURCE_DIR}/02_sycl_migrated_optimized/ && ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/02_sycl_migrated_optimized_host)
 endif()
 include_directories(${CMAKE_SOURCE_DIR}/02_sycl_migrated_optimized/include/)
diff --git a/Libraries/oneMKL/guided_cuBLAS_examples_SYCL_Migration/02_sycl_dpct_migrated/CMakeLists.txt b/Libraries/oneMKL/guided_cuBLAS_examples_SYCL_Migration/02_sycl_dpct_migrated/CMakeLists.txt
@@ -1,7 +1,7 @@
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -std=c++17")
-set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -lmkl_sycl -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -std=c++17 -qmkl")
+set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -qmkl -qmkl-sycl-impl=blas")
 
 include_directories(${CMAKE_SOURCE_DIR}/02_sycl_dpct_migrated/Common/)
 add_subdirectory("Level-1")
 add_subdirectory("Level-2")
-add_subdirectory("Level-3")
+add_subdirectory("Level-3")
diff --git a/Libraries/oneMKL/guided_cuRAND_examples_SYCL_migration/02_sycl_dpct_migrated/CMakeLists.txt b/Libraries/oneMKL/guided_cuRAND_examples_SYCL_migration/02_sycl_dpct_migrated/CMakeLists.txt
@@ -1,5 +1,5 @@
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -std=c++17")
-set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -lmkl_sycl -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -std=c++17 -qmkl")
+set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -qmkl -qmkl-sycl-impl=rng")
 
 include_directories(${CMAKE_SOURCE_DIR}/02_sycl_dpct_migrated/Common/)
 add_subdirectory("mrg32k3a")
@@ -9,4 +9,4 @@ add_subdirectory("philox")
 add_subdirectory("scrambled_sobol32")
 add_subdirectory("scrambled_sobol64")
 add_subdirectory("sobol32")
-add_subdirectory("sobol64")
+add_subdirectory("sobol64")
diff --git a/Libraries/oneMKL/guided_simpleCUBLAS_LU_SYCLMigration/02_sycl_migrated/CMakeLists.txt b/Libraries/oneMKL/guided_simpleCUBLAS_LU_SYCLMigration/02_sycl_migrated/CMakeLists.txt
@@ -1,20 +1,20 @@
 if(INTEL_MAX_GPU)
   #To build for Intel® Data Center GPU Max 1550 or 1100
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -fsycl-targets=intel_gpu_pvc -D INTEL_MAX_GPU")
-  set(CMAKE_EXE_LINKER_FLAGS "-qmkl ${CMAKE_EXE_LINKER_FLAGS}")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -fsycl-targets=intel_gpu_pvc -D INTEL_MAX_GPU -qmkl")
+  set(CMAKE_EXE_LINKER_FLAGS "-qmkl -qmkl-sycl-impl=lapack ${CMAKE_EXE_LINKER_FLAGS}")
 elseif (NVIDIA_GPU)
   #To build for Nvidia GPU like P100, A100, H100 ...
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -fsycl-targets=nvptx64-nvidia-cuda -D NVIDIA_GPU")
   set(CMAKE_EXE_LINKER_FLAGS "-L $ENV{ONEMKL_INSTALL_DIR}/lib/ -lonemkl ${CMAKE_EXE_LINKER_FLAGS}")
   include_directories($ENV{ONEMKL_INSTALL_DIR}/include/)
 elseif(FLOAT_TYPE)
   #To build for Gen11 as it doesn't support double precision
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -D FLOAT_TYPE")
-  set(CMAKE_EXE_LINKER_FLAGS "-qmkl ${CMAKE_EXE_LINKER_FLAGS}")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -D FLOAT_TYPE -qmkl")
+  set(CMAKE_EXE_LINKER_FLAGS "-qmkl -qmkl-sycl-impl=lapack ${CMAKE_EXE_LINKER_FLAGS}")
 else()
   #To build for Intel® UHD Graphics, Intel® Gen9, Gen11, Xeon CPU 
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl ")
-set(CMAKE_EXE_LINKER_FLAGS "-qmkl ${CMAKE_EXE_LINKER_FLAGS}")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -qmkl")
+set(CMAKE_EXE_LINKER_FLAGS "-qmkl -qmkl-sycl-impl=lapack ${CMAKE_EXE_LINKER_FLAGS}")
 endif()
 
 include_directories(${CMAKE_SOURCE_DIR}/02_sycl_migrated/Common/)
diff --git a/Libraries/oneMKL/matrix_mul_mkl/GNUmakefile b/Libraries/oneMKL/matrix_mul_mkl/GNUmakefile
@@ -9,10 +9,9 @@ run: matrix_mul_mkl
 	./matrix_mul_mkl double
 
 INCLUDE_COMMON=../../../common
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl=blas
 
-DPCPP_OPTS = -O2 $(MKL_COPTS) $(MKL_LIBS)
+DPCPP_OPTS = -O2 $(MKL_COPTS)
 
 matrix_mul_mkl: matrix_mul_mkl.cpp
 	icpx -fsycl -I$(INCLUDE_COMMON) $< -o $@ $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/matrix_mul_mkl/makefile b/Libraries/oneMKL/matrix_mul_mkl/makefile
@@ -8,7 +8,7 @@ run: matrix_mul_mkl.exe
 	.\matrix_mul_mkl.exe single
 	.\matrix_mul_mkl.exe double
 
-DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
+DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /Qmkl-sycl-impl=blas /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
 
 matrix_mul_mkl.exe: matrix_mul_mkl.cpp
 	icx-cl -fsycl matrix_mul_mkl.cpp /Fematrix_mul_mkl.exe $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/monte_carlo_european_opt/GNUmakefile b/Libraries/oneMKL/monte_carlo_european_opt/GNUmakefile
@@ -18,15 +18,14 @@ endif
 # setting initial random number generation on host
 init_on_host ?= 0
 
-MKL_COPTS = -DMKL_ILP64 $(GENERATOR) -qmkl=sequential -Wall -Wformat-security -Werror=format-security
-MKL_LIBS  = -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64 $(GENERATOR) -qmkl -qmkl-sycl-impl=rng -Wall -Wformat-security -Werror=format-security
 
-DPCPP_OPTS = $(MKL_COPTS) -fsycl -DINIT_ON_HOST=$(init_on_host) $(MKL_LIBS)
+DPCPP_OPTS = $(MKL_COPTS) -fsycl -DINIT_ON_HOST=$(init_on_host)
 
 montecarlo: src/montecarlo_main.cpp
 	icpx $< -o $@ $(DPCPP_OPTS)
 
 clean:
 	-rm -f montecarlo
 
-.PHONY: clean all
+.PHONY: clean all
diff --git a/Libraries/oneMKL/monte_carlo_european_opt/makefile b/Libraries/oneMKL/monte_carlo_european_opt/makefile
@@ -12,9 +12,10 @@ all: montecarlo
         INIT_ON_HOST=/DINIT_ON_HOST=1
 !endif
 
-DPCPP_OPTS=/I"$(MKLROOT)\include" /DMKL_ILP64 $(GENERATOR) -fsycl $(INIT_ON_HOST) -qmkl -lOpenCL
+DPCPP_OPTS=/I"$(MKLROOT)\include" /DMKL_ILP64 $(GENERATOR) /EHsc -fsycl $(INIT_ON_HOST) /Qmkl /Qmkl-sycl-impl=rng -lOpenCL
+
 montecarlo: src/montecarlo_main.cpp
-	icpx src/montecarlo_main.cpp /omontecarlo.exe $(DPCPP_OPTS)
+	icx src/montecarlo_main.cpp /omontecarlo.exe $(DPCPP_OPTS)
 
 clean:
 	del /q montecarlo.exe
diff --git a/Libraries/oneMKL/monte_carlo_pi/GNUmakefile b/Libraries/oneMKL/monte_carlo_pi/GNUmakefile
@@ -9,8 +9,7 @@ run: mc_pi mc_pi_usm mc_pi_device_api
 	./mc_pi_usm
 	./mc_pi_device_api
 
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl=rng
 
 DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations $(MKL_LIBS)
 
diff --git a/Libraries/oneMKL/monte_carlo_pi/makefile b/Libraries/oneMKL/monte_carlo_pi/makefile
@@ -9,7 +9,7 @@ run: mc_pi.exe mc_pi_usm.exe mc_pi_device_api.exe
 	.\mc_pi_usm.exe
 	.\mc_pi_device_api.exe
 
-DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /DMKL_ILP64 /EHsc -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations OpenCL.lib
+DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /Qmkl-sycl-impl=rng /DMKL_ILP64 /EHsc -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations OpenCL.lib
 
 mc_pi.exe: mc_pi.cpp
 	icx-cl -fsycl mc_pi.cpp /Femc_pi.exe $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/random_sampling_without_replacement/GNUmakefile b/Libraries/oneMKL/random_sampling_without_replacement/GNUmakefile
@@ -9,10 +9,9 @@ run: lottery lottery_usm lottery_device_api
 		./lottery_usm
 		./lottery_device_api
 
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl=rng
 
-DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations $(MKL_LIBS)
+DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations
 
 lottery: lottery.cpp
 		icpx $< -fsycl -o $@ $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/random_sampling_without_replacement/makefile b/Libraries/oneMKL/random_sampling_without_replacement/makefile
@@ -9,7 +9,7 @@ run: lottery.exe lottery_usm.exe lottery_device_api.exe
 	.\lottery_usm.exe
 	.\lottery_device_api.exe
 
-DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /DMKL_ILP64 /EHsc -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations OpenCL.lib
+DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /Qmkl-sycl-impl=rng /DMKL_ILP64 /EHsc -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations OpenCL.lib
 
 lottery.exe: lottery.cpp
 	icx-cl -fsycl lottery.cpp /Felottery.exe $(DPCPP_OPTS)
@@ -34,4 +34,4 @@ pseudo: clean run all
 # 		icpx $< -fsycl -o $@ $(DPCPP_OPTS)
 
 # lottery_device_api: lottery_device_api.cpp
-# 		icpx $< -fsycl -o $@ $(DPCPP_OPTS)
+# 		icpx $< -fsycl -o $@ $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/sparse_conjugate_gradient/GNUmakefile b/Libraries/oneMKL/sparse_conjugate_gradient/GNUmakefile
@@ -7,10 +7,9 @@ all: run
 run: sparse_cg
 	./sparse_cg
 
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl="blas,sparse"
 
-DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel $(MKL_LIBS)
+DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel
 
 sparse_cg: sparse_cg.cpp
 	icpx $< -fsycl -o $@ $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/sparse_conjugate_gradient/makefile b/Libraries/oneMKL/sparse_conjugate_gradient/makefile
@@ -7,7 +7,7 @@ all: run
 run: sparse_cg.exe
 	.\sparse_cg
 
-DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
+DPCPP_OPTS=/I"$(MKLROOT)\include" /Qmkl /Qmkl-sycl-impl="blas,sparse" /EHsc -fsycl-device-code-split=per_kernel OpenCL.lib
 
 sparse_cg.exe: sparse_cg.cpp
 	icx-cl -fsycl sparse_cg.cpp /Fesparse_cg.exe $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/student_t_test/GNUmakefile b/Libraries/oneMKL/student_t_test/GNUmakefile
@@ -8,10 +8,9 @@ run: t_test t_test_usm
 	./t_test
 	./t_test_usm
 
-MKL_COPTS = -DMKL_ILP64  -qmkl=sequential
-MKL_LIBS  = -lsycl -lOpenCL -lpthread -lm -ldl
+MKL_COPTS = -DMKL_ILP64  -qmkl -qmkl-sycl-impl="stats,rng"
 
-DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations $(MKL_LIBS)
+DPCPP_OPTS = $(MKL_COPTS) -fsycl-device-code-split=per_kernel -fno-sycl-early-optimizations
 
 t_test: t_test.cpp
 	icpx $< -fsycl -o $@ $(DPCPP_OPTS)
diff --git a/Libraries/oneMKL/student_t_test/makefile b/Libraries/oneMKL/student_t_test/makefile

Original file line number	Diff line number	Diff line change
`@@ -21,10 +21,9 @@`
`21`	`21`	`! lu_solve_omp_offload_ex1_timer.F90 -o lu_solve_ex1_timer`
`22`	`22`	`!`
`23`	`23`	`! Compile for GPU:`
`24`		`-! ifx -i8 -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -fsycl -free \`
`25`		`-! lu_solve_omp_offload_ex1_timer.F90 -o lu_solve_ex1_omp_timer \`
`26`		`-! -L${MKLROOT}/lib/intel64 -lmkl_sycl -lmkl_intel_ilp64 \`
`27`		`-! -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -ldl`
	`24`	`+! ifx -i8 -DMKL_ILP64 -qopenmp -fopenmp-targets=spir64 -free \`
	`25`	`+! -qmkl -qmkl-sycl-impl=lapack \`
	`26`	`+! lu_solve_omp_offload_ex1_timer.F90 -o lu_solve_ex1_omp_timer`
`28`	`27`	`!`
`29`	`28`	`! Compile with -DSP to use single precision instead of double precision.`
`30`	`29`	`!`