amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-03-23 02:37:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Updated sup[mt] Makefiles for variable dim ranges.

Browse Source 
Details:
- Updated test/sup/Makefile and test/supmt/Makefile to allow specifying
  different problem size ranges for the drivers where one, two, or three
  matrix dimensions is large. This will facilitate the generation of
  more meaningful graphs, particularly when two dimensions are tiny.
This commit is contained in:
Field G. Van Zee
2020-03-02 15:06:48 -06:00
parent 31f11a06ea
commit 90db88e572
2 changed files with 59 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
test/sup/Makefile
View File

@@ -142,14 +142,17 @@ VENDORP_LIB := $(MKLP_LIB)
#
# Single core
PS_BEGIN := 4
PS_MAX := 800
PS_INC := 4
PS_BEGIN_3L := 2
PS_MAX_3L := 400
PS_INC_3L := 2
# Multicore
P1_BEGIN := 120
P1_MAX := 6000
P1_INC := 120
PS_BEGIN_2L := 4
PS_MAX_2L := 800
PS_INC_2L := 4
PS_BEGIN_1L := 32
PS_MAX_1L := 6400
PS_INC_1L := 32
#
@@ -186,21 +189,17 @@ CXXFLAGS += -I$(EIGEN_INC)
CXXFLAGS_ST := -march=native $(subst -fopenmp,,$(CXXFLAGS))
CXXFLAGS_MT := -march=native $(CXXFLAGS)
# Single or multithreaded string
# Single or multithreaded string.
STR_ST := -DTHR_STR=\"st\"
STR_MT := -DTHR_STR=\"mt\"
# Number of trials per problem size.
N_TRIALS := -DN_TRIALS=3
# Problem size specification
PDEF_ST := -DP_BEGIN=$(PS_BEGIN) \
-DP_MAX=$(PS_MAX) \
-DP_INC=$(PS_INC)
PDEF_MT := -DP_BEGIN=$(P1_BEGIN) \
-DP_MAX=$(P1_MAX) \
-DP_INC=$(P1_INC)
# Problem size specification.
PDEF_ST_1L := -DP_BEGIN=$(PS_BEGIN_1L) -DP_MAX=$(PS_MAX_1L) -DP_INC=$(PS_INC_1L)
PDEF_ST_2L := -DP_BEGIN=$(PS_BEGIN_2L) -DP_MAX=$(PS_MAX_2L) -DP_INC=$(PS_INC_2L)
PDEF_ST_3L := -DP_BEGIN=$(PS_BEGIN_3L) -DP_MAX=$(PS_MAX_3L) -DP_INC=$(PS_INC_3L)
ifeq ($(E),1)
ERRCHK := -DERROR_CHECK
@@ -262,6 +261,15 @@ get-1of3 = $(word 1,$(subst _, ,$(1)))
get-2of3 = $(word 2,$(subst _, ,$(1)))
get-3of3 = $(word 3,$(subst _, ,$(1)))
# A function to return the correct PDEFS_ST variable given the shape string.
get-pdefs = $(strip $(subst l_l_l,$(PDEF_ST_3L), \
$(subst l_l_s,$(PDEF_ST_2L), \
$(subst l_s_l,$(PDEF_ST_2L), \
$(subst s_l_l,$(PDEF_ST_2L), \
$(subst s_s_l,$(PDEF_ST_1L), \
$(subst s_l_s,$(PDEF_ST_1L), \
$(subst l_s_s,$(PDEF_ST_1L),$(1)))))))))
# Datatype defs.
get-dt-cpp = $(strip \
$(if $(findstring s,$(1)),-DDT=BLIS_FLOAT -DIS_FLOAT,\
@@ -416,7 +424,7 @@ EIMPLS := eigen
# for the various implementations.
define make-st-rule
test_$(1)gemm_$(call stripu,$(2))_$(call stripu,$(3))_$(call get-shape-dim-str,$(4),$(5),$(6),$(7))_$(8)_st.o: test_gemm.c Makefile
$(CC) $(CFLAGS) $(ERRCHK) $(N_TRIALS) $(PDEF_ST) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_ST) -c $$< -o $$@
$(CC) $(CFLAGS) $(ERRCHK) $(N_TRIALS) $(call get-pdefs,$(4)) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_ST) -c $$< -o $$@
endef
# Instantiate the rule function make-st-rule() for each BLIS/BLAS/CBLAS
@@ -435,7 +443,7 @@ $(eval $(call make-st-rule,$(dt),$(tr),$(st),$(sh),$(sm),$(sn),$(sk),$(impl)))))
# for the various implementations.
define make-eigst-rule
test_$(1)gemm_$(call stripu,$(2))_$(call stripu,$(3))_$(call get-shape-dim-str,$(4),$(5),$(6),$(7))_$(8)_st.o: test_gemm.c Makefile
$(CXX) $(CXXFLAGS_ST) $(ERRCHK) $(N_TRIALS) $(PDEF_ST) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_ST) -c $$< -o $$@
$(CXX) $(CXXFLAGS_ST) $(ERRCHK) $(N_TRIALS) $(call get-pdefs,$(4)) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_ST) -c $$< -o $$@
endef
# Instantiate the rule function make-st-rule() for each Eigen implementation.

54
test/supmt/Makefile
View File

@@ -141,15 +141,18 @@ VENDORP_LIB := $(MKLP_LIB)
# --- Problem size definitions -------------------------------------------------
#
# Single core
PS_BEGIN := 4
PS_MAX := 800
PS_INC := 4
# Multicore
P1_BEGIN := 8
P1_MAX := 1600
P1_INC := 8
P1_BEGIN_3L := 4
P1_MAX_3L := 800
P1_INC_3L := 4
P1_BEGIN_2L := 8
P1_MAX_2L := 1600
P1_INC_2L := 8
P1_BEGIN_1L := 64
P1_MAX_1L := 12800
P1_INC_1L := 64
#
@@ -186,21 +189,17 @@ CXXFLAGS += -I$(EIGEN_INC)
CXXFLAGS_ST := -march=native $(subst -fopenmp,,$(CXXFLAGS))
CXXFLAGS_MT := -march=native $(CXXFLAGS)
# Single or multithreaded string
# Single or multithreaded string.
STR_ST := -DTHR_STR=\"st\"
STR_MT := -DTHR_STR=\"mt\"
# Number of trials per problem size.
N_TRIALS := -DN_TRIALS=3
# Problem size specification
PDEF_ST := -DP_BEGIN=$(PS_BEGIN) \
-DP_MAX=$(PS_MAX) \
-DP_INC=$(PS_INC)
PDEF_MT := -DP_BEGIN=$(P1_BEGIN) \
-DP_MAX=$(P1_MAX) \
-DP_INC=$(P1_INC)
# Problem size specification.
PDEF_MT_1L := -DP_BEGIN=$(P1_BEGIN_1L) -DP_MAX=$(P1_MAX_1L) -DP_INC=$(P1_INC_1L)
PDEF_MT_2L := -DP_BEGIN=$(P1_BEGIN_2L) -DP_MAX=$(P1_MAX_2L) -DP_INC=$(P1_INC_2L)
PDEF_MT_3L := -DP_BEGIN=$(P1_BEGIN_3L) -DP_MAX=$(P1_MAX_3L) -DP_INC=$(P1_INC_3L)
ifeq ($(E),1)
ERRCHK := -DERROR_CHECK
@@ -262,6 +261,15 @@ get-1of3 = $(word 1,$(subst _, ,$(1)))
get-2of3 = $(word 2,$(subst _, ,$(1)))
get-3of3 = $(word 3,$(subst _, ,$(1)))
# A function to return the correct PDEFS_ST variable given the shape string.
get-pdefs = $(strip $(subst l_l_l,$(PDEF_MT_3L), \
$(subst l_l_s,$(PDEF_MT_2L), \
$(subst l_s_l,$(PDEF_MT_2L), \
$(subst s_l_l,$(PDEF_MT_2L), \
$(subst s_s_l,$(PDEF_MT_1L), \
$(subst s_l_s,$(PDEF_MT_1L), \
$(subst l_s_s,$(PDEF_MT_1L),$(1)))))))))
# Datatype defs.
get-dt-cpp = $(strip \
$(if $(findstring s,$(1)),-DDT=BLIS_FLOAT -DIS_FLOAT,\
@@ -453,9 +461,11 @@ EIMPLS := eigen
# Define the function that will be used to instantiate compilation rules
# for the various single-threaded implementations.
# NOTE: The rules from this function are not yet working since this Makefile
# targets multithreaded, not single-threaded, drivers.
define make-st-rule
test_$(1)gemm_$(call stripu,$(2))_$(call stripu,$(3))_$(call get-shape-dim-str,$(4),$(5),$(6),$(7))_$(8)_st.o: test_gemm.c Makefile
$(CC) $(CFLAGS) $(ERRCHK) $(N_TRIALS) $(PDEF_ST) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_ST) -c $$< -o $$@
$(CC) $(CFLAGS) $(ERRCHK) $(N_TRIALS) $(call get-pdefs,$(4)) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_ST) -c $$< -o $$@
endef
# Instantiate the rule function make-st-rule() for each BLIS/BLAS/CBLAS
@@ -474,7 +484,7 @@ $(eval $(call make-st-rule,$(dt),$(tr),$(st),$(sh),$(sm),$(sn),$(sk),$(impl)))))
# for the various multithreaded implementations.
define make-mt-rule
test_$(1)gemm_$(call stripu,$(2))_$(call stripu,$(3))_$(call get-shape-dim-str,$(4),$(5),$(6),$(7))_$(8)_mt.o: test_gemm.c Makefile
$(CC) $(CFLAGS) $(ERRCHK) $(N_TRIALS) $(PDEF_MT) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_MT) -c $$< -o $$@
$(CC) $(CFLAGS) $(ERRCHK) $(N_TRIALS) $(call get-pdefs,$(4)) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_MT) -c $$< -o $$@
endef
# Instantiate the rule function make-mt-rule() for each BLIS/BLAS/CBLAS
@@ -491,9 +501,11 @@ $(eval $(call make-mt-rule,$(dt),$(tr),$(st),$(sh),$(sm),$(sn),$(sk),$(impl)))))
# Define the function that will be used to instantiate compilation rules
# for the single-threaded Eigen implementation.
# NOTE: The rules from this function are not yet working since this Makefile
# targets multithreaded, not single-threaded, drivers.
define make-eigst-rule
test_$(1)gemm_$(call stripu,$(2))_$(call stripu,$(3))_$(call get-shape-dim-str,$(4),$(5),$(6),$(7))_$(8)_st.o: test_gemm.c Makefile
$(CXX) $(CXXFLAGS_ST) $(ERRCHK) $(N_TRIALS) $(PDEF_ST) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_ST) -c $$< -o $$@
$(CXX) $(CXXFLAGS_ST) $(ERRCHK) $(N_TRIALS) $(call get-pdefs,$(4)) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_ST) -c $$< -o $$@
endef
# Instantiate the rule function make-st-rule() for each Eigen implementation.
@@ -511,7 +523,7 @@ $(eval $(call make-eigst-rule,$(dt),$(tr),$(st),$(sh),$(sm),$(sn),$(sk),$(impl))
# for the multithreaded Eigen implementation.
define make-eigmt-rule
test_$(1)gemm_$(call stripu,$(2))_$(call stripu,$(3))_$(call get-shape-dim-str,$(4),$(5),$(6),$(7))_$(8)_mt.o: test_gemm.c Makefile
$(CXX) $(CXXFLAGS_MT) $(ERRCHK) $(N_TRIALS) $(PDEF_MT) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_MT) -c $$< -o $$@
$(CXX) $(CXXFLAGS_MT) $(ERRCHK) $(N_TRIALS) $(call get-pdefs,$(4)) $(call get-dt-cpp,$(1)) $(call get-tra-defs,$(2)) $(call get-sto-defs,$(3)) $(call get-shape-defs,$(4),$(5),$(6),$(7)) $(call get-imp-defs,$(8)) $(STR_MT) -c $$< -o $$@
endef
# Instantiate the rule function make-st-rule() for each Eigen implementation.