Added no AVX option for code generation

DomenicoCacace · Dec 31, 2020 · 4184a91 · 4184a91
1 parent 773e3ae
commit 4184a91
Show file tree

Hide file tree

Showing 4 changed files with 8 additions and 2 deletions.
diff --git a/scripts/djb_generator/builder.py b/scripts/djb_generator/builder.py
@@ -134,9 +134,14 @@ def main():
     argParser = argparse.ArgumentParser()
     argParser.add_argument("-ws", "--wordsize", help="The number of bits addressable by the CPU at once (default: 256)", default=256, type=int, dest="ws")
     argParser.add_argument("-s", "--digit-size", help="The number of bits per digit (default: 64)", default=64, type=int, dest="size")
+    argParser.add_argument("-n", "--no-avx", help="Generates code for machines not supporting AVX intrinsics", action="store_false", dest="AVX")
     args = argParser.parse_args()
 
     # Setting memory word size related parameters
+    if not args.AVX:
+        args.ws = args.size
+        k.AVX = False
+
     k.WS = args.ws
     k.DIGIT_SIZE_b = args.size
     k.DIGIT_SIZE_B = k.DIGIT_SIZE_b >> 3

diff --git a/scripts/djb_generator/constants.py b/scripts/djb_generator/constants.py
@@ -18,6 +18,7 @@
 DIGIT_SIZE_B = 0
 DIGIT_SIZE_b =0
 WS = 0
+AVX = True
 
 # Maximum size for direct multiplications
 CHUNK_SIZE = 9

diff --git a/scripts/djb_generator/generator.py b/scripts/djb_generator/generator.py
@@ -257,7 +257,7 @@ def scalarprod(nr, res, na, a0, a1, nb, b0, b1, offset=0):
 
 # Generates the code for a base multiplication
 def GF2X_MUL(nr, res, na, a, nb, b):
-    if na < k.CHUNK_SIZE:
+    if k.AVX and na < k.CHUNK_SIZE:
         return "gf2x_mul_" + str(na) + "_avx(" + res + ", " + a + ", " + b +");"
 
     return "GF2X_MUL(" + str(nr) + ", " + res + ", " + str(na) + ", " + a + ", " + str(nb) + ", " + b + ");"

diff --git a/src/benchmarking/CMakeLists.txt b/src/benchmarking/CMakeLists.txt
@@ -7,7 +7,7 @@ set(OPTIMIZATION_FLAGS "-O2 -fipa-cp-clone -floop-interchange -fgcse-after-reloa
                         -fsplit-loops -fsplit-paths -ftree-loop-distribution -ftree-loop-vectorize \
                         -ftree-partial-pre -funswitch-loops -fvect-cost-model -fvect-cost-model=dynamic -fversion-loops-for-strides")
 
-set(CMAKE_C_FLAGS  "${CMAKE_C_FLAGS} -Wall -pedantic -Wuninitialized -fno-var-tracking-assignments -march=native -O0 -g3 -Wno-unused-function")
+set(CMAKE_C_FLAGS  "${CMAKE_C_FLAGS} -Wall -pedantic -Wuninitialized -fno-var-tracking-assignments -march=native -O3 -g3 -Wno-unused-function")
 
 set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS} -Wall -pedantic -Wuninitialized -march=native -g3")
 message("FLAGS" ${CMAKE_C_FLAGS})