aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request #1003 from ncos/knl-support

Browse Source 
KNL support via generic target.
This commit is contained in:
Dmitry Babokin
2015-04-03 15:47:44 +03:00
parent 5535bfc83f 74c1feed08
commit 8cc9d08f46
3 changed files with 3626 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
alloy.py
View File

@@ -293,8 +293,10 @@ def check_targets():
AVX = ["avx1-i32x4", "avx1-i32x8", "avx1-i32x16", "avx1-i64x4"] AVX = ["avx1-i32x4", "avx1-i32x8", "avx1-i32x16", "avx1-i64x4"]
AVX11 = ["avx1.1-i32x8","avx1.1-i32x16","avx1.1-i64x4"] AVX11 = ["avx1.1-i32x8","avx1.1-i32x16","avx1.1-i64x4"]
AVX2 = ["avx2-i32x8", "avx2-i32x16", "avx2-i64x4"] AVX2 = ["avx2-i32x8", "avx2-i32x16", "avx2-i64x4"]
KNL = ["knl"]
targets = [["AVX2", AVX2, False], ["AVX1.1", AVX11, False], ["AVX", AVX, False], ["SSE4", SSE4, False], ["SSE2", SSE2, False]] targets = [["AVX2", AVX2, False], ["AVX1.1", AVX11, False], ["AVX", AVX, False], ["SSE4", SSE4, False],
["SSE2", SSE2, False], ["KNL", KNL, False]]
f_lines = take_lines("check_isa.exe", "first") f_lines = take_lines("check_isa.exe", "first")
for i in range(0,5): for i in range(0,5):
if targets[i][0] in f_lines: if targets[i][0] in f_lines:
@@ -329,6 +331,8 @@ def check_targets():
# here we have SDE # here we have SDE
f_lines = take_lines(sde_exists + " -help", "all") f_lines = take_lines(sde_exists + " -help", "all")
for i in range(0,len(f_lines)): for i in range(0,len(f_lines)):
if targets[5][2] == False and "knl" in f_lines[i]:
answer_sde = answer_sde + [["-knl", "knl"]]
if targets[3][2] == False and "wsm" in f_lines[i]: if targets[3][2] == False and "wsm" in f_lines[i]:
answer_sde = answer_sde + [["-wsm", "sse4-i32x4"], ["-wsm", "sse4-i32x8"], ["-wsm", "sse4-i16x8"], ["-wsm", "sse4-i8x16"]] answer_sde = answer_sde + [["-wsm", "sse4-i32x4"], ["-wsm", "sse4-i32x8"], ["-wsm", "sse4-i16x8"], ["-wsm", "sse4-i8x16"]]
if targets[2][2] == False and "snb" in f_lines[i]: if targets[2][2] == False and "snb" in f_lines[i]:
@@ -587,6 +591,7 @@ def validation_run(only, only_targets, reference_branch, number, notify, update,
LLVM = [newest_LLVM, "trunk"] LLVM = [newest_LLVM, "trunk"]
gen_archs = ["x86-64"] gen_archs = ["x86-64"]
knc_archs = ["x86-64"] knc_archs = ["x86-64"]
knl_archs = ["x86-64"]
need_LLVM = check_LLVM(LLVM) need_LLVM = check_LLVM(LLVM)
for i in range(0,len(need_LLVM)): for i in range(0,len(need_LLVM)):
build_LLVM(need_LLVM[i], "", "", "", False, False, False, True, False, make, options.gcc_toolchain_path) build_LLVM(need_LLVM[i], "", "", "", False, False, False, True, False, make, options.gcc_toolchain_path)
@@ -608,14 +613,16 @@ def validation_run(only, only_targets, reference_branch, number, notify, update,
# sometimes clang++ is not avaluable. if --ispc-build-compiler = gcc we will pass in g++ compiler # sometimes clang++ is not avaluable. if --ispc-build-compiler = gcc we will pass in g++ compiler
if options.ispc_build_compiler == "gcc": if options.ispc_build_compiler == "gcc":
stability.compiler_exe = "g++" stability.compiler_exe = "g++"
# but 'knc' target is supported only by icpc, so set explicitly # but 'knc/knl' target is supported only by icpc, so set explicitly
if "knc" in targets[j]: if ("knc" in targets[j]) or ("knl" in targets[j]):
stability.compiler_exe = "icpc" stability.compiler_exe = "icpc"
# now set archs for targets # now set archs for targets
if "generic" in targets[j]: if "generic" in targets[j]:
arch = gen_archs arch = gen_archs
elif "knc" in targets[j]: elif "knc" in targets[j]:
arch = knc_archs arch = knc_archs
elif "knl" in targets[j]:
arch = knl_archs
else: else:
arch = archs arch = archs
for i1 in range(0,len(arch)): for i1 in range(0,len(arch)):
@@ -634,10 +641,16 @@ def validation_run(only, only_targets, reference_branch, number, notify, update,
for j in range(0,len(sde_targets)): for j in range(0,len(sde_targets)):
stability.target = sde_targets[j][1] stability.target = sde_targets[j][1]
stability.wrapexe = os.environ["SDE_HOME"] + "/sde " + sde_targets[j][0] + " -- " stability.wrapexe = os.environ["SDE_HOME"] + "/sde " + sde_targets[j][0] + " -- "
for i1 in range(0,len(archs)): if "knc" in stability.target:
arch = knc_archs
elif "knl" in stability.target:
arch = knl_archs
else:
arch = archs
for i1 in range(0,len(arch)):
for i2 in range(0,len(opts)): for i2 in range(0,len(opts)):
for i3 in range(dbg_begin,dbg_total): for i3 in range(dbg_begin,dbg_total):
stability.arch = archs[i1] stability.arch = arch[i1]
stability.no_opt = opts[i2] stability.no_opt = opts[i2]
stability.ispc_flags = ispc_flags_tmp stability.ispc_flags = ispc_flags_tmp
if (i3 != 0): if (i3 != 0):
@@ -888,8 +901,8 @@ if __name__ == '__main__':
"Try to build compiler with all LLVM\n\talloy.py -r --only=build\n" + "Try to build compiler with all LLVM\n\talloy.py -r --only=build\n" +
"Performance validation run with 10 runs of each test and comparing to branch 'old'\n\talloy.py -r --only=performance --compare-with=old --number=10\n" + "Performance validation run with 10 runs of each test and comparing to branch 'old'\n\talloy.py -r --only=performance --compare-with=old --number=10\n" +
"Validation run. Update fail_db.txt with new fails, send results to my@my.com\n\talloy.py -r --update-errors=F --notify='my@my.com'\n" + "Validation run. Update fail_db.txt with new fails, send results to my@my.com\n\talloy.py -r --update-errors=F --notify='my@my.com'\n" +
"Test KNC target (not tested when tested all supported targets, so should be set explicitly via --only-targets)\n\talloy.py -r --only='stability' --only-targets='knc'\n") "Test KNC target (not tested when tested all supported targets, so should be set explicitly via --only-targets)\n\talloy.py -r --only='stability' --only-targets='knc'\n" +
"Test KNL target (requires sde)\n\talloy.py -r --only='stability' --only-targets='knl'\n")
num_threads="%s" % multiprocessing.cpu_count() num_threads="%s" % multiprocessing.cpu_count()
parser = MyParser(usage="Usage: alloy.py -r/-b [options]", epilog=examples) parser = MyParser(usage="Usage: alloy.py -r/-b [options]", epilog=examples)

3583
examples/intrinsics/knl.h Normal file
View File

File diff suppressed because it is too large Load Diff

29
run_tests.py
View File

@@ -165,6 +165,9 @@ def run_test(testname):
if (options.target == "knc"): if (options.target == "knc"):
ispc_cmd = ispc_exe_rel + " --werror --nowrap %s --arch=%s --target=%s" % \ ispc_cmd = ispc_exe_rel + " --werror --nowrap %s --arch=%s --target=%s" % \
(filename, options.arch, "generic-16") (filename, options.arch, "generic-16")
elif (options.target == "knl"):
ispc_cmd = ispc_exe_rel + " --werror --nowrap %s --arch=%s --target=%s" % \
(filename, options.arch, "generic-16")
else: else:
ispc_cmd = ispc_exe_rel + " --werror --nowrap %s --arch=%s --target=%s" % \ ispc_cmd = ispc_exe_rel + " --werror --nowrap %s --arch=%s --target=%s" % \
(filename, options.arch, options.target) (filename, options.arch, options.target)
@@ -264,6 +267,9 @@ def run_test(testname):
if (options.target == "knc"): if (options.target == "knc"):
cc_cmd = "%s -O2 -I. %s %s test_static.cpp -DTEST_SIG=%d %s -o %s" % \ cc_cmd = "%s -O2 -I. %s %s test_static.cpp -DTEST_SIG=%d %s -o %s" % \
(options.compiler_exe, gcc_arch, "-mmic", match, obj_name, exe_name) (options.compiler_exe, gcc_arch, "-mmic", match, obj_name, exe_name)
elif (options.target == "knl"):
cc_cmd = "%s -O2 -I. %s %s test_static.cpp -DTEST_SIG=%d %s -o %s" % \
(options.compiler_exe, gcc_arch, "-xMIC-AVX512", match, obj_name, exe_name)
else: else:
cc_cmd = "%s -O2 -I. %s %s test_static.cpp -DTEST_SIG=%d %s -o %s" % \ cc_cmd = "%s -O2 -I. %s %s test_static.cpp -DTEST_SIG=%d %s -o %s" % \
(options.compiler_exe, gcc_arch, gcc_isa, match, obj_name, exe_name) (options.compiler_exe, gcc_arch, gcc_isa, match, obj_name, exe_name)
@@ -285,6 +291,9 @@ def run_test(testname):
if (options.target == "knc"): if (options.target == "knc"):
ispc_cmd = ispc_exe_rel + " --woff %s -o %s --arch=%s --target=%s" % \ ispc_cmd = ispc_exe_rel + " --woff %s -o %s --arch=%s --target=%s" % \
(filename, obj_name, options.arch, "generic-16") (filename, obj_name, options.arch, "generic-16")
elif (options.target == "knl"):
ispc_cmd = ispc_exe_rel + " --woff %s -o %s --arch=%s --target=%s" % \
(filename, obj_name, options.arch, "generic-16")
else: else:
ispc_cmd = ispc_exe_rel + " --woff %s -o %s --arch=%s --target=%s" % \ ispc_cmd = ispc_exe_rel + " --woff %s -o %s --arch=%s --target=%s" % \
(filename, obj_name, options.arch, options.target) (filename, obj_name, options.arch, options.target)
@@ -546,7 +555,7 @@ def verify():
"sse4-i8x16", "avx1-i32x4" "avx1-i32x8", "avx1-i32x16", "avx1-i64x4", "avx1.1-i32x8", "sse4-i8x16", "avx1-i32x4" "avx1-i32x8", "avx1-i32x16", "avx1-i64x4", "avx1.1-i32x8",
"avx1.1-i32x16", "avx1.1-i64x4", "avx2-i32x8", "avx2-i32x16", "avx2-i64x4", "avx1.1-i32x16", "avx1.1-i64x4", "avx2-i32x8", "avx2-i32x16", "avx2-i64x4",
"generic-1", "generic-4", "generic-8", "generic-1", "generic-4", "generic-8",
"generic-16", "generic-32", "generic-64", "knc"]] "generic-16", "generic-32", "generic-64", "knc", "knl"]]
for i in range (0,len(f_lines)): for i in range (0,len(f_lines)):
if f_lines[i][0] == "%": if f_lines[i][0] == "%":
continue continue
@@ -622,7 +631,7 @@ def run_tests(options1, args, print_version):
global is_generic_target global is_generic_target
is_generic_target = ((options.target.find("generic-") != -1 and is_generic_target = ((options.target.find("generic-") != -1 and
options.target != "generic-1" and options.target != "generic-x1") or options.target != "generic-1" and options.target != "generic-x1") or
options.target == "knc") options.target == "knc" or options.target == "knl")
global is_nvptx_target global is_nvptx_target
is_nvptx_target = (options.target.find("nvptx") != -1) is_nvptx_target = (options.target.find("nvptx") != -1)
@@ -650,10 +659,15 @@ def run_tests(options1, args, print_version):
elif options.target == "knc": elif options.target == "knc":
error("No knc #include specified; using examples/intrinsics/knc.h\n", 2) error("No knc #include specified; using examples/intrinsics/knc.h\n", 2)
options.include_file = "examples/intrinsics/knc.h" options.include_file = "examples/intrinsics/knc.h"
elif options.target == "knl":
error("No knl #include specified; using examples/intrinsics/knl.h\n", 2)
options.include_file = "examples/intrinsics/knl.h"
if options.compiler_exe == None: if options.compiler_exe == None:
if (options.target == "knc"): if (options.target == "knc"):
options.compiler_exe = "icpc" options.compiler_exe = "icpc"
elif (options.target == "knl"):
options.compiler_exe = "icpc"
elif is_windows: elif is_windows:
options.compiler_exe = "cl.exe" options.compiler_exe = "cl.exe"
else: else:
@@ -699,7 +713,7 @@ def run_tests(options1, args, print_version):
OS = "Linux" OS = "Linux"
if not (OS == 'Linux'): if not (OS == 'Linux'):
error ("knc target supported only on Linux", 1) error ("knc target is supported only on Linux", 1)
# if no specific test files are specified, run all of the tests in tests/, # if no specific test files are specified, run all of the tests in tests/,
# failing_tests/, and tests_errors/ # failing_tests/, and tests_errors/
if len(args) == 0: if len(args) == 0:
@@ -890,11 +904,12 @@ if __name__ == "__main__":
parser.add_option("-f", "--ispc-flags", dest="ispc_flags", help="Additional flags for ispc (-g, -O1, ...)", parser.add_option("-f", "--ispc-flags", dest="ispc_flags", help="Additional flags for ispc (-g, -O1, ...)",
default="") default="")
parser.add_option('-t', '--target', dest='target', parser.add_option('-t', '--target', dest='target',
help='Set compilation target (sse2-i32x4, sse2-i32x8, sse4-i32x4, sse4-i32x8, sse4-i16x8, sse4-i8x16, avx1-i32x8, avx1-i32x16, avx1.1-i32x8, avx1.1-i32x16, avx2-i32x8, avx2-i32x16, generic-x1, generic-x4, generic-x8, generic-x16, generic-x32, generic-x64, knc)', help=('Set compilation target (sse2-i32x4, sse2-i32x8, sse4-i32x4, sse4-i32x8, ' +
default="sse4") 'sse4-i16x8, sse4-i8x16, avx1-i32x8, avx1-i32x16, avx1.1-i32x8, avx1.1-i32x16, ' +
'avx2-i32x8, avx2-i32x16, generic-x1, generic-x4, generic-x8, generic-x16, ' +
'generic-x32, generic-x64, knc, knl)'), default="sse4")
parser.add_option('-a', '--arch', dest='arch', parser.add_option('-a', '--arch', dest='arch',
help='Set architecture (arm, x86, x86-64)', help='Set architecture (arm, x86, x86-64)',default="x86-64")
default="x86-64")
parser.add_option("-c", "--compiler", dest="compiler_exe", help="C/C++ compiler binary to use to run tests", parser.add_option("-c", "--compiler", dest="compiler_exe", help="C/C++ compiler binary to use to run tests",
default=None) default=None)
parser.add_option('-o', '--no-opt', dest='no_opt', help='Disable optimization', parser.add_option('-o', '--no-opt', dest='no_opt', help='Disable optimization',