Meson build

- tools/expand.py can take input/output files as arguments (for meson)
- tools/expand.py handles the C++ string wrapping
- drop int128 use

.gitignore

@@ -1,19 +1,12 @@
 gpuowl
 version.inc
-*-wrap.cpp
-*-wrap.cl
 version.new
 config.txt
 gpuowl.log
 worktodo.txt
-results.txt
-results.old
-sent.txt
-sent.old
 *.owl
 *.bak
 *.o
-*.so
 tmp/
 tmp?/
 cuda/
@@ -31,3 +24,4 @@ comgr.log
 \#*\#
 trashbin
 proof
+build/

Gpu.cpp

@@ -25,7 +25,6 @@
 #include <bitset>
 #include <limits>
 #include <iomanip>
-#include <quadmath.h>
 
 #ifndef M_PIl
 #define M_PIl 3.141592653589793238462643383279502884L
@@ -73,8 +72,8 @@ pair<T, T> root1(u32 N, u32 k) {
     assert(!(N&7));
     assert(k <= N/8);
     N /= 2;
-    __float128 angle = - M_PIq * k / N;
-    return {cosq(angle), sinq(angle)};
+    auto angle = - M_PIl * k / N;
+    return {cosl(angle), sinl(angle)};
   }
 }
 
@@ -143,12 +142,12 @@ u32 kAt(u32 H, u32 line, u32 col) { return (line + col * H) * 2; }
 
 auto weight(u32 N, u32 E, u32 H, u32 line, u32 col, u32 rep) {
   auto iN = 1 / (f128) N;
-  return exp2q(iN * extra(N, E, kAt(H, line, col) + rep));
+  return exp2l(iN * extra(N, E, kAt(H, line, col) + rep));
 }
 
 auto invWeight(u32 N, u32 E, u32 H, u32 line, u32 col, u32 rep) {
   auto iN = 1 / (f128) N;
-  return exp2q(- iN * extra(N, E, kAt(H, line, col) + rep));
+  return exp2l(- iN * extra(N, E, kAt(H, line, col) + rep));
 }
 
 double boundUnderOne(double x) { return std::min(x, nexttoward(1, 0)); }

Makefile

@@ -2,7 +2,7 @@ CXXFLAGS = -Wall -g -O3 -std=gnu++17
 
 LIBPATH = -L/opt/rocm-5.1.1/opencl/lib -L/opt/rocm-4.0.0/opencl/lib -L/opt/rocm-3.3.0/opencl/lib/x86_64 -L/opt/rocm/opencl/lib -L/opt/rocm/opencl/lib/x86_64 -L/opt/amdgpu-pro/lib/x86_64-linux-gnu -L.
 
-LDFLAGS = -lstdc++fs -lOpenCL -lgmp -pthread -lquadmath ${LIBPATH}
+LDFLAGS = -lstdc++fs -lOpenCL -lgmp -pthread ${LIBPATH}
 
 LINK = $(CXX) $(CXXFLAGS) -o $@ ${OBJS} ${LDFLAGS}
 
@@ -26,7 +26,7 @@ D:	D.o Pm1Plan.o log.o common.o timeutil.o
 	$(CXX) -o $@ $^ ${LDFLAGS}
 
 clean:
-	rm -f ${OBJS} gpuowl gpuowl-win.exe
+	rm -f ${OBJS} gpuowl gpuowl-win.exe gpuowl-wrap.cpp
 
 %.o : %.cpp
 %.o : %.cpp $(DEPDIR)/%.d gpuowl-wrap.cpp version.inc
@@ -41,11 +41,8 @@ version.inc: FORCE
 	diff -q -N version.new version.inc >/dev/null || mv version.new version.inc
 	echo Version: `cat version.inc`
 
-gpuowl-expanded.cl: gpuowl.cl
-	python3 tools/expand.py < gpuowl.cl > gpuowl-expanded.cl
-
-gpuowl-wrap.cpp: gpuowl-expanded.cl head.txt tail.txt
-	cat head.txt gpuowl-expanded.cl tail.txt > gpuowl-wrap.cpp
+gpuowl-wrap.cpp: gpuowl.cl
+	python3 tools/expand.py < gpuowl.cl > gpuowl-wrap.cpp
 
 FORCE:
 

README.md

@@ -99,13 +99,35 @@ exponent and optionally the assignment ID (AID) from PrimeNet. The `PRPDC=` pref
 * Run `gpuowl`. It prints progress report on stdout and in `gpuowl.log`, and writes result lines to `results.txt`
 * Submit the result lines from `results.txt` to https://www.mersenne.org/manual_result/ manual testing.
 
+
 ## Build
-To build simply invoke "`make`" (or look inside the Makefile for a manual build).
 
+Prerequisites (please install these):
 * the GNU Multiple Precision (GMP) 6.1 library `libgmp-dev`
 * a C++20 compiler (e.g. GCC, Clang)
 * an OpenCL implementation (which provides the **libOpenCL** library). Recommended: an AMD GPU with ROCm 1.7.
 
+### Meson build
+Example build steps on linux:
+```
+cd gpuowl
+mkdir build
+cd build
+meson ..
+ninja
+```
+
+What the previous commands do:
+- go to gpuowl source directory
+- create a subdirectory named "build"
+- go to the build directory
+- invoke meson, passing as argument the gpuowl source directory (.. in this situation)
+- run ninja to build
+
+### Make build
+To build simply invoke "`make`" (or look inside the Makefile for a manual build).
+
+
 ## See \"`gpuowl -h`\" for the command line options.
 
 ## Self-test

meson.build

@@ -0,0 +1,29 @@
+# -*- mode:Python -*-
+
+project('gpuowl', 'cpp', default_options:['cpp_std=c++17', 'buildtype=debugoptimized'])
+
+vcs_tag(input:'version.inc.in', output:'version.inc')
+
+python3 = import('python').find_installation('python3')
+expander = find_program('./tools/expand.py')
+
+wrap = generator(expander, output:'@[email protected]', arguments:['@INPUT@', '@OUTPUT@'])
+
+gpuowl_wrap = wrap.process('gpuowl.cl')
+
+srcs = 'ProofCache.cpp Proof.cpp Pm1Plan.cpp B1Accumulator.cpp Memlock.cpp log.cpp md5.cpp sha3.cpp AllocTrac.cpp GmpUtil.cpp FFTConfig.cpp Worktodo.cpp common.cpp main.cpp Gpu.cpp clwrap.cpp Task.cpp Saver.cpp timeutil.cpp Args.cpp state.cpp Signal.cpp'.split()
+
+cpp = meson.get_compiler('cpp')
+amdocl = cpp.find_library('amdocl64', dirs:['/opt/rocm-3.3.0/opencl/lib/x86_64', '/opt/rocm-5.1.1/opencl/lib', '/opt/rocm/opencl/lib'])
+
+
+executable('gpuowl', sources: srcs + [gpuowl_wrap], dependencies:[amdocl, dependency('gmp')])
+
+
+# Meson experiments below:
+
+#executable('gpuowl', sources:srcs, link_with:[amdocl, dependency('gmp')])
+# amdocl = declare_dependency(link_args : ['-L/opt/rocm-3.3.0/opencl/lib/x86_64', '-lamdocl64'])
+# dependency('amdocl64')
+# thread_dep = dependency('threads')
+# gpuowl_wrap = custom_target('gpuowl_wrap', command:[python3, '@INPUT@', '@OUTPUT@'], input:['tools/expand.py', 'gpuowl.cl'], output:'gpuowl-wrap.cpp')

tools/expand.py

@@ -2,11 +2,23 @@
 
 import sys
 
+# If any arguments are passed, expect <input-file> <output-file>
+# Otherwise read from stdin, write to stdout
+if len(sys.argv) > 1:
+    assert len(sys.argv) == 3, f'Use: f{sys.argv[0]} <input-file> <output-file>'
+    sys.stdin  = open(sys.argv[1])
+    sys.stdout = open(sys.argv[2], 'w')
+
+HEAD = 'const char *CL_SOURCE = R"clsource('
+TAIL = ')clsource";';
+
 lineNo = 0
 current = None
 body = None
 macros = {}
 
+print(HEAD)
+
 def err(text):
     print(f'#{lineNo}', text, file=sys.stderr)
     exit(1)
@@ -44,4 +56,7 @@ def err(text):
         if current:
             body += line
         else:
-            print(line,end='')
+            print(line, end='')
+
+print(TAIL)
+

version.inc.in

@@ -0,0 +1 @@
+"@VCS_TAG@"