mirrors/yosys
3
0
Fork 0
mirror of https://github.com/YosysHQ/yosys synced 2025-10-31 11:42:30 +00:00
Code Activity

cxxrtl: disable optimization of debug_items().

Browse source 
Implementing outlining has greatly increased the amount of debug
information in a typical build, and consequently exposed performance
issues in C++ compilers, which are similar for both GCC and Clang;
the compile time of Minerva SoC SRAM increased almost twofold.

Although one would expect the slowdown to be caused by the increased
use of templates in `debug_eval()`, it is actually almost entirely
attributable to optimizations and codegen for `debug_items()`.

Fortunately, it is neither possible nor desirable to optimize
`debug_items()`: in most cases it is called exactly once, and its
body is a linear sequence of calls with unique arguments.

This commit turns off optimizations for `debug_items()` on GCC and
Clang, improving -Os compile time of Minerva SoC SRAM by ~40% (!)
This commit is contained in:
whitequark 2020-12-13 18:16:55 +00:00
parent 4d40595d64
commit f75bc6c7aa
2 changed files with 15 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

17
backends/cxxrtl/cxxrtl.h
View file

@ -41,18 +41,29 @@
#include <backends/cxxrtl/cxxrtl_capi.h> #include <backends/cxxrtl/cxxrtl_capi.h>
#ifndef __has_attribute
# define __has_attribute(x) 0
#endif
// CXXRTL essentially uses the C++ compiler as a hygienic macro engine that feeds an instruction selector. // CXXRTL essentially uses the C++ compiler as a hygienic macro engine that feeds an instruction selector.
// It generates a lot of specialized template functions with relatively large bodies that, when inlined // It generates a lot of specialized template functions with relatively large bodies that, when inlined
// into the caller and (for those with loops) unrolled, often expose many new optimization opportunities. // into the caller and (for those with loops) unrolled, often expose many new optimization opportunities.
// Because of this, most of the CXXRTL runtime must be always inlined for best performance. // Because of this, most of the CXXRTL runtime must be always inlined for best performance.
#ifndef __has_attribute
# define __has_attribute(x) 0
#endif
#if __has_attribute(always_inline) #if __has_attribute(always_inline)
#define CXXRTL_ALWAYS_INLINE inline __attribute__((__always_inline__)) #define CXXRTL_ALWAYS_INLINE inline __attribute__((__always_inline__))
#else #else
#define CXXRTL_ALWAYS_INLINE inline #define CXXRTL_ALWAYS_INLINE inline
#endif #endif
// Conversely, some functions in the generated code are extremely large yet very cold, with both of these
// properties being extreme enough to confuse C++ compilers into spending pathological amounts of time
// on a futile (the code becomes worse) attempt to optimize the least important parts of code.
#if __has_attribute(optnone)
#define CXXRTL_EXTREMELY_COLD __attribute__((__optnone__))
#elif __has_attribute(optimize)
#define CXXRTL_EXTREMELY_COLD __attribute__((__optimize__(0)))
#else
#define CXXRTL_EXTREMELY_COLD
#endif
// CXXRTL uses assert() to check for C++ contract violations (which may result in e.g. undefined behavior // CXXRTL uses assert() to check for C++ contract violations (which may result in e.g. undefined behavior
// of the simulation code itself), and CXXRTL_ASSERT to check for RTL contract violations (which may at // of the simulation code itself), and CXXRTL_ASSERT to check for RTL contract violations (which may at

1
backends/cxxrtl/cxxrtl_backend.cc
View file

@ -2038,6 +2038,7 @@ struct CxxrtlWorker {
f << indent << "}\n"; f << indent << "}\n";
f << "\n"; f << "\n";
} }
f << indent << "CXXRTL_EXTREMELY_COLD\n";
f << indent << "void " << mangle(module) << "::debug_info(debug_items &items, std::string path) {\n"; f << indent << "void " << mangle(module) << "::debug_info(debug_items &items, std::string path) {\n";
dump_debug_info_method(module); dump_debug_info_method(module);
f << indent << "}\n"; f << indent << "}\n";