I did not replace riscv.clmulh/clmulr since those require a multiple
instruction pattern match. I wanted to ensure that -O0 will select the
correct instructions without relying on combines.
Extract ManualCodegen blocks from riscv_vector.td to dedicated helper
functions in RISCV.cpp to improve compilation times and code
organization.
This refactoring:
- Reduces riscv_vector_builtin_cg.inc from ~70,000 lines to ~30,000
lines
- Extracts lots of ManualCodegen blocks into helper functions in
RISCV.cpp
- Moves complex code generation logic from TableGen to C++
- Marks extracted functions with LLVM_ATTRIBUTE_NOINLINE to prevent
excessive inlining in EmitRISCVBuiltinExpr's large switch statement,
which would cause compilation time to increase significantly
Performance Impact on AMD Ryzen 9 3950X 16-Core with SSD (Release build)
with GCC 11:
Before: real 1m4.560s, user 0m0.529s, sys 0m0.175s
After: real 0m22.577s, user 0m0.498s, sys 0m0.152s
Which reduced around 65% of compilation time.
During this refactoring, I also found few more opportunities to optimize
and simplify the code generation logic, but I think leave to next PR
since it already change a lot of code.
Fix#88368
XAndesBFHCvt provides two builtins functions for converting between
float and bf16. Users can use them to convert bf16 values loaded from
memory to float, perform arithmetic operations, then convert them back
to bf16 and store them to memory.
The load/store and move operations for bf16 will be handled in a later
patch.
clang/lib/CodeGen/CGBuiltin.cpp is over 1MB long (>23k LoC), and can
take minutes to recompile (depending on compiler and host system) when
modified, and 5 seconds for clangd to update for every edit. Splitting
this file was discussed in this thread:
https://discourse.llvm.org/t/splitting-clang-s-cgbuiltin-cpp-over-23k-lines-long-takes-1min-to-compile/
and the idea has received a number of +1 votes, hence this change.
