Commit Graph

3200 Commits

Author SHA1 Message Date
whitequark c6f946a816 llvm_ir_generator: add fast-math flags to fcmp.
This is allowed in 3.8.
2016-04-14 18:10:30 +08:00
whitequark d4f1614a23 llvm_ir_generator: change !{→unconditionally_}dereferenceable.
Since LLVM 3.8, !dereferenceable is weaker, so we introduce
!unconditionally_dereferenceable (http://reviews.llvm.org/D18738)
to regain its functionality.
2016-04-14 18:10:17 +08:00
whitequark 75252ca5a4 llvm_ir_generator: fix DICompileUnit.language. 2016-04-14 18:10:17 +08:00
whitequark 31b5154222 conda: update llvmlite-artiq dependency.
Build 22 includes debug information support.
2016-04-14 18:10:17 +08:00
whitequark 89326fb189 compiler: purge generated functions from backtraces. 2016-04-14 18:09:59 +08:00
whitequark a2f6e81c50 ttl: mark constant attributes for TTL{In,InOut,ClockGen}. 2016-04-14 18:09:59 +08:00
whitequark 702e959033 llvm_ir_generator: add TBAA metadata for @now. 2016-04-14 18:09:59 +08:00
whitequark f958cba4ed llvm_ir_generator: update debug info emission for LLVM 3.8. 2016-04-14 18:09:36 +08:00
whitequark 7c520aa0c4 coredevice: format backtrace RA as +0xN, not 0xN.
The absolute address is somewhere in the 0x4000000 range; the one
that is displayed is an offset from the shared object base.
2016-04-14 18:09:36 +08:00
whitequark 66bbee51d8 conda: require llvmlite-artiq built for LLVM 3.8. 2016-04-14 18:09:09 +08:00
whitequark f26990aa57 compiler: emit verbose assembly via ARTIQ_DUMP_ASM. 2016-04-14 18:09:02 +08:00
whitequark c89c27e389 compiler: add analysis passes from TargetMachine.
This doesn't have any effect right now, but is the right thing to do.
2016-04-14 18:08:47 +08:00
whitequark 1120c264b1 compiler: mark loaded pointers as !dereferenceable.
Also, lower the bound for test_pulse_rate_dds, since we generate
better code for it now.
2016-04-14 18:08:47 +08:00
whitequark 03b6555d9d compiler: update for LLVM 3.7. 2016-04-14 18:08:28 +08:00
whitequark 932e680f3e compiler: use correct data layout. 2016-04-14 18:07:56 +08:00
whitequark f59fd8faec llvm_ir_generator: do not use 'coldcc' calling convention.
First, this calling convention doesn't actually exist in OR1K
and trying to use it in Asserts build causes an UNREACHABLE.

Second, I tried to introduce it and it does not appear to produce
any measurable benefit: not only OR1K has a ton of CSRs but also
it is quite hard, if not realistically impossible, to produce
the kind of register pressure that would be relieved by sparing
a few more CSRs for our exception raising function calls, since
temporaries don't have to be preserved before a noreturn call
and spilling over ten registers across an exceptional edge
is not something that the code we care about would do.

Third, it produces measurable drawbacks: it inflates code size
of check:* functions by adding spills. Of course, this could be
alleviated by making __artiq_raise coldcc as well, but what's
the point anyway?
2016-04-14 18:07:35 +08:00
whitequark e416246e78 llvm_ir_generator: mark loads as non-null where applicable. 2016-04-14 18:07:35 +08:00
whitequark 50ae17649d test: relax lit/embedding/syscall_flags.
We currently have broken debug info. In either case, debug info
is irrelevant to this test.
2016-04-14 18:07:35 +08:00
whitequark f7603dcb6f compiler: fix ARTIQ_DUMP_ELF. 2016-04-14 18:07:17 +08:00
whitequark 812e79b63d llvm_ir_generator: don't mark non-constant attribute loads as invariant.
Oops.
2016-04-14 18:07:13 +08:00
whitequark dcb0ffdd03 Commit missing parts of 1d8b0d46. 2016-04-14 18:07:04 +08:00
whitequark ee7e648cb0 compiler: allow specifying per-function "fast-math" flags.
Fixes #351.
2016-04-14 18:07:04 +08:00
whitequark 5fafcc1341 Commit missing parts of 6f5332f8. 2016-04-14 18:07:04 +08:00
whitequark f7d4a37df9 compiler: allow flagging syscalls, providing information to optimizer.
This also fixes a crash in test_cache introduced in 1d8b0d46.
2016-04-14 18:06:47 +08:00
whitequark c6b21652ba compiler: mark FFI functions as ModRef=Ref using TBAA metadata.
Fascinatingly, the fact that you can mark call instructions with
!tbaa metadata is completely undocumented. Regardless, it is true:
a !tbaa metadata for an "immutable" type will cause
AliasAnalysis::getModRefBehavior to return OnlyReadsMemory for that
call site.

Don't bother marking loads with TBAA yet since we already place
!load.invariant on them (which is as good as the TBAA "immutable"
flag) and after that we're limited by lack of !nonnull anyway.

Also, add TBAA analysis passes in our pipeline to actually engage it.
2016-04-14 18:06:47 +08:00
whitequark 0e0f81b509 compiler: mark loads of kernel constant attributes as load invariant.
Also, enable LICM, since it can take advantage of this.
2016-04-14 18:06:47 +08:00
whitequark 081edb27d7 coredevice: add some kernel_constant_attributes specifications. 2016-04-14 18:06:47 +08:00
whitequark b5fd257a33 compiler: do not write back kernel constant attributes.
Fixes #322.
2016-04-14 18:06:21 +08:00
whitequark 665e59e064 compiler: implement kernel constant attributes.
Part of #322.
2016-04-14 18:06:21 +08:00
whitequark 348e058c6f test_pulse_rate_dds: tighten upper bound to 400us. 2016-04-14 18:06:06 +08:00
whitequark 718d411dd5 compiler: run IPSCCP.
This doesn't do much, only frees some registers.
2016-04-14 18:05:57 +08:00
whitequark 019f528ea6 compiler: raise inliner threshold to the equivalent of -O3. 2016-04-14 18:05:57 +08:00
whitequark 3fa5762c10 compiler: extract runtime checks into separate cold functions.
This reduces register pressure as well as function size, which
favorably affects the inliner.
2016-04-14 18:05:57 +08:00
whitequark fcf2a73f82 test_pulse_rate: tighten upper bound to 1500ns. 2016-04-14 18:05:31 +08:00
whitequark 92f3dc705f llvm_ir_generator: generate code more amenable to LLVM's GlobalOpt.
This exposes almost all embedded methods to inlining, with massive
gains.
2016-04-14 18:05:10 +08:00
whitequark f2c92fffea compiler: make quoted functions independent of outer environment. 2016-04-14 18:04:42 +08:00
whitequark ccb1d54beb compiler: tune the LLVM optimizer pipeline (fixes #315). 2016-04-14 18:04:42 +08:00
whitequark 8fa4281470 compiler: significantly increase readability of LLVM and ARTIQ IRs. 2016-04-14 18:04:42 +08:00
whitequark e534941383 compiler: quote functions directly instead of going through a local. 2016-04-14 18:04:22 +08:00
whitequark f72e050af5 transforms.llvm_ir_generator: extract class function attributes.
This should give LLVM more visibility.
2016-04-14 18:04:22 +08:00
whitequark 00facbbc78 compiler: get rid of the GetConstructor opcode. 2016-04-14 18:04:22 +08:00
Sebastien Bourdeauducq 321ba57e84 manual/installing: --toolchain vivado 2016-04-14 01:25:48 +08:00
Sebastien Bourdeauducq 582efe5b91 typo 2016-04-14 01:17:47 +08:00
Sebastien Bourdeauducq 349ccfb633 gateware/nist_qc2: substitute FMC 2016-04-14 01:04:19 +08:00
Sebastien Bourdeauducq 71b9ba6ab7 manual/faq: list HITL TTL connections 2016-04-14 01:04:19 +08:00
Sebastien Bourdeauducq 317e6ea38d manual/git: commit needed before starting master, refresh. Closes #387 2016-04-14 01:04:19 +08:00
dhslichter 08e742ce68 Updated qc2 pinouts for SPI and 2x DDS bus, update docs 2016-04-13 18:39:50 +08:00
Robert Jördens 90876e0143 examples: move pdq2 frame selects away from TTLInOut ttl3 2016-04-12 19:41:16 +08:00
Robert Jördens 6552aa4c28 test: set inputs to input(), should close #383 2016-04-12 18:17:39 +08:00
Sebastien Bourdeauducq 936190033e gui/models: handle Qt calling DictSyncTreeSepModel.index with garbage inputs. Closes #388 2016-04-11 20:12:07 +08:00