From 4a7a15757d40b7dd5a5ecedfa5927ec7ed799d67 Mon Sep 17 00:00:00 2001 From: Maciej Fijalkowski Date: Mon, 1 Mar 2021 15:17:34 +0000 Subject: hopefully add vmprof support to pypy on aarch64 --- rpython/jit/backend/aarch64/assembler.py | 32 ++++++++++++++-------- rpython/jit/backend/aarch64/runner.py | 3 ++ rpython/jit/backend/aarch64/test/test_rvmprof.py | 6 ++++ .../jit/backend/aarch64/test/test_zrpy_vmprof.py | 7 +++++ 4 files changed, 36 insertions(+), 12 deletions(-) create mode 100644 rpython/jit/backend/aarch64/test/test_rvmprof.py create mode 100644 rpython/jit/backend/aarch64/test/test_zrpy_vmprof.py diff --git a/rpython/jit/backend/aarch64/assembler.py b/rpython/jit/backend/aarch64/assembler.py index 0212d6dd7d..050710f1fa 100644 --- a/rpython/jit/backend/aarch64/assembler.py +++ b/rpython/jit/backend/aarch64/assembler.py @@ -44,6 +44,9 @@ class AssemblerARM64(ResOpAssembler): assert len(set(inputargs)) == len(inputargs) self.setup(looptoken) + if self.cpu.HAS_CODEMAP: + self.codemap_builder.enter_portal_frame(jd_id, unique_id, + self.mc.get_relative_pos()) frame_info = self.datablockwrapper.malloc_aligned( jitframe.JITFRAMEINFO_SIZE, alignment=WORD) @@ -124,7 +127,10 @@ class AssemblerARM64(ResOpAssembler): assert len(set(inputargs)) == len(inputargs) self.setup(original_loop_token) - #self.codemap.inherit_code_from_position(faildescr.adr_jump_offset) + if self.cpu.HAS_CODEMAP: + self.codemap_builder.inherit_code_from_position( + faildescr.adr_jump_offset) + descr_number = compute_unique_id(faildescr) if log: operations = self._inject_debugging_code(faildescr, operations, @@ -1019,8 +1025,8 @@ class AssemblerARM64(ResOpAssembler): size = self.mc.get_relative_pos() res = self.mc.materialize(self.cpu, allblocks, self.cpu.gc_ll_descr.gcrootmap) - #self.cpu.codemap.register_codemap( - # self.codemap.get_final_bytecode(res, size)) + self.cpu.codemap.register_codemap( + self.codemap_builder.get_final_bytecode(res, size)) return res def patch_trace(self, faildescr, looptoken, bridge_addr, regalloc): @@ -1086,13 +1092,13 @@ class AssemblerARM64(ResOpAssembler): pmc.B_ofs_cond(self.mc.currpos() - pos, c.LS) def _call_header(self): - stack_size = (len(r.callee_saved_registers) + 4) * WORD + stack_size = (len(r.callee_saved_registers) + 8) * WORD self.mc.STP_rr_preindex(r.lr.value, r.fp.value, r.sp.value, -stack_size) for i in range(0, len(r.callee_saved_registers), 2): self.mc.STP_rri(r.callee_saved_registers[i].value, r.callee_saved_registers[i + 1].value, r.sp.value, - (i + 4) * WORD) + (i + 8) * WORD) if self.cpu.translate_support_code: self._call_header_vmprof() @@ -1118,14 +1124,16 @@ class AssemblerARM64(ResOpAssembler): offset = cintf.vmprof_tl_stack.getoffset() self.mc.LDR_ri(r.ip0.value, tloc.value, offset) # stack->next = old - self.mc.STR_ri(r.ip0.value, r.sp.value, 0) + self.mc.STR_ri(r.ip0.value, r.sp.value, 4 * WORD) # stack->value = my sp - self.mc.STR_ri(r.sp.value, r.sp.value, WORD) + self.mc.ADD_ri(r.ip1.value, r.sp.value, 0) + self.mc.STR_ri(r.ip1.value, r.sp.value, (4 + 1) * WORD) # stack->kind = VMPROF_JITTED_TAG self.mc.gen_load_int(r.ip0.value, VMPROF_JITTED_TAG) - self.mc.STR_ri(r.ip0.value, r.sp.value, WORD * 2) + self.mc.STR_ri(r.ip0.value, r.sp.value, (4 + 2) * WORD) # save in vmprof_tl_stack the new eax - self.mc.STR_ri(r.sp.value, tloc.value, offset) + self.mc.ADD_ri(r.ip0.value, r.sp.value, 4 * WORD) + self.mc.STR_ri(r.ip0.value, tloc.value, offset) def _assemble(self, regalloc, inputargs, operations): @@ -1367,13 +1375,13 @@ class AssemblerARM64(ResOpAssembler): self._call_footer_vmprof(mc) # pop all callee saved registers - stack_size = (len(r.callee_saved_registers) + 4) * WORD + stack_size = (len(r.callee_saved_registers) + 8) * WORD for i in range(0, len(r.callee_saved_registers), 2): mc.LDP_rri(r.callee_saved_registers[i].value, r.callee_saved_registers[i + 1].value, r.sp.value, - (i + 4) * WORD) + (i + 8) * WORD) mc.LDP_rr_postindex(r.lr.value, r.fp.value, r.sp.value, stack_size) @@ -1384,7 +1392,7 @@ class AssemblerARM64(ResOpAssembler): # ip0 = address of pypy_threadlocal_s mc.LDR_ri(r.ip0.value, r.sp.value, 3 * WORD) # ip1 = (our local vmprof_tl_stack).next - mc.LDR_ri(r.ip1.value, r.sp.value, 0) + mc.LDR_ri(r.ip1.value, r.sp.value, 4 * WORD) # save in vmprof_tl_stack the value eax offset = cintf.vmprof_tl_stack.getoffset() mc.STR_ri(r.ip1.value, r.ip0.value, offset) diff --git a/rpython/jit/backend/aarch64/runner.py b/rpython/jit/backend/aarch64/runner.py index ef69fff25d..063b85b483 100644 --- a/rpython/jit/backend/aarch64/runner.py +++ b/rpython/jit/backend/aarch64/runner.py @@ -14,6 +14,7 @@ class CPU_ARM64(AbstractLLCPU): gen_regs = r.all_regs float_regs = VFPRegisterManager.all_regs supports_floats = True + HAS_CODEMAP = True from rpython.jit.backend.aarch64.arch import JITFRAME_FIXED_SIZE @@ -29,6 +30,8 @@ class CPU_ARM64(AbstractLLCPU): def setup_once(self): self.assembler.setup_once() + if self.HAS_CODEMAP: + self.codemap.setup() def compile_bridge(self, faildescr, inputargs, operations, original_loop_token, log=True, logger=None): diff --git a/rpython/jit/backend/aarch64/test/test_rvmprof.py b/rpython/jit/backend/aarch64/test/test_rvmprof.py new file mode 100644 index 0000000000..e2e50e77aa --- /dev/null +++ b/rpython/jit/backend/aarch64/test/test_rvmprof.py @@ -0,0 +1,6 @@ +from rpython.jit.backend.test.test_rvmprof import BaseRVMProfTest +from rpython.jit.backend.aarch64.test.test_basic import JitAarch64Mixin + + +class TestRVMProfCall(JitAarch64Mixin, BaseRVMProfTest): + pass diff --git a/rpython/jit/backend/aarch64/test/test_zrpy_vmprof.py b/rpython/jit/backend/aarch64/test/test_zrpy_vmprof.py new file mode 100644 index 0000000000..68b5d6ba03 --- /dev/null +++ b/rpython/jit/backend/aarch64/test/test_zrpy_vmprof.py @@ -0,0 +1,7 @@ + +from rpython.jit.backend.llsupport.test.zrpy_vmprof_test import CompiledVmprofTest + +class TestZVMprof(CompiledVmprofTest): + + gcrootfinder = "shadowstack" + gc = "incminimark" \ No newline at end of file -- cgit v1.2.3-65-gdbad