[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
https://github.com/mattweingarten updated https://github.com/llvm/llvm-project/pull/94264 error: too big or took too long to generate ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -140,12 +140,16 @@ static cl::opt ClDebugMin("memprof-debug-min",
cl::desc("Debug min inst"),
static cl::opt ClDebugMax("memprof-debug-max", cl::desc("Debug max inst"),
cl::Hidden, cl::init(-1));
+static cl::opt ClHistogram("memprof-histogram",
mattweingarten wrote:
Use global variable `__memprof_histogram` in IR to set flag in runtime, and got
rid of runtime flag.
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
https://github.com/mattweingarten updated https://github.com/llvm/llvm-project/pull/94264 error: too big or took too long to generate ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -216,6 +228,35 @@ u64 GetShadowCount(uptr p, u32 size) {
return count;
}
+// Accumulates the access count from the shadow for the given pointer and size.
+// See memprof_mapping.h for an overview on histogram counters.
+u64 GetShadowCountHistogram(uptr p, u32 size) {
+ u8 *shadow = (u8 *)HISTOGRAM_MEM_TO_SHADOW(p);
+ u8 *shadow_end = (u8 *)HISTOGRAM_MEM_TO_SHADOW(p + size);
+ u64 count = 0;
+ for (; shadow <= shadow_end; shadow++)
+count += *shadow;
+ return count;
+}
+
+// If we use the normal approach from clearCountersWithoutHistogram, the
+// histogram will clear too much data and may overwrite shadow counters that
are
+// in use. Likely because of rounding up the shadow_end pointer.
+// See memprof_mapping.h for an overview on histogram counters.
+void clearCountersHistogram(uptr addr, uptr size) {
+ u8 *shadow_8 = (u8 *)HISTOGRAM_MEM_TO_SHADOW(addr);
+ u8 *shadow_end_8 = (u8 *)HISTOGRAM_MEM_TO_SHADOW(addr + size);
+ for (; shadow_8 < shadow_end_8; shadow_8++) {
mattweingarten wrote:
changed.
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -216,6 +228,35 @@ u64 GetShadowCount(uptr p, u32 size) {
return count;
}
+// Accumulates the access count from the shadow for the given pointer and size.
+// See memprof_mapping.h for an overview on histogram counters.
+u64 GetShadowCountHistogram(uptr p, u32 size) {
+ u8 *shadow = (u8 *)HISTOGRAM_MEM_TO_SHADOW(p);
+ u8 *shadow_end = (u8 *)HISTOGRAM_MEM_TO_SHADOW(p + size);
+ u64 count = 0;
+ for (; shadow <= shadow_end; shadow++)
+count += *shadow;
+ return count;
+}
+
+// If we use the normal approach from clearCountersWithoutHistogram, the
+// histogram will clear too much data and may overwrite shadow counters that
are
+// in use. Likely because of rounding up the shadow_end pointer.
+// See memprof_mapping.h for an overview on histogram counters.
+void clearCountersHistogram(uptr addr, uptr size) {
+ u8 *shadow_8 = (u8 *)HISTOGRAM_MEM_TO_SHADOW(addr);
+ u8 *shadow_end_8 = (u8 *)HISTOGRAM_MEM_TO_SHADOW(addr + size);
+ for (; shadow_8 < shadow_end_8; shadow_8++) {
+*shadow_8 = 0;
+ }
+}
+
+void clearCountersWithoutHistogram(uptr addr, uptr size) {
+ uptr shadow_beg = MEM_TO_SHADOW(addr);
+ uptr shadow_end = MEM_TO_SHADOW(addr + size - SHADOW_GRANULARITY) + 1;
+ REAL(memset)((void *)shadow_beg, 0, shadow_end - shadow_beg);
+}
+
// Clears the shadow counters (when memory is allocated).
void ClearShadow(uptr addr, uptr size) {
mattweingarten wrote:
This function uses the MEM_TO_SHADOW computed pointers "kind of", but in
reality it just rounds these pointers up to nearest page_sizes. So in effect,
full pages should be cleared no matter if it is with histogram or without.
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -38,4 +38,5 @@ MEMPROF_FLAG(bool, allocator_frees_and_returns_null_on_realloc_zero, true, MEMPROF_FLAG(bool, print_text, false, "If set, prints the heap profile in text format. Else use the raw binary serialization format.") MEMPROF_FLAG(bool, print_terse, false, - "If set, prints memory profile in a terse format. Only applicable if print_text = true.") + "If set, prints memory profile in a terse format. Only applicable " mattweingarten wrote: done. https://github.com/llvm/llvm-project/pull/94264 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -205,8 +205,14 @@ class RawMemProfReader final : public MemProfReader {
object::SectionedAddress getModuleOffset(uint64_t VirtualAddress);
+ llvm::SmallVector>
+ readMemInfoBlocks(const char *Ptr);
+
// The profiled binary.
object::OwningBinary Binary;
+ // Version of raw memprof binary currently being read. Defaults to most
update
mattweingarten wrote:
done.
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -96,19 +102,63 @@ llvm::SmallVector readSegmentEntries(const
char *Ptr) {
}
llvm::SmallVector>
-readMemInfoBlocks(const char *Ptr) {
+readMemInfoBlocksV3(const char *Ptr) {
using namespace support;
const uint64_t NumItemsToRead =
- endian::readNext(Ptr);
+ endian::readNext(Ptr);
+
llvm::SmallVector> Items;
for (uint64_t I = 0; I < NumItemsToRead; I++) {
const uint64_t Id =
-endian::readNext(Ptr);
-const MemInfoBlock MIB = *reinterpret_cast(Ptr);
+endian::readNext(Ptr);
+
+// We cheat a bit here and remove the const from cast to set the
+// Histogram Pointer to newly allocated buffer. We also cheat, since V3 and
+// V4 do not have the same fields. V3 is missing AccessHistogramSize and
+// AccessHistogram. This means we read "dirty" data in here, but it should
+// not segfault, since there will be callstack data placed after this in
the
+// binary format.
+MemInfoBlock MIB = *reinterpret_cast(Ptr);
+// Overwrite dirty data.
mattweingarten wrote:
My thought process is that the MemInfoBlock MIB is a allocated on the stack,
since its a local variable and not a pointer. Meaning the reinterpret_cast
copies the information from the data-buffer into the stack variable. Correct
me if I am wrong
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -610,13 +670,33 @@ RawMemProfReader::peekBuildIds(MemoryBuffer *DataBuffer) {
return BuildIds.takeVector();
}
+// FIXME: Add a schema for serializing similiar to IndexedMemprofReader. This
+// will help being able to deserialize different versions raw memprof versions
+// more easily.
+llvm::SmallVector>
+RawMemProfReader::readMemInfoBlocks(const char *Ptr) {
+ if (MemprofRawVersion == 3ULL) {
+errs() << "Reading V3\n";
mattweingarten wrote:
done.
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -508,7 +519,26 @@ void createProfileFileNameVar(Module &M) {
}
}
+// Set MemprofHistogramFlag as a Global veriable in IR. This makes it
accessible
+// to
mattweingarten wrote:
done.
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -508,7 +519,26 @@ void createProfileFileNameVar(Module &M) {
}
}
+// Set MemprofHistogramFlag as a Global veriable in IR. This makes it
accessible
+// to
+// the runtime, changing shadow count behavior.
+void createMemprofHistogramFlagVar(Module &M) {
+ const StringRef VarName(MemProfHistogramFlagVar);
+ Type *IntTy1 = Type::getInt1Ty(M.getContext());
+ auto MemprofHistogramFlag = new GlobalVariable(
+ M, IntTy1, true, GlobalValue::WeakAnyLinkage,
+ Constant::getIntegerValue(IntTy1, APInt(1, ClHistogram)), VarName);
+ // MemprofHistogramFlag->setVisibility(GlobalValue::HiddenVisibility);
mattweingarten wrote:
done.
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -0,0 +1,101 @@ +REQUIRES: x86_64-linux + +This is a copy of memprof-basict.test with slight changes to check that we can still read v3 of memprofraw. mattweingarten wrote: done https://github.com/llvm/llvm-project/pull/94264 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -0,0 +1,101 @@ +REQUIRES: x86_64-linux + +This is a copy of memprof-basict.test with slight changes to check that we can still read v3 of memprofraw. + +To update the inputs used below run Inputs/update_memprof_inputs.sh /path/to/updated/clang mattweingarten wrote: yes, good point! done. https://github.com/llvm/llvm-project/pull/94264 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -20,25 +20,25 @@ CHECK-NEXT: - CHECK: Records: CHECK-NEXT: - -CHECK-NEXT:FunctionGUID: 15505678318020221912 +CHECK-NEXT:FunctionGUID: 3873612792189045660 CHECK-NEXT:AllocSites: CHECK-NEXT:- CHECK-NEXT: Callstack: CHECK-NEXT: - -CHECK-NEXT:Function: 15505678318020221912 -CHECK-NEXT:SymbolName: qux +CHECK-NEXT:Function: 3873612792189045660 +CHECK-NEXT:SymbolName: _Z3quxi mattweingarten wrote: good question, I suspect this test has not been regenerated in a long time and something changed in meantime in memprof? I think it might have to do with the SymbolNames being mangled vs demangled, which changes the hash value? Is it a problem that the hash value changed? https://github.com/llvm/llvm-project/pull/94264 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -124,6 +124,13 @@ struct PortableMemInfoBlock {
OS << "" << #Name << ": " << Name << "\n";
#include "llvm/ProfileData/MIBEntryDef.inc"
#undef MIBEntryDef
+if (AccessHistogramSize > 0) {
+ OS << "" << "AccessHistogramValues" << ":";
+ for (uint32_t I = 0; I < AccessHistogramSize; ++I) {
+OS << " -" << ((uint64_t *)AccessHistogram)[I];
mattweingarten wrote:
I though only having space delimiting did not produce valid yaml, I double
checked and was wrong. Changed to space delimiting, thanks!
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
https://github.com/mattweingarten updated https://github.com/llvm/llvm-project/pull/94264 error: too big or took too long to generate ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
https://github.com/mattweingarten edited https://github.com/llvm/llvm-project/pull/94264 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
@@ -216,6 +228,35 @@ u64 GetShadowCount(uptr p, u32 size) {
return count;
}
+// Accumulates the access count from the shadow for the given pointer and size.
+// See memprof_mapping.h for an overview on histogram counters.
+u64 GetShadowCountHistogram(uptr p, u32 size) {
+ u8 *shadow = (u8 *)HISTOGRAM_MEM_TO_SHADOW(p);
+ u8 *shadow_end = (u8 *)HISTOGRAM_MEM_TO_SHADOW(p + size);
+ u64 count = 0;
+ for (; shadow <= shadow_end; shadow++)
+count += *shadow;
+ return count;
+}
+
+// If we use the normal approach from clearCountersWithoutHistogram, the
+// histogram will clear too much data and may overwrite shadow counters that
are
+// in use. Likely because of rounding up the shadow_end pointer.
+// See memprof_mapping.h for an overview on histogram counters.
+void clearCountersHistogram(uptr addr, uptr size) {
+ u8 *shadow_8 = (u8 *)HISTOGRAM_MEM_TO_SHADOW(addr);
+ u8 *shadow_end_8 = (u8 *)HISTOGRAM_MEM_TO_SHADOW(addr + size);
+ for (; shadow_8 < shadow_end_8; shadow_8++) {
+*shadow_8 = 0;
+ }
+}
+
+void clearCountersWithoutHistogram(uptr addr, uptr size) {
+ uptr shadow_beg = MEM_TO_SHADOW(addr);
+ uptr shadow_end = MEM_TO_SHADOW(addr + size - SHADOW_GRANULARITY) + 1;
+ REAL(memset)((void *)shadow_beg, 0, shadow_end - shadow_beg);
+}
+
// Clears the shadow counters (when memory is allocated).
void ClearShadow(uptr addr, uptr size) {
mattweingarten wrote:
Yes I agree, this is nicer.
Changed.
https://github.com/llvm/llvm-project/pull/94264
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[clang-tools-extra] [compiler-rt] [lldb] [llvm] [Memprof] Adds the option to collect AccessCountHistograms for memprof. (PR #94264)
https://github.com/mattweingarten updated https://github.com/llvm/llvm-project/pull/94264 error: too big or took too long to generate ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
