hist: Support values creater than 4G

Before we only supported values up to 4G. Unclear why that was. We have
a whole u64 to store bucket position. Supporting up to INT64_MAX still
only takes a u8. So might as well support up to INT64_MAX.

This fixes #856.

src/ast/passes/codegen_llvm.cpp

@@ -3066,7 +3066,7 @@ Function *CodegenLLVM::createLog2Function()
   //   result++;
   //   if (n == 0) return result;
   //   result++;
-  //   for (int i = 4; i >= 0; i--)
+  //   for (int i = 5; i >= 0; i--)
   //   {
   //     shift = (n >= (1<<(1<<i))) << i;
   //     n >>= shift;
@@ -3114,10 +3114,16 @@ Function *CodegenLLVM::createLog2Function()
 
   // power-of-2 index, offset by +2
   b_.CreateStore(b_.getInt64(2), result);
-  for (int i = 4; i >= 0; i--)
+  for (int i = 5; i >= 0; i--)
   {
     Value *n = b_.CreateLoad(b_.getInt64Ty(), n_alloc);
-    Value *shift = b_.CreateShl(b_.CreateIntCast(b_.CreateICmpSGE(b_.CreateIntCast(n, b_.getInt64Ty(), false), b_.getInt64(1 << (1<<i))), b_.getInt64Ty(), false), i);
+    Value *shift = b_.CreateShl(
+        b_.CreateIntCast(
+            b_.CreateICmpSGE(b_.CreateIntCast(n, b_.getInt64Ty(), false),
+                             b_.getInt64(1ULL << (1 << i))),
+            b_.getInt64Ty(),
+            false),
+        i);
     b_.CreateStore(b_.CreateLShr(n, shift), n_alloc);
     b_.CreateStore(b_.CreateAdd(b_.CreateLoad(b_.getInt64Ty(), result), shift),
                    result);

src/output.cpp

@@ -712,8 +712,8 @@ std::string JsonOutput::hist_to_str(const std::vector<uint64_t> &values,
     }
     else
     {
-      long low = 1 << (i-2);
-      long high = (1 << (i-2+1)) - 1;
+      long low = 1ULL << (i - 2);
+      long high = (1ULL << (i - 2 + 1)) - 1;
       res << "\"min\": " << low << ", \"max\": " << high << ", ";
     }
     res << "\"count\": " << values.at(i) / div;

tests/codegen/llvm/call_hist.ll

@@ -81,52 +81,61 @@ hist.is_zero:                                     ; preds = %hist.is_not_less_th
 hist.is_not_zero:                                 ; preds = %hist.is_not_less_than_zero
   store i64 2, i64* %1, align 8
   %11 = load i64, i64* %2, align 8
-  %12 = icmp sge i64 %11, 65536
+  %12 = icmp sge i64 %11, 4294967296
   %13 = zext i1 %12 to i64
-  %14 = shl i64 %13, 4
+  %14 = shl i64 %13, 5
   %15 = lshr i64 %11, %14
   store i64 %15, i64* %2, align 8
   %16 = load i64, i64* %1, align 8
   %17 = add i64 %16, %14
   store i64 %17, i64* %1, align 8
   %18 = load i64, i64* %2, align 8
-  %19 = icmp sge i64 %18, 256
+  %19 = icmp sge i64 %18, 65536
   %20 = zext i1 %19 to i64
-  %21 = shl i64 %20, 3
+  %21 = shl i64 %20, 4
   %22 = lshr i64 %18, %21
   store i64 %22, i64* %2, align 8
   %23 = load i64, i64* %1, align 8
   %24 = add i64 %23, %21
   store i64 %24, i64* %1, align 8
   %25 = load i64, i64* %2, align 8
-  %26 = icmp sge i64 %25, 16
+  %26 = icmp sge i64 %25, 256
   %27 = zext i1 %26 to i64
-  %28 = shl i64 %27, 2
+  %28 = shl i64 %27, 3
   %29 = lshr i64 %25, %28
   store i64 %29, i64* %2, align 8
   %30 = load i64, i64* %1, align 8
   %31 = add i64 %30, %28
   store i64 %31, i64* %1, align 8
   %32 = load i64, i64* %2, align 8
-  %33 = icmp sge i64 %32, 4
+  %33 = icmp sge i64 %32, 16
   %34 = zext i1 %33 to i64
-  %35 = shl i64 %34, 1
+  %35 = shl i64 %34, 2
   %36 = lshr i64 %32, %35
   store i64 %36, i64* %2, align 8
   %37 = load i64, i64* %1, align 8
   %38 = add i64 %37, %35
   store i64 %38, i64* %1, align 8
   %39 = load i64, i64* %2, align 8
-  %40 = icmp sge i64 %39, 2
+  %40 = icmp sge i64 %39, 4
   %41 = zext i1 %40 to i64
-  %42 = shl i64 %41, 0
+  %42 = shl i64 %41, 1
   %43 = lshr i64 %39, %42
   store i64 %43, i64* %2, align 8
   %44 = load i64, i64* %1, align 8
   %45 = add i64 %44, %42
   store i64 %45, i64* %1, align 8
-  %46 = load i64, i64* %1, align 8
-  ret i64 %46
+  %46 = load i64, i64* %2, align 8
+  %47 = icmp sge i64 %46, 2
+  %48 = zext i1 %47 to i64
+  %49 = shl i64 %48, 0
+  %50 = lshr i64 %46, %49
+  store i64 %50, i64* %2, align 8
+  %51 = load i64, i64* %1, align 8
+  %52 = add i64 %51, %49
+  store i64 %52, i64* %1, align 8
+  %53 = load i64, i64* %1, align 8
+  ret i64 %53
 }
 
 ; Function Attrs: argmemonly nofree nosync nounwind willreturn

tests/runtime/call

@@ -199,6 +199,11 @@ EXPECT @bytes: *\n[\[(].*
 AFTER ./testprogs/syscall read
 TIMEOUT 5
 
+NAME hist_10g
+RUN {{BPFTRACE}} -q -f json -e 'BEGIN { @ = hist(10 * 1024 * 1024 * 1024); exit(); }' | python3 -c 'import sys,json; print(json.load(sys.stdin) == json.load(open("runtime/outputs/hist_10g.json")))'
+EXPECT ^True$
+TIMEOUT 1
+
 NAME lhist
 PROG kretprobe:vfs_read { @bytes = lhist(retval, 0, 10000, 1000); exit()}
 EXPECT @bytes: *\n[\[(].*

tests/runtime/outputs/hist_10g.json

@@ -0,0 +1,12 @@
+{
+  "type": "hist",
+  "data": {
+    "@": [
+      {
+        "min": 8589934592,
+        "max": 17179869183,
+        "count": 1
+      }
+    ]
+  }
+}