Skip to content

Commit b037d0f

Browse files
[BOLT][docs] Expand Heatmaps.md (#98162)
Improve documentation on heatmaps. Add example for X axis labels.
1 parent b1bcb7c commit b037d0f

File tree

2 files changed

+56
-13
lines changed

2 files changed

+56
-13
lines changed

bolt/docs/HeatmapHeader.png

75 KB
Loading

bolt/docs/Heatmaps.md

Lines changed: 56 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,9 @@
11
# Code Heatmaps
22

33
BOLT has gained the ability to print code heatmaps based on
4-
sampling-based LBR profiles generated by `perf`. The output is produced
5-
in colored ASCII to be displayed in a color-capable terminal. It looks
6-
something like this:
4+
sampling-based profiles generated by `perf`, either with `LBR` data or not.
5+
The output is produced in colored ASCII to be displayed in a color-capable
6+
terminal. It looks something like this:
77

88
![](./Heatmap.png)
99

@@ -32,21 +32,64 @@ $ llvm-bolt-heatmap -p perf.data <executable>
3232
```
3333

3434
By default the heatmap will be dumped to *stdout*. You can change it
35-
with `-o <heatmapfile>` option. Each character/block in the heatmap
36-
shows the execution data accumulated for corresponding 64 bytes of
37-
code. You can change this granularity with a `-block-size` option.
38-
E.g. set it to 4096 to see code usage grouped by 4K pages.
39-
Other useful options are:
35+
with `-o <heatmapfile>` option.
4036

41-
```bash
42-
-line-size=<uint> - number of entries per line (default 256)
43-
-max-address=<uint> - maximum address considered valid for heatmap (default 4GB)
44-
-print-mappings=<bool> - print mappings in legend, between characters/blocks and text sections (default false)
45-
```
4637

4738
If you prefer to look at the data in a browser (or would like to share
4839
it that way), then you can use an HTML conversion tool. E.g.:
4940

5041
```bash
5142
$ aha -b -f <heatmapfile> > <heatmapfile>.html
5243
```
44+
45+
---
46+
47+
## Background on heatmaps:
48+
A heatmap is effectively a histogram that is rendered into a grid for better
49+
visualization.
50+
In theory we can generate a heatmap using any binary and a perf profile.
51+
52+
Each block/character in the heatmap shows the execution data accumulated for
53+
corresponding 64 bytes of code. You can change this granularity with a
54+
`-block-size` option.
55+
E.g. set it to 4096 to see code usage grouped by 4K pages.
56+
57+
58+
When a block is shown as a dot, it means that no samples were found for that
59+
address.
60+
When it is shown as a letter, it indicates a captured sample on a particular
61+
text section of the binary.
62+
To show a mapping between letters and text sections in the legend, use
63+
`-print-mappings`.
64+
When a sampled address does not belong to any of the text sections, the
65+
characters 'o' or 'O' will be shown.
66+
67+
The legend shows by default the ranges in the heatmap according to the number
68+
of samples per block.
69+
A color is assigned per range, except the first two ranges that distinguished by
70+
lower and upper case letters.
71+
72+
On the Y axis, each row/line starts with an actual address of the binary.
73+
Consecutive lines in the heatmap advance by the same amount, with the binary
74+
size covered by a line dependent on the block size and the line size.
75+
An empty new line is inserted for larger gaps between samples.
76+
77+
On the X axis, the horizontally emitted hex numbers can help *estimate* where
78+
in the line the samples lie, but they cannot be combined to provide a full
79+
address, as they are relative to both the bucket and line sizes.
80+
81+
In the example below, the highlighted `0x100` column is not an offset to each
82+
row's address, but instead, it points to the middle of the line.
83+
For the generation, the default bucket size was used with a line size of 128.
84+
85+
86+
![](./HeatmapHeader.png)
87+
88+
89+
Some useful options are:
90+
91+
```
92+
-line-size=<uint> - number of entries per line (default 256)
93+
-max-address=<uint> - maximum address considered valid for heatmap (default 4GB)
94+
-print-mappings - print mappings in the legend, between characters/blocks and text sections (default false)
95+
```

0 commit comments

Comments
 (0)