[llvm-objdump] Add the --visualize-jumps option #74858
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-llvm-binary-utilities Author: None (ostannard) ChangesThis is a feature which GNU objdump currently has, which prints a control-flow graph showing all of the direct branches alongside the disassembly. This should work for any architecture which implements the MCInstrAnalysis class, I've tested it on ARM, Thumb and AArch64. There are a few (known) differences between this and the GNU objdump
Related issue: #60172 Demo: Patch is 60.76 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/74858.diff 19 Files Affected:
diff --git a/llvm/lib/Support/FormattedStream.cpp b/llvm/lib/Support/FormattedStream.cpp
index c0d28435099570..48b02889c6d3e5 100644
--- a/llvm/lib/Support/FormattedStream.cpp
+++ b/llvm/lib/Support/FormattedStream.cpp
@@ -45,6 +45,7 @@ void formatted_raw_ostream::UpdatePosition(const char *Ptr, size_t Size) {
break;
case '\t':
// Assumes tab stop = 8 characters.
+ Column += 1;
Column += (8 - (Column & 0x7)) & 0x7;
break;
}
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-ascii.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-ascii.txt
new file mode 100644
index 00000000000000..592e50d98f8ff7
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-ascii.txt
@@ -0,0 +1,31 @@
+
+<stdin>: file format elf64-littleaarch64
+
+Disassembly of section .text:
+
+0000000000000000 <test_func>:
+ 0: 94000000 bl 0x0 <test_func>
+ 4: 14000000 b 0x4 <test_func+0x4>
+ 8: /-- 14000001 b 0xc <test_func+0xc>
+ c: +-> d503201f nop
+ 10: \-- 17ffffff b 0xc <test_func+0xc>
+ 14: 14000000 b 0x14 <test_func+0x14>
+ 18: /-- 54000040 b.eq 0x20 <test_func+0x20>
+ 1c: +-- b4000020 cbz x0, 0x20 <test_func+0x20>
+ 20: \-> d503201f nop
+ 24: /------ 14000005 b 0x38 <test_func+0x38>
+ 28: | /---- 14000003 b 0x34 <test_func+0x34>
+ 2c: | | /-- 14000001 b 0x30 <test_func+0x30>
+ 30: | | \-> d503201f nop
+ 34: | \---> d503201f nop
+ 38: \-----> d503201f nop
+ 3c: /-- 14000002 b 0x44 <test_func+0x44>
+ 40: /-|-- 14000002 b 0x48 <test_func+0x48>
+ 44: | \-> d503201f nop
+ 48: \---> d503201f nop
+ 4c: /-- 14000001 b 0x50 <test_func+0x50>
+ 50: /-|-- 14000001 b 0x54 <test_func+0x54>
+ 54: \---> d503201f nop
+ 58: /-- 14000002 b 0x60 <test_func+0x60>
+ 5c: | 94000000 bl 0x5c <test_func+0x5c>
+ 60: \-> d503201f nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-color.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-color.txt
new file mode 100644
index 00000000000000..925c5ef52a8d22
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-color.txt
@@ -0,0 +1,31 @@
+
+<stdin>: file format elf64-littleaarch64
+
+Disassembly of section .text:
+
+0000000000000000 <test_func>:
+ 0: �[0m 94000000 bl 0x0 <test_func>
+ 4: �[0m 14000000 b 0x4 <test_func+0x4>
+ 8: �[0;31m ╭�[0;31m──�[0m 14000001 b 0xc <test_func+0xc>
+ c: �[0;31m ├�[0;31m─>�[0m d503201f nop
+ 10: �[0;31m ╰�[0;31m──�[0m 17ffffff b 0xc <test_func+0xc>
+ 14: �[0m 14000000 b 0x14 <test_func+0x14>
+ 18: �[0;32m ╭�[0;32m──�[0m 54000040 b.eq 0x20 <test_func+0x20>
+ 1c: �[0;32m ├�[0;32m──�[0m b4000020 cbz x0, 0x20 <test_func+0x20>
+ 20: �[0;32m ╰�[0;32m─>�[0m d503201f nop
+ 24: �[0;33m ╭�[0;33m──�[0;33m──�[0;33m──�[0m 14000005 b 0x38 <test_func+0x38>
+ 28: �[0;33m │�[0;34m ╭�[0;34m──�[0;34m──�[0m 14000003 b 0x34 <test_func+0x34>
+ 2c: �[0;33m │�[0;34m │�[0;35m ╭�[0;35m──�[0m 14000001 b 0x30 <test_func+0x30>
+ 30: �[0;33m │�[0;34m │�[0;35m ╰�[0;35m─>�[0m d503201f nop
+ 34: �[0;33m │�[0;34m ╰�[0;34m──�[0;34m─>�[0m d503201f nop
+ 38: �[0;33m ╰�[0;33m──�[0;33m──�[0;33m─>�[0m d503201f nop
+ 3c: �[0;36m ╭�[0;36m──�[0m 14000002 b 0x44 <test_func+0x44>
+ 40: �[0;31m ╭�[0;31m─�[0;36m│�[0;31m──�[0m 14000002 b 0x48 <test_func+0x48>
+ 44: �[0;31m │�[0;36m ╰�[0;36m─>�[0m d503201f nop
+ 48: �[0;31m ╰�[0;31m──�[0;31m─>�[0m d503201f nop
+ 4c: �[0;32m ╭�[0;32m──�[0m 14000001 b 0x50 <test_func+0x50>
+ 50: �[0;33m ╭�[0;33m─�[0;32m│�[0;33m──�[0m 14000001 b 0x54 <test_func+0x54>
+ 54: �[0;33m ╰�[0;33m──�[0;33m─>�[0m d503201f nop
+ 58: �[0;34m ╭�[0;34m──�[0m 14000002 b 0x60 <test_func+0x60>
+ 5c: �[0;34m │ �[0m 94000000 bl 0x5c <test_func+0x5c>
+ 60: �[0;34m ╰�[0;34m─>�[0m d503201f nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-relocs.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-relocs.txt
new file mode 100644
index 00000000000000..75ec5741fdaae2
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-relocs.txt
@@ -0,0 +1,34 @@
+
+<stdin>: file format elf64-littleaarch64
+
+Disassembly of section .text:
+
+0000000000000000 <test_func>:
+ 0: 94000000 bl 0x0 <test_func>
+ 0000000000000000: R_AARCH64_CALL26 extern_func
+ 4: 14000000 b 0x4 <test_func+0x4>
+ 0000000000000004: R_AARCH64_JUMP26 extern_func
+ 8: ╭── 14000001 b 0xc <test_func+0xc>
+ c: ├─> d503201f nop
+ 10: ╰── 17ffffff b 0xc <test_func+0xc>
+ 14: 14000000 b 0x14 <test_func+0x14>
+ 18: ╭── 54000040 b.eq 0x20 <test_func+0x20>
+ 1c: ├── b4000020 cbz x0, 0x20 <test_func+0x20>
+ 20: ╰─> d503201f nop
+ 24: ╭────── 14000005 b 0x38 <test_func+0x38>
+ 28: │ ╭──── 14000003 b 0x34 <test_func+0x34>
+ 2c: │ │ ╭── 14000001 b 0x30 <test_func+0x30>
+ 30: │ │ ╰─> d503201f nop
+ 34: │ ╰───> d503201f nop
+ 38: ╰─────> d503201f nop
+ 3c: ╭── 14000002 b 0x44 <test_func+0x44>
+ 40: ╭─│── 14000002 b 0x48 <test_func+0x48>
+ 44: │ ╰─> d503201f nop
+ 48: ╰───> d503201f nop
+ 4c: ╭── 14000001 b 0x50 <test_func+0x50>
+ 50: ╭─│── 14000001 b 0x54 <test_func+0x54>
+ 54: ╰───> d503201f nop
+ 58: ╭── 14000002 b 0x60 <test_func+0x60>
+ 5c: │ 94000000 bl 0x5c <test_func+0x5c>
+ │ 000000000000005c: R_AARCH64_CALL26 extern_func
+ 60: ╰─> d503201f nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode.txt
new file mode 100644
index 00000000000000..4cabdc92d61c1d
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode.txt
@@ -0,0 +1,31 @@
+
+<stdin>: file format elf64-littleaarch64
+
+Disassembly of section .text:
+
+0000000000000000 <test_func>:
+ 0: 94000000 bl 0x0 <test_func>
+ 4: 14000000 b 0x4 <test_func+0x4>
+ 8: ╭── 14000001 b 0xc <test_func+0xc>
+ c: ├─> d503201f nop
+ 10: ╰── 17ffffff b 0xc <test_func+0xc>
+ 14: 14000000 b 0x14 <test_func+0x14>
+ 18: ╭── 54000040 b.eq 0x20 <test_func+0x20>
+ 1c: ├── b4000020 cbz x0, 0x20 <test_func+0x20>
+ 20: ╰─> d503201f nop
+ 24: ╭────── 14000005 b 0x38 <test_func+0x38>
+ 28: │ ╭──── 14000003 b 0x34 <test_func+0x34>
+ 2c: │ │ ╭── 14000001 b 0x30 <test_func+0x30>
+ 30: │ │ ╰─> d503201f nop
+ 34: │ ╰───> d503201f nop
+ 38: ╰─────> d503201f nop
+ 3c: ╭── 14000002 b 0x44 <test_func+0x44>
+ 40: ╭─│── 14000002 b 0x48 <test_func+0x48>
+ 44: │ ╰─> d503201f nop
+ 48: ╰───> d503201f nop
+ 4c: ╭── 14000001 b 0x50 <test_func+0x50>
+ 50: ╭─│── 14000001 b 0x54 <test_func+0x54>
+ 54: ╰───> d503201f nop
+ 58: ╭── 14000002 b 0x60 <test_func+0x60>
+ 5c: │ 94000000 bl 0x5c <test_func+0x5c>
+ 60: ╰─> d503201f nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-ascii.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-ascii.txt
new file mode 100644
index 00000000000000..7e006841980534
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-ascii.txt
@@ -0,0 +1,27 @@
+
+<stdin>: file format elf32-littlearm
+
+Disassembly of section .text:
+
+00000000 <test_func>:
+ 0: ebfffffe bl 0x0 <test_func> @ imm = #-0x8
+ 4: eafffffe b 0x4 <test_func+0x4> @ imm = #-0x8
+ 8: /-- eaffffff b 0xc <test_func+0xc> @ imm = #-0x4
+ c: +-> e320f000 nop
+ 10: \-- eafffffd b 0xc <test_func+0xc> @ imm = #-0xc
+ 14: eafffffe b 0x14 <test_func+0x14> @ imm = #-0x8
+ 18: /-- 0affffff beq 0x1c <test_func+0x1c> @ imm = #-0x4
+ 1c: \-> e320f000 nop
+ 20: /------ ea000003 b 0x34 <test_func+0x34> @ imm = #0xc
+ 24: | /---- ea000001 b 0x30 <test_func+0x30> @ imm = #0x4
+ 28: | | /-- eaffffff b 0x2c <test_func+0x2c> @ imm = #-0x4
+ 2c: | | \-> e320f000 nop
+ 30: | \---> e320f000 nop
+ 34: \-----> e320f000 nop
+ 38: /-- ea000000 b 0x40 <test_func+0x40> @ imm = #0x0
+ 3c: /-|-- ea000000 b 0x44 <test_func+0x44> @ imm = #0x0
+ 40: | \-> e320f000 nop
+ 44: \---> e320f000 nop
+ 48: /---- eaffffff b 0x4c <test_func+0x4c> @ imm = #-0x4
+ 4c: \-|-> eaffffff b 0x50 <test_func+0x50> @ imm = #-0x4
+ 50: \-> e320f000 nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-unicode.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-unicode.txt
new file mode 100644
index 00000000000000..62194b4b33c172
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-unicode.txt
@@ -0,0 +1,27 @@
+
+<stdin>: file format elf32-littlearm
+
+Disassembly of section .text:
+
+00000000 <test_func>:
+ 0: ebfffffe bl 0x0 <test_func> @ imm = #-0x8
+ 4: eafffffe b 0x4 <test_func+0x4> @ imm = #-0x8
+ 8: ╭── eaffffff b 0xc <test_func+0xc> @ imm = #-0x4
+ c: ├─> e320f000 nop
+ 10: ╰── eafffffd b 0xc <test_func+0xc> @ imm = #-0xc
+ 14: eafffffe b 0x14 <test_func+0x14> @ imm = #-0x8
+ 18: ╭── 0affffff beq 0x1c <test_func+0x1c> @ imm = #-0x4
+ 1c: ╰─> e320f000 nop
+ 20: ╭────── ea000003 b 0x34 <test_func+0x34> @ imm = #0xc
+ 24: │ ╭──── ea000001 b 0x30 <test_func+0x30> @ imm = #0x4
+ 28: │ │ ╭── eaffffff b 0x2c <test_func+0x2c> @ imm = #-0x4
+ 2c: │ │ ╰─> e320f000 nop
+ 30: │ ╰───> e320f000 nop
+ 34: ╰─────> e320f000 nop
+ 38: ╭── ea000000 b 0x40 <test_func+0x40> @ imm = #0x0
+ 3c: ╭─│── ea000000 b 0x44 <test_func+0x44> @ imm = #0x0
+ 40: │ ╰─> e320f000 nop
+ 44: ╰───> e320f000 nop
+ 48: ╭──── eaffffff b 0x4c <test_func+0x4c> @ imm = #-0x4
+ 4c: ╰─│─> eaffffff b 0x50 <test_func+0x50> @ imm = #-0x4
+ 50: ╰─> e320f000 nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-ascii.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-ascii.txt
new file mode 100644
index 00000000000000..f18062167dd97d
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-ascii.txt
@@ -0,0 +1,34 @@
+
+<stdin>: file format elf32-littlearm
+
+Disassembly of section .text:
+
+00000000 <test_func>:
+ 0: f7ff fffe bl 0x0 <test_func> @ imm = #-0x4
+ 4: f7ff bffe b.w 0x4 <test_func+0x4> @ imm = #-0x4
+ 8: /-- e7ff b 0xa <test_func+0xa> @ imm = #-0x2
+ a: +-> bf00 nop
+ c: \-- e7fd b 0xa <test_func+0xa> @ imm = #-0x6
+ e: e7fe b 0xe <test_func+0xe> @ imm = #-0x4
+ 10: /-- f000 b807 b.w 0x22 <test_func+0x22> @ imm = #0xe
+ 14: +-- d005 beq 0x22 <test_func+0x22> @ imm = #0xa
+ 16: +-- f040 8004 bne.w 0x22 <test_func+0x22> @ imm = #0x8
+ 1a: +-- b110 cbz r0, 0x22 <test_func+0x22> @ imm = #0x4
+ 1c: | bfd8 it le
+ 1e: +-- e000 ble 0x22 <test_func+0x22> @ imm = #0x0
+ 20: | bf00 nop
+ 22: \-> bf00 nop
+ 24: bf00 nop
+ 26: /------ e003 b 0x30 <test_func+0x30> @ imm = #0x6
+ 28: | /---- e001 b 0x2e <test_func+0x2e> @ imm = #0x2
+ 2a: | | /-- e7ff b 0x2c <test_func+0x2c> @ imm = #-0x2
+ 2c: | | \-> bf00 nop
+ 2e: | \---> bf00 nop
+ 30: \-----> bf00 nop
+ 32: /-- e000 b 0x36 <test_func+0x36> @ imm = #0x0
+ 34: /-|-- e000 b 0x38 <test_func+0x38> @ imm = #0x0
+ 36: | \-> bf00 nop
+ 38: \---> bf00 nop
+ 3a: /-- e7ff b 0x3c <test_func+0x3c> @ imm = #-0x2
+ 3c: /-|-- e7ff b 0x3e <test_func+0x3e> @ imm = #-0x2
+ 3e: \---> bf00 nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-unicode.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-unicode.txt
new file mode 100644
index 00000000000000..b171fb1c5d28c2
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-unicode.txt
@@ -0,0 +1,34 @@
+
+<stdin>: file format elf32-littlearm
+
+Disassembly of section .text:
+
+00000000 <test_func>:
+ 0: f7ff fffe bl 0x0 <test_func> @ imm = #-0x4
+ 4: f7ff bffe b.w 0x4 <test_func+0x4> @ imm = #-0x4
+ 8: ╭── e7ff b 0xa <test_func+0xa> @ imm = #-0x2
+ a: ├─> bf00 nop
+ c: ╰── e7fd b 0xa <test_func+0xa> @ imm = #-0x6
+ e: e7fe b 0xe <test_func+0xe> @ imm = #-0x4
+ 10: ╭── f000 b807 b.w 0x22 <test_func+0x22> @ imm = #0xe
+ 14: ├── d005 beq 0x22 <test_func+0x22> @ imm = #0xa
+ 16: ├── f040 8004 bne.w 0x22 <test_func+0x22> @ imm = #0x8
+ 1a: ├── b110 cbz r0, 0x22 <test_func+0x22> @ imm = #0x4
+ 1c: │ bfd8 it le
+ 1e: ├── e000 ble 0x22 <test_func+0x22> @ imm = #0x0
+ 20: │ bf00 nop
+ 22: ╰─> bf00 nop
+ 24: bf00 nop
+ 26: ╭────── e003 b 0x30 <test_func+0x30> @ imm = #0x6
+ 28: │ ╭──── e001 b 0x2e <test_func+0x2e> @ imm = #0x2
+ 2a: │ │ ╭── e7ff b 0x2c <test_func+0x2c> @ imm = #-0x2
+ 2c: │ │ ╰─> bf00 nop
+ 2e: │ ╰───> bf00 nop
+ 30: ╰─────> bf00 nop
+ 32: ╭── e000 b 0x36 <test_func+0x36> @ imm = #0x0
+ 34: ╭─│── e000 b 0x38 <test_func+0x38> @ imm = #0x0
+ 36: │ ╰─> bf00 nop
+ 38: ╰───> bf00 nop
+ 3a: ╭── e7ff b 0x3c <test_func+0x3c> @ imm = #-0x2
+ 3c: ╭─│── e7ff b 0x3e <test_func+0x3e> @ imm = #-0x2
+ 3e: ╰───> bf00 nop
diff --git a/llvm/test/tools/llvm-objdump/visualize-jumps-aarch64.s b/llvm/test/tools/llvm-objdump/visualize-jumps-aarch64.s
new file mode 100644
index 00000000000000..f78b3e27975556
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/visualize-jumps-aarch64.s
@@ -0,0 +1,69 @@
+// RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \
+// RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=unicode - | \
+// RUN: diff - %p/Inputs/visualize-jumps-aarch64-unicode.txt
+
+// RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \
+// RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=ascii - | \
+// RUN: diff - %p/Inputs/visualize-jumps-aarch64-ascii.txt
+
+// RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \
+// RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=unicode,color - | \
+// RUN: diff - %p/Inputs/visualize-jumps-aarch64-unicode-color.txt
+
+// RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \
+// RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=unicode --reloc - | \
+// RUN: diff - %p/Inputs/visualize-jumps-aarch64-unicode-relocs.txt
+
+test_func:
+ // Relocated instructions don't get control-flow edges.
+ bl extern_func
+ b extern_func
+
+ // Two branches to the same label, one forward and one backward.
+ b .Llabel1
+.Llabel1:
+ nop
+ b .Llabel1
+
+ // Branch to self, no CFG edge shown
+ b .
+
+ // Conditional branches
+ b.eq .Llabel2
+ cbz x0, .Llabel2
+.Llabel2:
+ nop
+
+ // Branches are sorted with shorter ones to the right, to reduce number of
+ // crossings, and keep the lines for short branches short themselves.
+ b .Llabel5
+ b .Llabel4
+ b .Llabel3
+.Llabel3:
+ nop
+.Llabel4:
+ nop
+.Llabel5:
+ nop
+
+ // Sometimes crossings can't be avoided.
+ b .Llabel6
+ b .Llabel7
+.Llabel6:
+ nop
+.Llabel7:
+ nop
+
+ // TODO If a branch goes to another branch instruction, we don't have a way
+ // to represent that. Can we improve on this?
+ b .Llabel8
+.Llabel8:
+ b .Llabel9
+.Llabel9:
+ nop
+
+ // Graph lines need to be drawn on the same output line as relocations.
+ b .Llabel10
+ bl extern_func
+.Llabel10:
+ nop
diff --git a/llvm/test/tools/llvm-objdump/visualize-jumps-arm.s b/llvm/test/tools/llvm-objdump/visualize-jumps-arm.s
new file mode 100644
index 00000000000000..6855e6ff84e32a
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/visualize-jumps-arm.s
@@ -0,0 +1,54 @@
+// RUN: llvm-mc < %s -triple armv8a -filetype=obj | \
+// RUN: llvm-objdump --triple armv8a -d --visualize-jumps=unicode - | \
+// RUN: diff - %p/Inputs/visualize-jumps-arm-unicode.txt
+
+// RUN: llvm-mc < %s -triple armv8a -filetype=obj | \
+// RUN: llvm-objdump --triple armv8a -d --visualize-jumps=ascii - | \
+// RUN: diff - %p/Inputs/visualize-jumps-arm-ascii.txt
+
+test_func:
+ // Relocated instructions don't get control-flow edges.
+ bl extern_func
+ b extern_func
+
+ // Two branches to the same label, one forward and one backward.
+ b .Llabel1
+.Llabel1:
+ nop
+ b .Llabel1
+
+ // Branch to self, no CFG edge shown
+ b .
+
+ // Conditional branches
+ beq .Llabel2
+.Llabel2:
+ nop
+
+ // Branches are sorted with shorter ones to the right, to reduce number of
+ // crossings, and keep the lines for short branches short themselves.
+ b .Llabel5
+ b .Llabel4
+ b .Llabel3
+.Llabel3:
+ nop
+.Llabel4:
+ nop
+.Llabel5:
+ nop
+
+ // Sometimes crossings can't be avoided.
+ b .Llabel6
+ b .Llabel7
+.Llabel6:
+ nop
+.Llabel7:
+ nop
+
+ // TODO If a branch goes to another branch instruction, we don't have a way
+ // to represent that. Can we improve on this?
+ b .Llabel8
+.Llabel8:
+ b .Llabel9
+.Llabel9:
+ nop
diff --git a/llvm/test/tools/llvm-objdump/visualize-jumps-thumb.s b/llvm/test/tools/llvm-objdump/visualize-jumps-thumb.s
new file mode 100644
index 00000000000000..8be73d3aa4c946
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/visualize-jumps-thumb.s
@@ -0,0 +1,63 @@
+// RUN: llvm-mc < %s -triple thumbv8a -filetype=obj | \
+// RUN: llvm-objdump --triple thumbv8a -d --visualize-jumps=unicode - | \
+// RUN: diff - %p/Inputs/visualize-jumps-thumb-unicode.txt
+
+// RUN: llvm-mc < %s -triple thumbv8a -filetype=obj | \
+// RUN: llvm-objdump --triple thumbv8a -d --visualize-jumps=ascii - | \
+// RUN: diff - %p/Inputs/visualize-jumps-thumb-ascii.txt
+
+test_func:
+ // Relocated instructions don't get control-flow edges.
+ bl extern_func
+ b extern_func
+
+ // Two branches to the same label, one forward and one backward.
+ b .Llabel1
+.Llabel1:
+ nop
+ b .Llabel1
+
+ // Branch to self, no CFG edge shown
+ b .
+
+ // Different branch instructions
+ b.w .Llabel2
+ beq .Llabel2
+ bne.w .Llabel2
+ cbz r0, .Llabel2
+ it le
+ ble .Llabel2
+ nop
+.Llabel2:
+ nop
+.Llabel2.1:
+ nop
+
+ // Branches are sorted with shorter ones to the right, to reduce number of
+ // crossings, and keep the lines for short branches short themselves.
+ b .Llabel5
+ b .Llabel4
+ b .Llabel3
+.Llabel3:
+ nop
+.Llabel4:
+ nop
+.Llabel5:
+ nop
+
+ // Sometimes crossings can't be avoided.
+ b ....
[truncated]
|
|
@llvm/pr-subscribers-llvm-support Author: None (ostannard) ChangesThis is a feature which GNU objdump currently has, which prints a control-flow graph showing all of the direct branches alongside the disassembly. This should work for any architecture which implements the MCInstrAnalysis class, I've tested it on ARM, Thumb and AArch64. There are a few (known) differences between this and the GNU objdump
Related issue: #60172 Demo: Patch is 60.76 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/74858.diff 19 Files Affected:
diff --git a/llvm/lib/Support/FormattedStream.cpp b/llvm/lib/Support/FormattedStream.cpp
index c0d28435099570..48b02889c6d3e5 100644
--- a/llvm/lib/Support/FormattedStream.cpp
+++ b/llvm/lib/Support/FormattedStream.cpp
@@ -45,6 +45,7 @@ void formatted_raw_ostream::UpdatePosition(const char *Ptr, size_t Size) {
break;
case '\t':
// Assumes tab stop = 8 characters.
+ Column += 1;
Column += (8 - (Column & 0x7)) & 0x7;
break;
}
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-ascii.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-ascii.txt
new file mode 100644
index 00000000000000..592e50d98f8ff7
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-ascii.txt
@@ -0,0 +1,31 @@
+
+<stdin>: file format elf64-littleaarch64
+
+Disassembly of section .text:
+
+0000000000000000 <test_func>:
+ 0: 94000000 bl 0x0 <test_func>
+ 4: 14000000 b 0x4 <test_func+0x4>
+ 8: /-- 14000001 b 0xc <test_func+0xc>
+ c: +-> d503201f nop
+ 10: \-- 17ffffff b 0xc <test_func+0xc>
+ 14: 14000000 b 0x14 <test_func+0x14>
+ 18: /-- 54000040 b.eq 0x20 <test_func+0x20>
+ 1c: +-- b4000020 cbz x0, 0x20 <test_func+0x20>
+ 20: \-> d503201f nop
+ 24: /------ 14000005 b 0x38 <test_func+0x38>
+ 28: | /---- 14000003 b 0x34 <test_func+0x34>
+ 2c: | | /-- 14000001 b 0x30 <test_func+0x30>
+ 30: | | \-> d503201f nop
+ 34: | \---> d503201f nop
+ 38: \-----> d503201f nop
+ 3c: /-- 14000002 b 0x44 <test_func+0x44>
+ 40: /-|-- 14000002 b 0x48 <test_func+0x48>
+ 44: | \-> d503201f nop
+ 48: \---> d503201f nop
+ 4c: /-- 14000001 b 0x50 <test_func+0x50>
+ 50: /-|-- 14000001 b 0x54 <test_func+0x54>
+ 54: \---> d503201f nop
+ 58: /-- 14000002 b 0x60 <test_func+0x60>
+ 5c: | 94000000 bl 0x5c <test_func+0x5c>
+ 60: \-> d503201f nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-color.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-color.txt
new file mode 100644
index 00000000000000..925c5ef52a8d22
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-color.txt
@@ -0,0 +1,31 @@
+
+<stdin>: file format elf64-littleaarch64
+
+Disassembly of section .text:
+
+0000000000000000 <test_func>:
+ 0: �[0m 94000000 bl 0x0 <test_func>
+ 4: �[0m 14000000 b 0x4 <test_func+0x4>
+ 8: �[0;31m ╭�[0;31m──�[0m 14000001 b 0xc <test_func+0xc>
+ c: �[0;31m ├�[0;31m─>�[0m d503201f nop
+ 10: �[0;31m ╰�[0;31m──�[0m 17ffffff b 0xc <test_func+0xc>
+ 14: �[0m 14000000 b 0x14 <test_func+0x14>
+ 18: �[0;32m ╭�[0;32m──�[0m 54000040 b.eq 0x20 <test_func+0x20>
+ 1c: �[0;32m ├�[0;32m──�[0m b4000020 cbz x0, 0x20 <test_func+0x20>
+ 20: �[0;32m ╰�[0;32m─>�[0m d503201f nop
+ 24: �[0;33m ╭�[0;33m──�[0;33m──�[0;33m──�[0m 14000005 b 0x38 <test_func+0x38>
+ 28: �[0;33m │�[0;34m ╭�[0;34m──�[0;34m──�[0m 14000003 b 0x34 <test_func+0x34>
+ 2c: �[0;33m │�[0;34m │�[0;35m ╭�[0;35m──�[0m 14000001 b 0x30 <test_func+0x30>
+ 30: �[0;33m │�[0;34m │�[0;35m ╰�[0;35m─>�[0m d503201f nop
+ 34: �[0;33m │�[0;34m ╰�[0;34m──�[0;34m─>�[0m d503201f nop
+ 38: �[0;33m ╰�[0;33m──�[0;33m──�[0;33m─>�[0m d503201f nop
+ 3c: �[0;36m ╭�[0;36m──�[0m 14000002 b 0x44 <test_func+0x44>
+ 40: �[0;31m ╭�[0;31m─�[0;36m│�[0;31m──�[0m 14000002 b 0x48 <test_func+0x48>
+ 44: �[0;31m │�[0;36m ╰�[0;36m─>�[0m d503201f nop
+ 48: �[0;31m ╰�[0;31m──�[0;31m─>�[0m d503201f nop
+ 4c: �[0;32m ╭�[0;32m──�[0m 14000001 b 0x50 <test_func+0x50>
+ 50: �[0;33m ╭�[0;33m─�[0;32m│�[0;33m──�[0m 14000001 b 0x54 <test_func+0x54>
+ 54: �[0;33m ╰�[0;33m──�[0;33m─>�[0m d503201f nop
+ 58: �[0;34m ╭�[0;34m──�[0m 14000002 b 0x60 <test_func+0x60>
+ 5c: �[0;34m │ �[0m 94000000 bl 0x5c <test_func+0x5c>
+ 60: �[0;34m ╰�[0;34m─>�[0m d503201f nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-relocs.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-relocs.txt
new file mode 100644
index 00000000000000..75ec5741fdaae2
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode-relocs.txt
@@ -0,0 +1,34 @@
+
+<stdin>: file format elf64-littleaarch64
+
+Disassembly of section .text:
+
+0000000000000000 <test_func>:
+ 0: 94000000 bl 0x0 <test_func>
+ 0000000000000000: R_AARCH64_CALL26 extern_func
+ 4: 14000000 b 0x4 <test_func+0x4>
+ 0000000000000004: R_AARCH64_JUMP26 extern_func
+ 8: ╭── 14000001 b 0xc <test_func+0xc>
+ c: ├─> d503201f nop
+ 10: ╰── 17ffffff b 0xc <test_func+0xc>
+ 14: 14000000 b 0x14 <test_func+0x14>
+ 18: ╭── 54000040 b.eq 0x20 <test_func+0x20>
+ 1c: ├── b4000020 cbz x0, 0x20 <test_func+0x20>
+ 20: ╰─> d503201f nop
+ 24: ╭────── 14000005 b 0x38 <test_func+0x38>
+ 28: │ ╭──── 14000003 b 0x34 <test_func+0x34>
+ 2c: │ │ ╭── 14000001 b 0x30 <test_func+0x30>
+ 30: │ │ ╰─> d503201f nop
+ 34: │ ╰───> d503201f nop
+ 38: ╰─────> d503201f nop
+ 3c: ╭── 14000002 b 0x44 <test_func+0x44>
+ 40: ╭─│── 14000002 b 0x48 <test_func+0x48>
+ 44: │ ╰─> d503201f nop
+ 48: ╰───> d503201f nop
+ 4c: ╭── 14000001 b 0x50 <test_func+0x50>
+ 50: ╭─│── 14000001 b 0x54 <test_func+0x54>
+ 54: ╰───> d503201f nop
+ 58: ╭── 14000002 b 0x60 <test_func+0x60>
+ 5c: │ 94000000 bl 0x5c <test_func+0x5c>
+ │ 000000000000005c: R_AARCH64_CALL26 extern_func
+ 60: ╰─> d503201f nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode.txt
new file mode 100644
index 00000000000000..4cabdc92d61c1d
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-aarch64-unicode.txt
@@ -0,0 +1,31 @@
+
+<stdin>: file format elf64-littleaarch64
+
+Disassembly of section .text:
+
+0000000000000000 <test_func>:
+ 0: 94000000 bl 0x0 <test_func>
+ 4: 14000000 b 0x4 <test_func+0x4>
+ 8: ╭── 14000001 b 0xc <test_func+0xc>
+ c: ├─> d503201f nop
+ 10: ╰── 17ffffff b 0xc <test_func+0xc>
+ 14: 14000000 b 0x14 <test_func+0x14>
+ 18: ╭── 54000040 b.eq 0x20 <test_func+0x20>
+ 1c: ├── b4000020 cbz x0, 0x20 <test_func+0x20>
+ 20: ╰─> d503201f nop
+ 24: ╭────── 14000005 b 0x38 <test_func+0x38>
+ 28: │ ╭──── 14000003 b 0x34 <test_func+0x34>
+ 2c: │ │ ╭── 14000001 b 0x30 <test_func+0x30>
+ 30: │ │ ╰─> d503201f nop
+ 34: │ ╰───> d503201f nop
+ 38: ╰─────> d503201f nop
+ 3c: ╭── 14000002 b 0x44 <test_func+0x44>
+ 40: ╭─│── 14000002 b 0x48 <test_func+0x48>
+ 44: │ ╰─> d503201f nop
+ 48: ╰───> d503201f nop
+ 4c: ╭── 14000001 b 0x50 <test_func+0x50>
+ 50: ╭─│── 14000001 b 0x54 <test_func+0x54>
+ 54: ╰───> d503201f nop
+ 58: ╭── 14000002 b 0x60 <test_func+0x60>
+ 5c: │ 94000000 bl 0x5c <test_func+0x5c>
+ 60: ╰─> d503201f nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-ascii.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-ascii.txt
new file mode 100644
index 00000000000000..7e006841980534
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-ascii.txt
@@ -0,0 +1,27 @@
+
+<stdin>: file format elf32-littlearm
+
+Disassembly of section .text:
+
+00000000 <test_func>:
+ 0: ebfffffe bl 0x0 <test_func> @ imm = #-0x8
+ 4: eafffffe b 0x4 <test_func+0x4> @ imm = #-0x8
+ 8: /-- eaffffff b 0xc <test_func+0xc> @ imm = #-0x4
+ c: +-> e320f000 nop
+ 10: \-- eafffffd b 0xc <test_func+0xc> @ imm = #-0xc
+ 14: eafffffe b 0x14 <test_func+0x14> @ imm = #-0x8
+ 18: /-- 0affffff beq 0x1c <test_func+0x1c> @ imm = #-0x4
+ 1c: \-> e320f000 nop
+ 20: /------ ea000003 b 0x34 <test_func+0x34> @ imm = #0xc
+ 24: | /---- ea000001 b 0x30 <test_func+0x30> @ imm = #0x4
+ 28: | | /-- eaffffff b 0x2c <test_func+0x2c> @ imm = #-0x4
+ 2c: | | \-> e320f000 nop
+ 30: | \---> e320f000 nop
+ 34: \-----> e320f000 nop
+ 38: /-- ea000000 b 0x40 <test_func+0x40> @ imm = #0x0
+ 3c: /-|-- ea000000 b 0x44 <test_func+0x44> @ imm = #0x0
+ 40: | \-> e320f000 nop
+ 44: \---> e320f000 nop
+ 48: /---- eaffffff b 0x4c <test_func+0x4c> @ imm = #-0x4
+ 4c: \-|-> eaffffff b 0x50 <test_func+0x50> @ imm = #-0x4
+ 50: \-> e320f000 nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-unicode.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-unicode.txt
new file mode 100644
index 00000000000000..62194b4b33c172
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-arm-unicode.txt
@@ -0,0 +1,27 @@
+
+<stdin>: file format elf32-littlearm
+
+Disassembly of section .text:
+
+00000000 <test_func>:
+ 0: ebfffffe bl 0x0 <test_func> @ imm = #-0x8
+ 4: eafffffe b 0x4 <test_func+0x4> @ imm = #-0x8
+ 8: ╭── eaffffff b 0xc <test_func+0xc> @ imm = #-0x4
+ c: ├─> e320f000 nop
+ 10: ╰── eafffffd b 0xc <test_func+0xc> @ imm = #-0xc
+ 14: eafffffe b 0x14 <test_func+0x14> @ imm = #-0x8
+ 18: ╭── 0affffff beq 0x1c <test_func+0x1c> @ imm = #-0x4
+ 1c: ╰─> e320f000 nop
+ 20: ╭────── ea000003 b 0x34 <test_func+0x34> @ imm = #0xc
+ 24: │ ╭──── ea000001 b 0x30 <test_func+0x30> @ imm = #0x4
+ 28: │ │ ╭── eaffffff b 0x2c <test_func+0x2c> @ imm = #-0x4
+ 2c: │ │ ╰─> e320f000 nop
+ 30: │ ╰───> e320f000 nop
+ 34: ╰─────> e320f000 nop
+ 38: ╭── ea000000 b 0x40 <test_func+0x40> @ imm = #0x0
+ 3c: ╭─│── ea000000 b 0x44 <test_func+0x44> @ imm = #0x0
+ 40: │ ╰─> e320f000 nop
+ 44: ╰───> e320f000 nop
+ 48: ╭──── eaffffff b 0x4c <test_func+0x4c> @ imm = #-0x4
+ 4c: ╰─│─> eaffffff b 0x50 <test_func+0x50> @ imm = #-0x4
+ 50: ╰─> e320f000 nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-ascii.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-ascii.txt
new file mode 100644
index 00000000000000..f18062167dd97d
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-ascii.txt
@@ -0,0 +1,34 @@
+
+<stdin>: file format elf32-littlearm
+
+Disassembly of section .text:
+
+00000000 <test_func>:
+ 0: f7ff fffe bl 0x0 <test_func> @ imm = #-0x4
+ 4: f7ff bffe b.w 0x4 <test_func+0x4> @ imm = #-0x4
+ 8: /-- e7ff b 0xa <test_func+0xa> @ imm = #-0x2
+ a: +-> bf00 nop
+ c: \-- e7fd b 0xa <test_func+0xa> @ imm = #-0x6
+ e: e7fe b 0xe <test_func+0xe> @ imm = #-0x4
+ 10: /-- f000 b807 b.w 0x22 <test_func+0x22> @ imm = #0xe
+ 14: +-- d005 beq 0x22 <test_func+0x22> @ imm = #0xa
+ 16: +-- f040 8004 bne.w 0x22 <test_func+0x22> @ imm = #0x8
+ 1a: +-- b110 cbz r0, 0x22 <test_func+0x22> @ imm = #0x4
+ 1c: | bfd8 it le
+ 1e: +-- e000 ble 0x22 <test_func+0x22> @ imm = #0x0
+ 20: | bf00 nop
+ 22: \-> bf00 nop
+ 24: bf00 nop
+ 26: /------ e003 b 0x30 <test_func+0x30> @ imm = #0x6
+ 28: | /---- e001 b 0x2e <test_func+0x2e> @ imm = #0x2
+ 2a: | | /-- e7ff b 0x2c <test_func+0x2c> @ imm = #-0x2
+ 2c: | | \-> bf00 nop
+ 2e: | \---> bf00 nop
+ 30: \-----> bf00 nop
+ 32: /-- e000 b 0x36 <test_func+0x36> @ imm = #0x0
+ 34: /-|-- e000 b 0x38 <test_func+0x38> @ imm = #0x0
+ 36: | \-> bf00 nop
+ 38: \---> bf00 nop
+ 3a: /-- e7ff b 0x3c <test_func+0x3c> @ imm = #-0x2
+ 3c: /-|-- e7ff b 0x3e <test_func+0x3e> @ imm = #-0x2
+ 3e: \---> bf00 nop
diff --git a/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-unicode.txt b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-unicode.txt
new file mode 100644
index 00000000000000..b171fb1c5d28c2
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/Inputs/visualize-jumps-thumb-unicode.txt
@@ -0,0 +1,34 @@
+
+<stdin>: file format elf32-littlearm
+
+Disassembly of section .text:
+
+00000000 <test_func>:
+ 0: f7ff fffe bl 0x0 <test_func> @ imm = #-0x4
+ 4: f7ff bffe b.w 0x4 <test_func+0x4> @ imm = #-0x4
+ 8: ╭── e7ff b 0xa <test_func+0xa> @ imm = #-0x2
+ a: ├─> bf00 nop
+ c: ╰── e7fd b 0xa <test_func+0xa> @ imm = #-0x6
+ e: e7fe b 0xe <test_func+0xe> @ imm = #-0x4
+ 10: ╭── f000 b807 b.w 0x22 <test_func+0x22> @ imm = #0xe
+ 14: ├── d005 beq 0x22 <test_func+0x22> @ imm = #0xa
+ 16: ├── f040 8004 bne.w 0x22 <test_func+0x22> @ imm = #0x8
+ 1a: ├── b110 cbz r0, 0x22 <test_func+0x22> @ imm = #0x4
+ 1c: │ bfd8 it le
+ 1e: ├── e000 ble 0x22 <test_func+0x22> @ imm = #0x0
+ 20: │ bf00 nop
+ 22: ╰─> bf00 nop
+ 24: bf00 nop
+ 26: ╭────── e003 b 0x30 <test_func+0x30> @ imm = #0x6
+ 28: │ ╭──── e001 b 0x2e <test_func+0x2e> @ imm = #0x2
+ 2a: │ │ ╭── e7ff b 0x2c <test_func+0x2c> @ imm = #-0x2
+ 2c: │ │ ╰─> bf00 nop
+ 2e: │ ╰───> bf00 nop
+ 30: ╰─────> bf00 nop
+ 32: ╭── e000 b 0x36 <test_func+0x36> @ imm = #0x0
+ 34: ╭─│── e000 b 0x38 <test_func+0x38> @ imm = #0x0
+ 36: │ ╰─> bf00 nop
+ 38: ╰───> bf00 nop
+ 3a: ╭── e7ff b 0x3c <test_func+0x3c> @ imm = #-0x2
+ 3c: ╭─│── e7ff b 0x3e <test_func+0x3e> @ imm = #-0x2
+ 3e: ╰───> bf00 nop
diff --git a/llvm/test/tools/llvm-objdump/visualize-jumps-aarch64.s b/llvm/test/tools/llvm-objdump/visualize-jumps-aarch64.s
new file mode 100644
index 00000000000000..f78b3e27975556
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/visualize-jumps-aarch64.s
@@ -0,0 +1,69 @@
+// RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \
+// RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=unicode - | \
+// RUN: diff - %p/Inputs/visualize-jumps-aarch64-unicode.txt
+
+// RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \
+// RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=ascii - | \
+// RUN: diff - %p/Inputs/visualize-jumps-aarch64-ascii.txt
+
+// RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \
+// RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=unicode,color - | \
+// RUN: diff - %p/Inputs/visualize-jumps-aarch64-unicode-color.txt
+
+// RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \
+// RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=unicode --reloc - | \
+// RUN: diff - %p/Inputs/visualize-jumps-aarch64-unicode-relocs.txt
+
+test_func:
+ // Relocated instructions don't get control-flow edges.
+ bl extern_func
+ b extern_func
+
+ // Two branches to the same label, one forward and one backward.
+ b .Llabel1
+.Llabel1:
+ nop
+ b .Llabel1
+
+ // Branch to self, no CFG edge shown
+ b .
+
+ // Conditional branches
+ b.eq .Llabel2
+ cbz x0, .Llabel2
+.Llabel2:
+ nop
+
+ // Branches are sorted with shorter ones to the right, to reduce number of
+ // crossings, and keep the lines for short branches short themselves.
+ b .Llabel5
+ b .Llabel4
+ b .Llabel3
+.Llabel3:
+ nop
+.Llabel4:
+ nop
+.Llabel5:
+ nop
+
+ // Sometimes crossings can't be avoided.
+ b .Llabel6
+ b .Llabel7
+.Llabel6:
+ nop
+.Llabel7:
+ nop
+
+ // TODO If a branch goes to another branch instruction, we don't have a way
+ // to represent that. Can we improve on this?
+ b .Llabel8
+.Llabel8:
+ b .Llabel9
+.Llabel9:
+ nop
+
+ // Graph lines need to be drawn on the same output line as relocations.
+ b .Llabel10
+ bl extern_func
+.Llabel10:
+ nop
diff --git a/llvm/test/tools/llvm-objdump/visualize-jumps-arm.s b/llvm/test/tools/llvm-objdump/visualize-jumps-arm.s
new file mode 100644
index 00000000000000..6855e6ff84e32a
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/visualize-jumps-arm.s
@@ -0,0 +1,54 @@
+// RUN: llvm-mc < %s -triple armv8a -filetype=obj | \
+// RUN: llvm-objdump --triple armv8a -d --visualize-jumps=unicode - | \
+// RUN: diff - %p/Inputs/visualize-jumps-arm-unicode.txt
+
+// RUN: llvm-mc < %s -triple armv8a -filetype=obj | \
+// RUN: llvm-objdump --triple armv8a -d --visualize-jumps=ascii - | \
+// RUN: diff - %p/Inputs/visualize-jumps-arm-ascii.txt
+
+test_func:
+ // Relocated instructions don't get control-flow edges.
+ bl extern_func
+ b extern_func
+
+ // Two branches to the same label, one forward and one backward.
+ b .Llabel1
+.Llabel1:
+ nop
+ b .Llabel1
+
+ // Branch to self, no CFG edge shown
+ b .
+
+ // Conditional branches
+ beq .Llabel2
+.Llabel2:
+ nop
+
+ // Branches are sorted with shorter ones to the right, to reduce number of
+ // crossings, and keep the lines for short branches short themselves.
+ b .Llabel5
+ b .Llabel4
+ b .Llabel3
+.Llabel3:
+ nop
+.Llabel4:
+ nop
+.Llabel5:
+ nop
+
+ // Sometimes crossings can't be avoided.
+ b .Llabel6
+ b .Llabel7
+.Llabel6:
+ nop
+.Llabel7:
+ nop
+
+ // TODO If a branch goes to another branch instruction, we don't have a way
+ // to represent that. Can we improve on this?
+ b .Llabel8
+.Llabel8:
+ b .Llabel9
+.Llabel9:
+ nop
diff --git a/llvm/test/tools/llvm-objdump/visualize-jumps-thumb.s b/llvm/test/tools/llvm-objdump/visualize-jumps-thumb.s
new file mode 100644
index 00000000000000..8be73d3aa4c946
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/visualize-jumps-thumb.s
@@ -0,0 +1,63 @@
+// RUN: llvm-mc < %s -triple thumbv8a -filetype=obj | \
+// RUN: llvm-objdump --triple thumbv8a -d --visualize-jumps=unicode - | \
+// RUN: diff - %p/Inputs/visualize-jumps-thumb-unicode.txt
+
+// RUN: llvm-mc < %s -triple thumbv8a -filetype=obj | \
+// RUN: llvm-objdump --triple thumbv8a -d --visualize-jumps=ascii - | \
+// RUN: diff - %p/Inputs/visualize-jumps-thumb-ascii.txt
+
+test_func:
+ // Relocated instructions don't get control-flow edges.
+ bl extern_func
+ b extern_func
+
+ // Two branches to the same label, one forward and one backward.
+ b .Llabel1
+.Llabel1:
+ nop
+ b .Llabel1
+
+ // Branch to self, no CFG edge shown
+ b .
+
+ // Different branch instructions
+ b.w .Llabel2
+ beq .Llabel2
+ bne.w .Llabel2
+ cbz r0, .Llabel2
+ it le
+ ble .Llabel2
+ nop
+.Llabel2:
+ nop
+.Llabel2.1:
+ nop
+
+ // Branches are sorted with shorter ones to the right, to reduce number of
+ // crossings, and keep the lines for short branches short themselves.
+ b .Llabel5
+ b .Llabel4
+ b .Llabel3
+.Llabel3:
+ nop
+.Llabel4:
+ nop
+.Llabel5:
+ nop
+
+ // Sometimes crossings can't be avoided.
+ b ....
[truncated]
|
|
|
omg yassss!!!! |
| T->Column = Column; | ||
| MaxColumn = std::max(MaxColumn, Column); | ||
|
|
||
| #if 1 |
|
|
||
| IndentToColumn(STI, OS, DisassemblyColumn::ControlFlow); | ||
|
|
||
| // TODO: What happens if an instruction has both incoming and outgoing edges? |
There was a problem hiding this comment.
so, a branch to a branch? isn't that what the Tee is used for?
There was a problem hiding this comment.
The tee is used when there are multiple branches to the same destination, which get one ControlFlowTarget object, and one line which is visually joined together. The case this doesn't handle correctly at the moment is where the destination of a branch is itself a branch instruction, so one disassembly line needs both an outgoing line and an incoming line, which should be unconnected and in different colors. I haven't yet though of a good way to represent that with either ASCII or box-drawing characters. There's an example of the current output for this in the test files.
Comment on lines
568
to
569
| bool IsASCII = | ||
| (OutputMode & VisualizeJumpsMode::CharsMask) == VisualizeJumpsMode::CharsASCII; |
There was a problem hiding this comment.
nearly every reference to OutputMode is verbose, and a few of the lambdas it's captured in look redundant. Consider making helper methods for these comparisons. I suspect you can avoid the lambdas with such methods.
There was a problem hiding this comment.
Done, I've made this a struct of two enums instead of a single bit-packed enum.
Comment on lines
243
to
250
| raw_ostream::Colors PickColor() { | ||
| if ((OutputMode & VisualizeJumpsMode::ColorMask) == | ||
| VisualizeJumpsMode::Off) | ||
| return raw_ostream::RESET; | ||
| auto Ret = LineColors[NextColorIdx]; | ||
| NextColorIdx = (NextColorIdx + 1) % (sizeof(LineColors) / sizeof(LineColors[0])); | ||
| return Ret; | ||
| } |
There was a problem hiding this comment.
consider moving this definition and a few of the larger definitions above to the .cpp file.
There was a problem hiding this comment.
at that point, you might not even need to declare LineColors in this header.
Comment on lines
517
to
519
| if (Col == DisassemblyColumn::Assembly) { | ||
| return Indent; | ||
| } |
There was a problem hiding this comment.
remove braces for single statement ifs.
Comment on lines
1379
to
1380
| std::vector<RelocationRef>::const_iterator RelCur = Relocs.begin(); | ||
| std::vector<RelocationRef>::const_iterator RelEnd = Relocs.end(); |
There was a problem hiding this comment.
Feel free to use const auto here.
Comment on lines
3400
to
3401
| VisualizeJumps = (VisualizeJumpsMode)(VisualizeJumpsMode::CharsUnicode | | ||
| VisualizeJumpsMode::ColorAuto); |
jh7370
reviewed
Dec 11, 2023
| @@ -0,0 +1,69 @@ | |||
| // RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \ | |||
| // RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=unicode - | \ | |||
| // RUN: diff - %p/Inputs/visualize-jumps-aarch64-unicode.txt | |||
There was a problem hiding this comment.
Rather than have the expected files in a separate location, please include them within this file. You can use split-file to divide a test file into multiple components (e.g. assembly and expected output).
That being said, a better method of testing would be to use FileCheck, with the expected input being the check patterns used by FileCheck.
There was a problem hiding this comment.
I used FileCheck for testing similar output with --debug-vars and was never very happy with that, because the disassembly contains tabs, so the output ends up getting misaligned in the text editor. I did it this way so that column 0 of the expected output is at column 0 of the test file, so the test looks exactly like the output.
Thanks for pointing out split-file, I didn't know about that. That looks like the best solution to this, so I'll switch the tests over to using it.
There was a problem hiding this comment.
because the disassembly contains tabs, so the output ends up getting misaligned in the text editor. I did it this way so that column 0 of the expected output is at column 0 of the test file, so the test looks exactly like the output.
Assuming you have --strict-whitespace and --match-full-lines on, I'm not sure I understand where this can go wrong, if you're careful with how you line up the colon following the check prefixes? I have a soft preference for FileCheck, as it has better diagnostics when things don't match, and it's easier to omit stuff irrelevant to what is under test.
This moves all of the logic for deciding the widths of different parts of the disassembly and indenting to them into one place. This will make it easier to add new columns. It also simplifies some things, because we were already using formatted_raw_ostream almost everywhere, which tracks the current column number for us.
This is a feature which GNU objdump currently has, which prints a control-flow graph showing all of the direct branches alongside the disassembly. This should work for any architecture which implements the MCInstrAnalysis class, I've tested it on ARM, Thumb and AArch64. There are a few (known) differences between this and the GNU objdump version: * This has the option to draw the lines using unicode line-drawing charcters, instead of ASCII. This is on by default because I find the connected lines much easier to read. * I haven't yet implemented the extended-color mode yet. With GNU objdump this often results in some very similar colors, so we might want to carefully pick a pallete which minimises that.
8edb1fd to
cb709ed
Compare
I gave this a quick shot for RISCV and it didn't work, though RISCV does subclass MCInstrAnalysis. Is there more to this than just implementing that class? Can you update the description to better document what else might be missing for other architectures? I also tested this on some linux kernel code; test arm64, arm (32b), and x86 all look great. Thumb2 code wasn't working. Can I send you an object file? |
jh7370
reviewed
Dec 14, 2023
| @@ -291,6 +291,36 @@ OPTIONS | |||
|
|
|||
| Target triple to disassemble for, see ``--version`` for available targets. | |||
|
|
|||
| .. option:: --visualize-jumps=<modes> | |||
|
|
|||
| Display a control-flow graph which shows the targets of branch instructions to the left of disasembly. ``modes`` is a comma-separated list of options, which configure the character set and used to print the graph. | |||
There was a problem hiding this comment.
Probably should be wrapped at a sensible width.
|
|
||
| .. option:: off | ||
|
|
||
| Disable control-flow graph |
There was a problem hiding this comment.
Nit: all of these should end with "."
|
|
||
| If ``modes`` is omitted, the default is ``unicode,auto`` | ||
|
|
||
| .. option:: off |
There was a problem hiding this comment.
Have you built and inspected the documentation? I don't have a currently working doc build, so I can't myself, but I'd be slightly concerned that this would look like a general command, rather than an argument of the --visualize-jumps option. --disassembler-color uses a different format. Perhaps that should be what you follow for consistency?
Oh, hang on, looking downwards, I see the --x86-asm-syntax=<style> is formatted this way. I personally prefer the bullet points of --disassembler-color (plus it's used in another place too), so I still think you should change.
| @@ -45,6 +45,7 @@ void formatted_raw_ostream::UpdatePosition(const char *Ptr, size_t Size) { | |||
| break; | |||
| case '\t': | |||
| // Assumes tab stop = 8 characters. | |||
| Column += 1; | |||
There was a problem hiding this comment.
I'm concerned this will have a wider impact on tools than just llvm-objdump. What's it actually doing?
| nop | ||
| b .Llabel1 | ||
|
|
||
| // Branch to self, no CFG edge shown |
There was a problem hiding this comment.
Nit: missing full stop, here and in other comments.
| // Check for relocations which apply to this instruction. | ||
| bool Relocated = false; | ||
| while (RelCur != RelEnd) { | ||
| // FIXME RelAdjustment for executables & shared objects |
There was a problem hiding this comment.
Similar comment to the earlier TODOs/FIXME: will this be actioned in a follow-up PR?
| @@ -2209,7 +2320,7 @@ disassembleObject(ObjectFile &Obj, const ObjectFile &DbgObj, | |||
| printBTFRelocation(FOS, *BTF, {Index, Section.getIndex()}, LVP); | |||
|
|
|||
| // Hexagon does this in pretty printer | |||
| if (Obj.getArch() != Triple::hexagon) { | |||
| if (Obj.getArch() != Triple::hexagon && InlineRelocs) { | |||
| } else if (Part == "unicode") { | ||
| Chars = VisualizeJumpsMode::Unicode; | ||
| } else { | ||
| reportCmdLineError("'" + Part + "' is not a valid value for '" + |
Comment on lines
+156
to
+158
| unsigned GetColumnIndent(MCSubtargetInfo const &STI, DisassemblyColumn Col); | ||
| void IndentToColumn(MCSubtargetInfo const &STI, formatted_raw_ostream &OS, | ||
| DisassemblyColumn Col); |
There was a problem hiding this comment.
Suggested change
| unsigned GetColumnIndent(MCSubtargetInfo const &STI, DisassemblyColumn Col); | |
| void IndentToColumn(MCSubtargetInfo const &STI, formatted_raw_ostream &OS, | |
| DisassemblyColumn Col); | |
| unsigned getColumnIndent(MCSubtargetInfo const &STI, DisassemblyColumn Col); | |
| void indentToColumn(MCSubtargetInfo const &STI, formatted_raw_ostream &OS, | |
| DisassemblyColumn Col); |
| @@ -0,0 +1,69 @@ | |||
| // RUN: llvm-mc < %s -triple aarch64 -filetype=obj | \ | |||
| // RUN: llvm-objdump --triple aarch64 -d --visualize-jumps=unicode - | \ | |||
| // RUN: diff - %p/Inputs/visualize-jumps-aarch64-unicode.txt | |||
There was a problem hiding this comment.
because the disassembly contains tabs, so the output ends up getting misaligned in the text editor. I did it this way so that column 0 of the expected output is at column 0 of the test file, so the test looks exactly like the output.
Assuming you have --strict-whitespace and --match-full-lines on, I'm not sure I understand where this can go wrong, if you're careful with how you line up the colon following the check prefixes? I have a soft preference for FileCheck, as it has better diagnostics when things don't match, and it's easier to omit stuff irrelevant to what is under test.
I think that's the only target-specific code I've relied on. This is all in llvm-objdump.cpp:collectLocalBranchTargets, which is shared with the
One of the test files is Thumb2, but I'll look at your object file and see if there's something which prevents this working. |
jthackray
approved these changes
Jan 3, 2024
| case LineChar::Tee: | ||
| return IsASCII ? "+" : u8"\u251c"; | ||
| case LineChar::Arrow: | ||
| return ">"; |
There was a problem hiding this comment.
Or u8"\u2192" (aka Unicode "RIGHTWARDS ARROW")?
| llvm_unreachable("Unhandled LineChar enum"); | ||
| } | ||
|
|
||
| #define C(id) getLineChar(LineChar::id) |
There was a problem hiding this comment.
Nit: would prefer not to have a #define if possible
|
Hi @ostannard , |
MaskRay
reviewed
Jul 15, 2024
| @@ -0,0 +1,204 @@ | |||
| # RUN: rm -rf %t && split-file %s %t && cd %t | |||
|
|
|||
| // RUN: llvm-mc < input.s -triple aarch64 -filetype=obj | \ | |||
There was a problem hiding this comment.
Omit < for llvm-mc
REQUIRES: aarch64-registered-target is needed unless you place it in ELF/AArch64
| MetaVarName<"mode,...">, | ||
| HelpText<"Print a control flow graph along side disassembly. " | ||
| "Color modes: auto (default), nocolor, color. " | ||
| "Character modes: unicode (default), ascii.">; |
Quite a few comments from @jh7370 and @jthackray need to be addressed first. I just left some comments, too. |
MaskRay
reviewed
Jul 15, 2024
|
|
||
| Display a control-flow graph which shows the targets of branch instructions to the left of disasembly. ``modes`` is a comma-separated list of options, which configure the character set and used to print the graph. | ||
|
|
||
| If ``modes`` is omitted, the default is ``unicode,auto`` |
There was a problem hiding this comment.
Unlike .td help messages, the documentation descriptions should have a trailing period.
|
@ostannard this seems very useful. I'd love to see it get merged. |
+1 in 2025:) This PR needs a rebase. A bunch of @jh7370 's comments need to be addressed |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is a feature which GNU objdump currently has, which prints a control-flow graph showing all of the direct branches alongside the disassembly.
This should work for any architecture which implements the MCInstrAnalysis class, I've tested it on ARM, Thumb and AArch64.
There are a few (known) differences between this and the GNU objdump
version:
Related issue: #60172
Demo: