iaca
This package helps one use Intel's IACA tool with go programs. It does this by processing the output of go tool objdump. To demonstate, first build an example binary (one exists at github.com/zeebo/iaca/example) which just contains
package main
var x [4]uint64
func main() {
x[0]++
x[1]++
x[2]++
x[3]++
x[0]++
x[1]++
x[2]++
x[3]++
}One can use the objdump tool to inspect the generated code.
$ go tool objdump -s main.main example
TEXT main.main(SB) /home/jeff/go/src/github.com/zeebo/iaca/example/main.go
main.go:6 0x44f930 488b0529e50800 MOVQ main.x(SB), AX
main.go:6 0x44f937 488d4801 LEAQ 0x1(AX), CX
main.go:6 0x44f93b 48890d1ee50800 MOVQ CX, main.x(SB)
main.go:7 0x44f942 488b0d1fe50800 MOVQ main.x+8(SB), CX
main.go:7 0x44f949 488d5101 LEAQ 0x1(CX), DX
main.go:7 0x44f94d 48891514e50800 MOVQ DX, main.x+8(SB)
main.go:8 0x44f954 488b1515e50800 MOVQ main.x+16(SB), DX
main.go:8 0x44f95b 488d5a01 LEAQ 0x1(DX), BX
main.go:8 0x44f95f 48891d0ae50800 MOVQ BX, main.x+16(SB)
main.go:9 0x44f966 488b1d0be50800 MOVQ main.x+24(SB), BX
main.go:10 0x44f96d 4883c002 ADDQ $0x2, AX
main.go:10 0x44f971 488905e8e40800 MOVQ AX, main.x(SB)
main.go:11 0x44f978 488d4102 LEAQ 0x2(CX), AX
main.go:11 0x44f97c 488905e5e40800 MOVQ AX, main.x+8(SB)
main.go:12 0x44f983 488d4202 LEAQ 0x2(DX), AX
main.go:12 0x44f987 488905e2e40800 MOVQ AX, main.x+16(SB)
main.go:13 0x44f98e 488d4302 LEAQ 0x2(BX), AX
main.go:13 0x44f992 488905dfe40800 MOVQ AX, main.x+24(SB)
main.go:14 0x44f999 c3 RET
Running the output of this package over that produces the binary data with the iaca markers inserted.
$ go tool objdump -s main.main a.out | go run github.com/zeebo/iaca | xxd
00000000: bb6f 0000 0064 6790 488b 0529 e508 0048 .o...dg.H..)...H
00000010: 8d48 0148 890d 1ee5 0800 488b 0d1f e508 .H.H......H.....
00000020: 0048 8d51 0148 8915 14e5 0800 488b 1515 .H.Q.H......H...
00000030: e508 0048 8d5a 0148 891d 0ae5 0800 488b ...H.Z.H......H.
00000040: 1d0b e508 0048 83c0 0248 8905 e8e4 0800 .....H...H......
00000050: 488d 4102 4889 05e5 e408 0048 8d42 0248 H.A.H......H.B.H
00000060: 8905 e2e4 0800 488d 4302 4889 05df e408 ......H.C.H.....
00000070: 00c3 bbde 0000 0064 6790 .......dg.
The starting marker is the byte sequence bb6f000000646790 and the ending marker is the byte sequence bbde000000646790, and in between is the same byte sequence as what was produced by go tool objdump. Then, Intel's iaca tool works like expected.
$ iaca <( go tool objdump -s main.main a.out | go run github.com/zeebo/iaca )
Intel(R) Architecture Code Analyzer Version - v3.0-28-g1ba2cbb build date: 2017-10-23;16:42:45
Analyzed File - /dev/fd/63
Binary Format - 64Bit
Architecture - SKL
Analysis Type - Throughput
Throughput Analysis Report
--------------------------
Block Throughput: 7.00 Cycles Throughput Bottleneck: Backend
Loop Count: 22
Port Binding In Cycles Per Iteration:
--------------------------------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
--------------------------------------------------------------------------------------------------
| Cycles | 0.5 0.0 | 3.0 | 4.0 3.0 | 4.0 2.0 | 7.0 | 3.0 | 0.5 | 4.0 |
--------------------------------------------------------------------------------------------------
DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3)
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion occurred
# - ESP Tracking sync uop was issued
@ - SSE instruction followed an AVX256/AVX512 instruction, dozens of cycles penalty is expected
X - instruction not supported, was not accounted in Analysis
| Num Of | Ports pressure in cycles | |
| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
-----------------------------------------------------------------------------------------
| 1 | | | | 1.0 1.0 | | | | | mov rax, qword ptr [rip+0x8e529]
| 1 | | 1.0 | | | | | | | lea rcx, ptr [rax+0x1]
| 2^ | | | | | 1.0 | | | 1.0 | mov qword ptr [rip+0x8e51e], rcx
| 1 | | | 1.0 1.0 | | | | | | mov rcx, qword ptr [rip+0x8e51f]
| 1 | | | | | | 1.0 | | | lea rdx, ptr [rcx+0x1]
| 2^ | | | | | 1.0 | | | 1.0 | mov qword ptr [rip+0x8e514], rdx
| 1 | | | | 1.0 1.0 | | | | | mov rdx, qword ptr [rip+0x8e515]
| 1 | | 1.0 | | | | | | | lea rbx, ptr [rdx+0x1]
| 2^ | | | | | 1.0 | | | 1.0 | mov qword ptr [rip+0x8e50a], rbx
| 1 | | | 1.0 1.0 | | | | | | mov rbx, qword ptr [rip+0x8e50b]
| 1 | 0.5 | | | | | | 0.5 | | add rax, 0x2
| 2^ | | | | 1.0 | 1.0 | | | | mov qword ptr [rip+0x8e4e8], rax
| 1 | | | | | | 1.0 | | | lea rax, ptr [rcx+0x2]
| 2^ | | | | | 1.0 | | | 1.0 | mov qword ptr [rip+0x8e4e5], rax
| 1 | | 1.0 | | | | | | | lea rax, ptr [rdx+0x2]
| 2^ | | | 1.0 | | 1.0 | | | | mov qword ptr [rip+0x8e4e2], rax
| 1 | | | | | | 1.0 | | | lea rax, ptr [rbx+0x2]
| 2^ | | | | 1.0 | 1.0 | | | | mov qword ptr [rip+0x8e4df], rax
| 3^ | | | 1.0 1.0 | | | | | | ret
Total Num Of Uops: 28
Analysis Notes:
Backend allocation was stalled due to unavailable allocation resources.
If you just want to check some hot kernel inside of some function, just trim the go tool objdump output to the portion you care about.