compiler-optimizations-deep

Name: compiler-optimizations-deep
Author: mohitmishra786

61installs19forks

How to install compiler-optimizations-deep?

mohitmishra786/low-level-dev-skills/compiler-optimizations-deep

$npx -y skills add mohitmishra786/low-level-dev-skills --skill compiler-optimizations-deep

Installs into the current project.

Summary

Deep compiler optimizations skill for RA, ISel, and PGO. Use when explaining register allocation, instruction selection, LICM, vectorization limits, or profile-guided optimization beyond -O3. Activates on queries about register allocation, instruction selection, LICM, auto-vector

Files · 1

View on GitHub

SKILL.md

1	# Compiler Optimizations (Deep)
2
3	## Purpose
4
5	Explain optimization phases beyond flags: mid-level IR opts, register allocation, instruction selection/scheduling, vectorization boundaries, PGO, and post-link BOLT — bridging `skills/compilers/pgo` and LLVM/GCC internals.
6
7	## When to Use
8
9	- `-O3` did not vectorize a hot loop
10	- Teaching why register pressure causes spills
11	- Planning PGO or BOLT deployment
12	- Understanding pass interaction (e.g., LICM before vectorize)
13
14	## Workflow
15
16	### 1. Compiler pipeline map
17
18	```
19	Frontend → LLVM IR / GCC GIMPLE
20	├── Mid-level: DCE, GVN, LICM, inlining
21	├── Loop opts: unroll, vectorize
22	├── Codegen prep: legalize types
23	├── Instruction selection (DAG → machine ops)
24	├── Register allocation (greedy, linear scan)
25	└── Peephole / scheduling
26	```
27
28	### 2. Vectorization failure triage
29
30	```bash
31	clang -O3 -Rpass=loop-vectorize -Rpass-missed=loop-vectorize foo.c
32	```
33
34	\| Miss reason \| Typical fix \|
35	\|-------------\|-------------\|
36	\| Unknown trip count \| peel loop; assert count \|
37	\| Dependence \| reorder / separate accumulators \|
38	\| Function call in loop \| inline or outline \|
39	\| Alignment unknown \| `__builtin_assume_aligned` \|
40
41	### 3. Register allocation intuition
42
43	When live ranges exceed physical registers, the allocator spills to stack slots — costly loads/stores. Reducing live ranges (splitting variables, rematerialization) helps.
44
45	GCC/LLVM both use graph coloring variants (LLVM "greedy regalloc").
46
47	### 4. PGO workflow (Clang)
48
49	```bash
50	clang -fprofile-instr-generate -O2 -o app foo.c
51	./app # training workload
52	llvm-profdata merge default.profraw -o default.profdata
53	clang -fprofile-instr-use=default.profdata -O2 -o app_pgo foo.c
54	```
55
56	Improves branch layout, inlining, and vectorization thresholds.
57
58	See `skills/compilers/pgo` for GCC and BOLT.
59
60	### 5. BOLT (post-link)
61
62	```bash
63	llvm-bolt -instrument app -o app.inst
64	./app.inst
65	llvm-bolt -data=perf.fdata -reorder-blocks=+ -o app.bolt app
66	```
67
68	Optimizes layout after linker — needs relocations (`-Wl,--emit-relocs`).
69
70	### 6. LICM example
71
72	Loop-invariant code motion hoists `x * scale` out of inner loop when legal — reduces work per iteration.
73
74	### 7. Agent usage
75
76	```
77	/compiler-optimizations-deep Why did LLVM fail to vectorize this reduction loop?
78	```
79
80	## Common Problems
81
82	\| Symptom \| Cause \| Fix \|
83	\|---------\|-------\|-----\|
84	\| PGO no gain \| Unrepresentative training \| Match production input \|
85	\| BOLT crash \| Stripped binary \| Keep symbols + relocs \|
86	\| Spills in asm \| Register pressure \| Simplify live ranges \|
87	\| `-O3` slower \| Code bloat / cache \| Try `-O2` or PGO \|
88	\| Different GCC/Clang \| Pass ordering differs \| Compare IR + asm \|
89
90	## Related Skills
91
92	- `skills/compilers/pgo` — PGO and BOLT detail
93	- `skills/compiler-internals/llvm-ir-and-passes` — IR-level opts
94	- `skills/compiler-internals/code-generation-and-backends` — ISel and backends
95	- `skills/computer-architecture/cpu-pipelines-and-hazards` — scheduling context
96	- `skills/low-level-programming/simd-intrinsics` — manual vectorization

People who liked this also viewed

6 picks

Type

compiler-optimizations-deep

61installs19forks

View on GitHub

How to install compiler-optimizations-deep?

mohitmishra786/low-level-dev-skills/compiler-optimizations-deep

$npx -y skills add mohitmishra786/low-level-dev-skills --skill compiler-optimizations-deep

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Compiler Optimizations (Deep)
2
3	## Purpose
4
5	Explain optimization phases beyond flags: mid-level IR opts, register allocation, instruction selection/scheduling, vectorization boundaries, PGO, and post-link BOLT — bridging `skills/compilers/pgo` and LLVM/GCC internals.
6
7	## When to Use
8
9	- `-O3` did not vectorize a hot loop
10	- Teaching why register pressure causes spills
11	- Planning PGO or BOLT deployment
12	- Understanding pass interaction (e.g., LICM before vectorize)
13
14	## Workflow
15
16	### 1. Compiler pipeline map
17
18	```
19	Frontend → LLVM IR / GCC GIMPLE
20	├── Mid-level: DCE, GVN, LICM, inlining
21	├── Loop opts: unroll, vectorize
22	├── Codegen prep: legalize types
23	├── Instruction selection (DAG → machine ops)
24	├── Register allocation (greedy, linear scan)
25	└── Peephole / scheduling
26	```
27
28	### 2. Vectorization failure triage
29
30	```bash
31	clang -O3 -Rpass=loop-vectorize -Rpass-missed=loop-vectorize foo.c
32	```
33
34	\| Miss reason \| Typical fix \|
35	\|-------------\|-------------\|
36	\| Unknown trip count \| peel loop; assert count \|
37	\| Dependence \| reorder / separate accumulators \|
38	\| Function call in loop \| inline or outline \|
39	\| Alignment unknown \| `__builtin_assume_aligned` \|
40
41	### 3. Register allocation intuition
42
43	When live ranges exceed physical registers, the allocator spills to stack slots — costly loads/stores. Reducing live ranges (splitting variables, rematerialization) helps.
44
45	GCC/LLVM both use graph coloring variants (LLVM "greedy regalloc").
46
47	### 4. PGO workflow (Clang)
48
49	```bash
50	clang -fprofile-instr-generate -O2 -o app foo.c
51	./app # training workload
52	llvm-profdata merge default.profraw -o default.profdata
53	clang -fprofile-instr-use=default.profdata -O2 -o app_pgo foo.c
54	```
55
56	Improves branch layout, inlining, and vectorization thresholds.
57
58	See `skills/compilers/pgo` for GCC and BOLT.
59
60	### 5. BOLT (post-link)
61
62	```bash
63	llvm-bolt -instrument app -o app.inst
64	./app.inst
65	llvm-bolt -data=perf.fdata -reorder-blocks=+ -o app.bolt app
66	```
67
68	Optimizes layout after linker — needs relocations (`-Wl,--emit-relocs`).
69
70	### 6. LICM example
71
72	Loop-invariant code motion hoists `x * scale` out of inner loop when legal — reduces work per iteration.
73
74	### 7. Agent usage
75
76	```
77	/compiler-optimizations-deep Why did LLVM fail to vectorize this reduction loop?
78	```
79
80	## Common Problems
81
82	\| Symptom \| Cause \| Fix \|
83	\|---------\|-------\|-----\|
84	\| PGO no gain \| Unrepresentative training \| Match production input \|
85	\| BOLT crash \| Stripped binary \| Keep symbols + relocs \|
86	\| Spills in asm \| Register pressure \| Simplify live ranges \|
87	\| `-O3` slower \| Code bloat / cache \| Try `-O2` or PGO \|
88	\| Different GCC/Clang \| Pass ordering differs \| Compare IR + asm \|
89
90	## Related Skills
91
92	- `skills/compilers/pgo` — PGO and BOLT detail
93	- `skills/compiler-internals/llvm-ir-and-passes` — IR-level opts
94	- `skills/compiler-internals/code-generation-and-backends` — ISel and backends
95	- `skills/computer-architecture/cpu-pipelines-and-hazards` — scheduling context
96	- `skills/low-level-programming/simd-intrinsics` — manual vectorization

People who liked this also viewed

6 picks

Type