Feat: Babybear GPU prover

[Velaciela]

related PR: #987

Performance

• fibonacci $\text{fib}(2^{20})$: runs 12M steps (warmup + rerun)
• machine(node-19): AMD EPYC 7702 64-Core, 1 x RTX 3090

metric	CPU(bb31)	GPU(bb31)	speedup	-	CPU(gl64)	GPU(gl64)	speedup
Create Proof(total)	12.6s	4.18s	3.01x	-	13.9s	4.87s	2.85x
batch commit	2.31s	514ms	4.49x	-	6.81s	1.20s	5.68x
transfer pk	-	5.17ms	N/A	-	-	8.20ms	N/A
main and tower proof	9.56s	3.48s	2.74x	-	6.11s	3.45s	1.77x
opening	773ms	179ms	4.31x	-	1.01s	214ms	4.71x

trace: Babybear

### CPU-BB31 ### ZKVM_create_proof [ 12.6s | 0.00% / 100.00% ]
┝━ batch commit to traces [ 2.31s | 18.29% ] profiling_1: true
┝━ transfer pk to device [ 2.99µs | 0.00% ] profiling_1: true
┝━ main_proofs [ 9.56s | 0.01% / 75.60% ] profiling_1: true
│  ┝━ create_chip_proof [ 2.03s | 0.35% / 16.04% ] table_name: "ADD"
│  │  ┝━ per_layer_gen_witness [ 111ms | 0.87% ] profiling_2: true
│  │  ┝━ prove_tower_relation [ 1.65s | 0.00% / 13.04% ] profiling_2: true
│  │  │  ┝━ build_tower_witness [ 398ms | 3.15% ] profiling_2: true
│  │  │  ┕━ prove_tower_relation [ 1.25s | 9.89% ] profiling_2: true
│  │  ┕━ layer_proof [ 224ms | 1.77% ] profiling_2: true
...
┕━ pcs_opening [ 773ms | 6.11% ] profiling_1: true


### GPU-BB31 ### ZKVM_create_proof [ 4.18s | 0.00% / 100.00% ]
┝━ batch commit to traces [ 514ms | 0.45% / 12.28% ] profiling_1: true
│  ┝━ [gpu] rmms_h2d  = 231.46ms
│  ┝━ [gpu] polys_d2d = 10.37ms
│  ┝━ [gpu] encode    = 42.93ms
│  ┕━ [gpu] mmcs      = 92.70ms
┝━ main_proofs [ 3.48s | 7.31% / 83.30% ] profiling_1: true
│  ┝━ create_chip_proof [ 677ms | 0.00% / 16.19% ] table_name: "ADD"
│  │  ┝━ per_layer_gen_witness [ 5.26ms | 0.13% ] profiling_2: true
│  │  ┝━ prove_tower_relation [ 357ms | 0.00% / 8.54% ] profiling_2: true
│  │  │  ┝━ build_tower_witness [ 62.3ms | 1.49% ] profiling_2: true
│  │  │  ┝━ extract_out_evals_from_gpu_towers [ 3.99ms | 0.10% ] profiling_2: true
│  │  │  ┕━ prove_tower_relation [ 291ms | 6.95% ] profiling_2: true
│  │  ┕━ layer_proof [ 315ms | 7.52% ] profiling_2: true
...
┕━ pcs_opening [ 179ms | 4.29% ] profiling_1: true

trace: Goldilocks

### CPU-GL64 ### ZKVM_create_proof [ 13.9s | 0.00% / 100.00% ]
┝━ batch commit to traces [ 6.81s | 48.85% ] profiling_1: true
┝━ main_proofs [ 6.11s | 0.41% / 43.87% ] profiling_1: true
│  ┝━ create_chip_proof [ 1.30s | 0.50% / 9.35% ] table_name: "ADD"
│  │  ┝━ per_layer_gen_witness [ 107ms | 0.77% ] profiling_2: true
│  │  ┝━ prove_tower_relation [ 945ms | 0.00% / 6.78% ] profiling_2: true
│  │  │  ┝━ build_tower_witness [ 334ms | 2.40% ] profiling_2: true
│  │  │  ┕━ prove_tower_relation [ 610ms | 4.38% ] profiling_2: true
│  │  ┕━ layer_proof [ 180ms | 1.29% ] profiling_2: true
...
┕━ pcs_opening [ 1.01s | 7.28% ] profiling_1: true

### GPU-GL64 ### ZKVM_create_proof [ 4.87s | 0.00% / 100.00% ]
┝━ batch commit to traces [ 1.20s | 0.90% / 24.58% ] profiling_1: true
│  ┝━ [gpu] rmms_h2d  = 420.56ms
│  ┝━ [gpu] polys_d2d = 15.84ms
│  ┝━ [gpu] encode    = 84.63ms
│  ┕━ [gpu] mmcs      = 431.68ms
┝━ transfer pk to device [ 8.20ms | 0.17% ] profiling_1: true
┝━ main_proofs [ 3.45s | 7.33% / 70.86% ] profiling_1: true
│  ┝━ create_chip_proof [ 657ms | 0.00% / 13.49% ] table_name: "ADD"
│  │  ┝━ per_layer_gen_witness [ 6.87ms | 0.14% ] profiling_2: true
│  │  ┝━ prove_tower_relation [ 346ms | 0.00% / 7.11% ] profiling_2: true
│  │  │  ┝━ build_tower_witness [ 61.4ms | 1.26% ] profiling_2: true
│  │  │  ┝━ extract_out_evals_from_gpu_towers [ 3.16ms | 0.06% ] profiling_2: true
│  │  │  ┕━ prove_tower_relation [ 282ms | 5.78% ] profiling_2: true
│  │  ┕━ layer_proof [ 304ms | 6.24% ] profiling_2: true
...
┕━ pcs_opening [ 214ms | 4.39% ] profiling_1: true

e2e test

# fibonacci
RUST_LOG=debug cargo run --release --features gpu --package ceno_zkvm --bin e2e -- --platform=ceno --hints=15 --public-io=2400 --profiling=2 --field=baby-bear examples/target/riscv32im-ceno-zkvm-elf/release/examples/fibonacci 2>&1 | tee "fib_perf_gpu.log"

# keccak
RUST_LOG=debug cargo run --release --features gpu --package ceno_zkvm --bin e2e -- --platform=ceno --profiling=2 --field=baby-bear examples/target/riscv32im-ceno-zkvm-elf/release/examples/keccak_syscall 2>&1 | tee "keccak_perf_gpu.log"