[AMD] Register 7 JIT kernel unit tests for AMD nightly CI (#28967)

test/registered/jit/minimax/test_minimax_decode_topk.py

@@ -10,10 +10,11 @@ import pytest
 import torch
 
 from sglang.jit_kernel.minimax_decode_topk import minimax_decode_topk
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=40, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=40, suite="base-b-kernel-unit-1-gpu-b200")
+register_amd_ci(est_time=15, suite="nightly-amd-kernel-1-gpu", nightly=True)
 
 
 def _ref(score, seq_lens, block_size, topk):

test/registered/jit/minimax/test_minimax_store_kv_index.py

@@ -9,10 +9,11 @@ import pytest
 import torch
 
 from sglang.jit_kernel.minimax_store_kv_index import store_kv_index
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=30, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=30, suite="base-b-kernel-unit-1-gpu-b200")
+register_amd_ci(est_time=10, suite="nightly-amd-kernel-1-gpu", nightly=True)
 
 dev = "cuda"
 HEAD_DIM = 128

test/registered/jit/test_deepseek_v4_compress_state_runtime_shapes.py

@@ -98,10 +98,11 @@ from sglang.jit_kernel.tests.deepseek_v4.common import (
     make_legacy_context,
     to_seq_extend,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.utils import is_in_ci
 
 register_cuda_ci(est_time=30, suite="base-b-kernel-unit-1-gpu-large")
+register_amd_ci(est_time=25, suite="nightly-amd-kernel-1-gpu", nightly=True)
 
 Mode = Literal["decode", "prefill"]
 ShapeTier = Literal["ci", "smoke", "full"]

test/registered/jit/test_fused_store_index_cache.py

@@ -22,7 +22,7 @@ from typing import Optional, Tuple
 import pytest
 import torch
 
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 try:
     from sglang.jit_kernel.fused_store_index_cache import (
@@ -50,6 +50,7 @@ except ImportError:
 
 register_cuda_ci(est_time=24, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)
+register_amd_ci(est_time=24, suite="nightly-amd-kernel-1-gpu", nightly=True)
 
 PAGE_SIZE = 64
 HEAD_DIM = 128

test/registered/jit/test_fused_verify_triton_gdn.py

@@ -11,7 +11,7 @@ import sys
 import pytest
 import torch
 
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 try:
     from sglang.srt.layers.attention.fla.fused_gdn_gating import fused_gdn_gating
@@ -28,6 +28,7 @@ except ImportError:
 
 register_cuda_ci(est_time=6, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)
+register_amd_ci(est_time=10, suite="nightly-amd-kernel-1-gpu", nightly=True)
 
 
 def _make_tensors(N, T, H, HV, K, V, device="cuda", seed=2025):

test/registered/jit/test_kvcacheio_asymmetric.py

@@ -5,9 +5,10 @@ import pytest
 import torch
 
 from sglang.srt.mem_cache.memory_pool_host import AsymmetricMHATokenToKVPoolHost
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=10, suite="base-b-kernel-unit-1-gpu-large")
+register_amd_ci(est_time=10, suite="nightly-amd-kernel-1-gpu", nightly=True)
 
 # These tests use AsymmetricMHATokenToKVPoolHost methods and let that class call
 # the real sgl-kernel transfer ops.

test/registered/jit/test_mla_kv_pack_quantize_fp8.py

@@ -5,9 +5,10 @@ import torch
 
 from sglang.jit_kernel.mla_kv_pack_quantize_fp8 import mla_kv_pack_quantize_fp8
 from sglang.jit_kernel.utils import get_ci_test_range
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=60, suite="base-b-kernel-unit-1-gpu-large")
+register_amd_ci(est_time=60, suite="nightly-amd-kernel-1-gpu", nightly=True)
 
 DEVICE = "cuda"