[Kernel][Helion] Disable warp specialization in rms_norm_per_block_quant B200 configs (#48797)

Signed-off-by: Shangdi Yu <shangdiy@meta.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Shangdi Yu committed Jul 17, 2026 at 14:39 UTC c9be3a8aa1f01f7efdc99dd9439451617ef54a00
1 file changed +45 -45
vllm/kernels/helion/configs/rms_norm_per_block_quant/nvidia_b200.json
+45 -45
@@ -655,7 +655,7 @@
655 "range_warp_specializes": [
656 null,
657 null,
658 - true,
658 + false,
659 null
660 ],
661 "range_num_stages": [],
@@ -1705,7 +1705,7 @@
1705 "range_warp_specializes": [
1706 null,
1707 false,
1708 - true,
1708 + false,
1709 null
1710 ],
1711 "range_num_stages": [],
@@ -1838,14 +1838,14 @@
1838 ],
1839 "range_unroll_factors": [
1840 0,
1841 - 4,
1841 + 2,
1842 0,
1843 1
1844 ],
1845 "range_warp_specializes": [
1846 null,
1847 null,
1848 - true,
1848 + false,
1849 null
1850 ],
1851 "range_num_stages": [],
@@ -1872,7 +1872,7 @@
1872 "last",
1873 ""
1874 ],
1875 - "num_warps": 2,
1875 + "num_warps": 4,
1876 "num_stages": 6,
1877 "indexing": [
1878 "tensor_descriptor",
@@ -1881,7 +1881,7 @@
1881 "tensor_descriptor",
1882 "tensor_descriptor",
1883 "tensor_descriptor",
1884 - "pointer",
1884 + "tensor_descriptor",
1885 "tensor_descriptor",
1886 "pointer"
1887 ],
@@ -2610,49 +2610,49 @@
2610 ],
2611 "range_unroll_factors": [
2612 0,
2613 - 3,
2613 + 0,
2614 0,
2615 0
2616 ],
2617 "range_warp_specializes": [
2618 null,
2619 + false,
2620 null,
2620 - true,
2621 null
2622 ],
2623 "range_num_stages": [],
2624 "range_multi_buffers": [
2625 null,
2626 - true,
2626 + null,
2627 null,
2628 null
2629 ],
2630 "range_flattens": [
2631 null,
2632 + false,
2633 null,
2633 - true,
2634 null
2635 ],
2636 "static_ranges": [
2637 true
2638 ],
2639 "load_eviction_policies": [
2640 - "last",
2640 "first",
2641 + "",
2642 + "",
2643 "first",
2644 "first",
2644 - "",
2645 "last"
2646 ],
2647 - "num_warps": 1,
2648 - "num_stages": 3,
2647 + "num_warps": 4,
2648 + "num_stages": 7,
2649 "indexing": [
2650 + "pointer",
2651 "tensor_descriptor",
2651 - "tensor_descriptor",
2652 - "tensor_descriptor",
2653 - "tensor_descriptor",
2652 + "pointer",
2653 "pointer",
2654 "tensor_descriptor",
2655 + "pointer",
2656 "tensor_descriptor",
2657 "tensor_descriptor",
2658 "tensor_descriptor"
@@ -2680,14 +2680,14 @@
2680 ],
2681 "range_unroll_factors": [
2682 0,
2683 - 4,
2683 + 0,
2684 0,
2685 0
2686 ],
2687 "range_warp_specializes": [
2688 - null,
2688 null,
2689 true,
2690 + null,
2691 null
2692 ],
2693 "range_num_stages": [],
@@ -2707,14 +2707,14 @@
2707 true
2708 ],
2709 "load_eviction_policies": [
2710 - "",
2710 + "last",
2711 "",
2712 "first",
2713 "last",
2714 "last",
2715 ""
2716 ],
2717 - "num_warps": 1,
2717 + "num_warps": 2,
2718 "num_stages": 3,
2719 "indexing": [
2720 "pointer",
@@ -2723,7 +2723,7 @@
2723 "tensor_descriptor",
2724 "tensor_descriptor",
2725 "pointer",
2726 - "tensor_descriptor",
2726 + "pointer",
2727 "pointer",
2728 "pointer"
2729 ],
@@ -2809,7 +2809,7 @@
2809 },
2810 "config": {
2811 "block_sizes": [
2812 - 8192,
2812 + 2048,
2813 8
2814 ],
2815 "loop_orders": [
@@ -2821,13 +2821,13 @@
2821 "range_unroll_factors": [
2822 0,
2823 3,
2824 - 0,
2825 - 2
2824 + 1,
2825 + 0
2826 ],
2827 "range_warp_specializes": [
2828 null,
2829 null,
2830 - true,
2830 + false,
2831 null
2832 ],
2833 "range_num_stages": [],
@@ -2841,10 +2841,10 @@
2841 null,
2842 null,
2843 false,
2844 - false
2844 + null
2845 ],
2846 "static_ranges": [
2847 - false
2847 + true
2848 ],
2849 "load_eviction_policies": [
2850 "",
@@ -2854,7 +2854,7 @@
2854 "",
2855 "last"
2856 ],
2857 - "num_warps": 1,
2857 + "num_warps": 2,
2858 "num_stages": 3,
2859 "indexing": [
2860 "pointer",
@@ -2884,20 +2884,20 @@
2884 ],
2885 "loop_orders": [
2886 [
2887 - 0,
2888 - 1
2887 + 1,
2888 + 0
2889 ]
2890 ],
2891 "range_unroll_factors": [
2892 0,
2893 - 3,
2893 + 1,
2894 0,
2895 - 2
2895 + 0
2896 ],
2897 "range_warp_specializes": [
2898 null,
2899 + false,
2900 null,
2900 - true,
2901 null
2902 ],
2903 "range_num_stages": [],
@@ -2908,37 +2908,37 @@
2908 null
2909 ],
2910 "range_flattens": [
2911 - null,
2911 null,
2912 false,
2914 - false
2913 + false,
2914 + null
2915 ],
2916 "static_ranges": [
2917 - false
2917 + true
2918 ],
2919 "load_eviction_policies": [
2920 + "last",
2921 "",
2922 "",
2923 "first",
2923 - "last",
2924 - "",
2924 + "first",
2925 "last"
2926 ],
2927 - "num_warps": 1,
2928 - "num_stages": 3,
2927 + "num_warps": 2,
2928 + "num_stages": 7,
2929 "indexing": [
2930 "pointer",
2931 "pointer",
2932 - "tensor_descriptor",
2933 - "tensor_descriptor",
2932 + "pointer",
2933 "pointer",
2934 "tensor_descriptor",
2935 + "pointer",
2936 "tensor_descriptor",
2937 "pointer",
2938 - "pointer"
2938 + "tensor_descriptor"
2939 ],
2940 "atomic_indexing": [],
2941 "pid_type": "flat"
2942 }
2943 }
2944 -]
\ No newline at end of file
2944 +]