From 2eab1132062caecf8954d351cc2c45d4bec4a77f Mon Sep 17 00:00:00 2001 From: "Wen-Heng (Jack) Chung" Date: Mon, 17 Feb 2025 17:54:18 -0600 Subject: [PATCH] [ROCm] Add additional block quant GEMM tuning configs for AMD GPUs. (#3616) Co-authored-by: HAI --- ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 164 ++++++++++++++++++ 9 files changed, 1476 insertions(+) create mode 100644 python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json create mode 100644 python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json create mode 100644 python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json create mode 100644 python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json create mode 100644 python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json create mode 100644 python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json create mode 100644 python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json create mode 100644 python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json create mode 100644 python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json diff --git a/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..6f5adbb93 --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..6f5adbb93 --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..6f5adbb93 --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=24576,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..4225c78eb --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..4225c78eb --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..4225c78eb --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..dfe5c1e43 --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 8, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..dfe5c1e43 --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 8, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..dfe5c1e43 --- /dev/null +++ b/python/sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,164 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 8, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + } +}