address comments

Signed-off-by: leslie-fang25 <[email protected]>

Author: leslie-fang25

cpp/tensorrt_llm/kernels/quantization.cu

@@ -147,10 +147,8 @@ void invokeFP4Quantization(int b, int m, int n, T const* input, float const* SFS
         // The number of blocks for m. The m dimension will be padded to 128 for swizzled layout.
         int numBlocksForM = layout == QuantizationSFLayout::SWIZZLED ? PadUpFn(m, 128) : m;
         int gridSize = std::min(numBlocksForM, multiProcessorCount * numBlocksPerSM);
-        if (gridSize == 0)
-        {
-            gridSize = 1;
-        }
+        // Ensure gridSize is not zero.
+        gridSize = std::max(1, gridSize);
         dim3 grid(gridSize);
 
         // Launch the cvt kernel.
@@ -171,10 +169,8 @@ void invokeFP4Quantization(int b, int m, int n, T const* input, float const* SFS
         // The number of blocks for m. The m dimension will be padded to 128 for swizzled layout.
         int numBlocksForM = layout == QuantizationSFLayout::SWIZZLED ? PadUpFn(m, 128) : m;
         int gridSize = std::min(numBlocksForM, multiProcessorCount * numBlocksPerSM);
-        if (gridSize == 0)
-        {
-            gridSize = 1;
-        }
+        // Ensure gridSize is not zero.
+        gridSize = std::max(1, gridSize);
         dim3 grid(gridSize);
 
         // Launch the cvt kernel.