chore: remove debug tracing from gatherQ8 and UploadWeights

Author: dndungu

compute/gpu_engine.go

@@ -569,16 +569,6 @@ func (e *GPUEngine[T]) UploadWeights(tensors []*tensor.TensorNumeric[float32]) e
 			"device", fmt.Sprintf("%d", e.deviceID),
 			"method", method)
 	}
-	// Check for sticky CUDA errors after UploadWeights.
-	if debugGPU || os.Getenv("UPLOAD_TRACE") == "1" {
-		if e.stream != nil {
-			if syncErr := e.stream.Synchronize(); syncErr != nil {
-				fmt.Fprintf(os.Stderr, "[UPLOAD] CUDA sync error after UploadWeights: %v\n", syncErr)
-			} else {
-				fmt.Fprintf(os.Stderr, "[UPLOAD] CUDA context clean after UploadWeights\n")
-			}
-		}
-	}
 	return nil
 }
 
@@ -3421,9 +3411,6 @@ func (e *GPUEngine[T]) gatherQ8(
 	devQ8 unsafe.Pointer,
 ) error {
 	e.setDevice()
-	if debugGPU || os.Getenv("UPLOAD_TRACE") == "1" {
-		fmt.Fprintf(os.Stderr, "[GATHER_Q8] called: V=%d D=%d devQ8=%p\n", params.Shape()[0], params.Shape()[1], devQ8)
-	}
 
 	pShape := params.Shape()
 	V := pShape[0]
@@ -3436,25 +3423,18 @@ func (e *GPUEngine[T]) gatherQ8(
 	}
 
 	// Upload indices as int32 to GPU.
-	trace := debugGPU || os.Getenv("UPLOAD_TRACE") == "1"
 	idx32 := make([]int32, N)
 	for i, id := range idxData {
 		idx32[i] = int32(id)
 	}
 	idxBytes := N * 4
 	devIdx, err := e.pool.Alloc(e.deviceID, idxBytes)
 	if err != nil {
-		if trace {
-			fmt.Fprintf(os.Stderr, "[GATHER_Q8] pool.Alloc(idx %d) FAILED: %v — CPU fallback\n", idxBytes, err)
-		}
 		return e.cpu.Gather(context.Background(), params, indices, output)
 	}
 	defer e.pool.Free(e.deviceID, devIdx, idxBytes)
 
 	if err := e.runtime.Memcpy(devIdx, unsafe.Pointer(&idx32[0]), idxBytes, gpuapi.MemcpyHostToDevice); err != nil {
-		if trace {
-			fmt.Fprintf(os.Stderr, "[GATHER_Q8] Memcpy(idx H2D %d bytes) FAILED: %v — CPU fallback\n", idxBytes, err)
-		}
 		return e.cpu.Gather(context.Background(), params, indices, output)
 	}
 
@@ -3463,23 +3443,14 @@ func (e *GPUEngine[T]) gatherQ8(
 	outBytes := outElems * f32Size
 	devOut, err := e.pool.Alloc(e.deviceID, outBytes)
 	if err != nil {
-		if trace {
-			fmt.Fprintf(os.Stderr, "[GATHER_Q8] pool.Alloc(out %d) FAILED: %v — CPU fallback\n", outBytes, err)
-		}
 		return e.cpu.Gather(context.Background(), params, indices, output)
 	}
 
 	// Launch Q8 gather kernel.
 	if err := e.kernels.GatherQ8F32(devQ8, devIdx, devOut, N, D, V, e.stream); err != nil {
-		if debugGPU || os.Getenv("UPLOAD_TRACE") == "1" {
-			fmt.Fprintf(os.Stderr, "[GATHER_Q8] kernel FAILED: %v — fallback to CPU\n", err)
-		}
 		e.pool.Free(e.deviceID, devOut, outBytes)
 		return e.cpu.Gather(context.Background(), params, indices, output)
 	}
-	if debugGPU || os.Getenv("UPLOAD_TRACE") == "1" {
-		fmt.Fprintf(os.Stderr, "[GATHER_Q8] kernel OK, devOut=%p elems=%d\n", devOut, outElems)
-	}
 
 	// Write result into output tensor as GPUStorage (pool-backed).
 	gs, err := tensor.NewGPUStorageFromPool[float32](devOut, outElems, e.pool, e.deviceID)