Packed backward pass speedup via unrolled camera position indexing

gsplat/cuda/_wrapper.py

@@ -356,6 +356,7 @@ def fully_fused_projection(
         - **batch_ids**. The batch indices of the projected Gaussians. Int32 tensor of shape [nnz].
         - **camera_ids**. The camera indices of the projected Gaussians. Int32 tensor of shape [nnz].
         - **gaussian_ids**. The column indices of the projected Gaussians. Int32 tensor of shape [nnz].
+        - **indptr**. CSR-style index pointer into gaussian_ids for batch-camera pairs. Int32 tensor of shape [B*C+1].
         - **radii**. The maximum radius of the projected Gaussians in pixel unit. Int32 tensor of shape [nnz, 2].
         - **means**. Projected Gaussian means in 2D. [nnz, 2]
         - **depths**. The z-depth of the projected Gaussians. [nnz]
@@ -1239,9 +1240,11 @@ def fully_fused_projection_with_ut(
         radial_coeffs.contiguous() if radial_coeffs is not None else None,
         tangential_coeffs.contiguous() if tangential_coeffs is not None else None,
         thin_prism_coeffs.contiguous() if thin_prism_coeffs is not None else None,
-        ftheta_coeffs.to_cpp()
-        if ftheta_coeffs is not None
-        else FThetaCameraDistortionParameters.to_cpp_default(),
+        (
+            ftheta_coeffs.to_cpp()
+            if ftheta_coeffs is not None
+            else FThetaCameraDistortionParameters.to_cpp_default()
+        ),
     )
     if not calc_compensations:
         compensations = None
@@ -1509,9 +1512,13 @@ def backward(
         tile_size = ctx.tile_size
         ftheta_coeffs = ctx.ftheta_coeffs
 
-        (v_means, v_quats, v_scales, v_colors, v_opacities,) = _make_lazy_cuda_func(
-            "rasterize_to_pixels_from_world_3dgs_bwd"
-        )(
+        (
+            v_means,
+            v_quats,
+            v_scales,
+            v_colors,
+            v_opacities,
+        ) = _make_lazy_cuda_func("rasterize_to_pixels_from_world_3dgs_bwd")(
             means,
             quats,
             scales,
@@ -1659,6 +1666,7 @@ def forward(
             batch_ids,
             camera_ids,
             gaussian_ids,
+            indptr,
             radii,
             means2d,
             depths,
@@ -1672,6 +1680,7 @@ def backward(
         v_batch_ids,
         v_camera_ids,
         v_gaussian_ids,
+        v_indptr,
         v_radii,
         v_means2d,
         v_depths,

gsplat/rendering.py

@@ -432,6 +432,7 @@ def reshape_view(C: int, world_view: torch.Tensor, N_world: list) -> torch.Tenso
             batch_ids,
             camera_ids,
             gaussian_ids,
+            indptr,
             radii,
             means2d,
             depths,
@@ -446,7 +447,7 @@ def reshape_view(C: int, world_view: torch.Tensor, N_world: list) -> torch.Tenso
         opacities = torch.broadcast_to(
             opacities[..., None, :], batch_dims + (C, N)
         )  # [..., C, N]
-        batch_ids, camera_ids, gaussian_ids = None, None, None
+        indptr, batch_ids, camera_ids, gaussian_ids = None, None, None, None
         image_ids = None
 
     if compensations is not None:
@@ -493,10 +494,27 @@ def reshape_view(C: int, world_view: torch.Tensor, N_world: list) -> torch.Tenso
             campos_rs = torch.inverse(viewmats_rs)[..., :3, 3]
             campos = 0.5 * (campos + campos_rs)  # [..., C, 3]
         if packed:
-            dirs = (
-                means.view(B, N, 3)[batch_ids, gaussian_ids]
-                - campos.view(B, C, 3)[batch_ids, camera_ids]
-            )  # [nnz, 3]
+            nnz = batch_ids.shape[0]
+            means_flat = means.view(B, N, 3)
+            campos_flat = campos.view(B, C, 3)
+
+            # Compute dirs in B*C steps to avoid many-to-one indexing of cameras in backward pass
+            dirs = torch.empty((nnz, 3), dtype=means.dtype, device=device)
+            indptr_cpu = indptr.cpu()
+            for b_idx in range(B):
+                for c_idx in range(C):
+                    bc_idx = b_idx * C + c_idx
+                    start_idx = indptr_cpu[bc_idx].item()
+                    end_idx = indptr_cpu[bc_idx + 1].item()
+                    if start_idx == end_idx:
+                        continue
+
+                    # Get the gaussian indices for this batch-camera pair and compute dirs
+                    gids = gaussian_ids[start_idx:end_idx]
+                    dirs[start_idx:end_idx] = (
+                        means_flat[b_idx, gids] - campos_flat[b_idx, c_idx]
+                    )
+
             masks = (radii > 0).all(dim=-1)  # [nnz]
             if colors.dim() == num_batch_dims + 3:
                 # Turn [..., N, K, 3] into [nnz, 3]

tests/test_basic.py

@@ -306,6 +306,7 @@ def test_fully_fused_projection_packed(
             batch_ids,
             camera_ids,
             gaussian_ids,
+            indptr,
             radii,
             means2d,
             depths,
@@ -344,6 +345,7 @@ def test_fully_fused_projection_packed(
             batch_ids,
             camera_ids,
             gaussian_ids,
+            indptr,
             radii,
             means2d,
             depths,