fix(turboquant): rebase hip-guard patch onto f2726891 (anchor moved)

The bump to turboquant f27268914 shifted the ggml-cuda.cu anchors the
hip-guard patch targeted, so apply-patches failed on the copy2d hunk.
Regenerated the patch against the new source:

- Hunk 1 (ggml_cuda_copy2d_across_devices 3D-peer fast path) is now
  HIP-ported upstream in the fork itself (it wraps the peer path in
  #if !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA) with a
  cudaMemcpy2DAsync #else fallback), so that hunk was DROPPED as
  now-upstream.
- Hunk 2 (ggml_backend_cuda_device_event_new using plain cudaEventCreate)
  is still present upstream and still breaks the HIP shim, so it is kept
  and rebased onto the moved anchor (cudaEventCreateWithFlags +
  cudaEventDisableTiming).

Verified with `git apply --check` and apply-patches.sh against a fresh
turboquant @ f27268914 checkout: applies cleanly.

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Assisted-by: Claude:opus-4.8 [Claude Code]

Author: mudler

backend/cpp/turboquant/patches/0001-hip-guard-copy2d-peer-fastpath.patch

@@ -5,51 +5,34 @@ that ggml's HIP (and MUSA) compatibility layer does not provide, breaking
 the -gpu-rocm-hipblas-turboquant build:
 
   1. ggml_cuda_copy2d_across_devices() (host-staged cross-device copy for
-     split mul_mat output) uses the CUDA 3D-peer copy APIs
+     split mul_mat output) used the CUDA 3D-peer copy APIs
      cudaMemcpy3DPeerParms / make_cudaPitchedPtr / make_cudaExtent /
-     cudaMemcpy3DPeerAsync. HIP genuinely does not support these (see the
-     fork's own comment "HIP does not support cudaMemcpy3DPeerAsync"), so
-     guard the peer fast path with #if !defined(GGML_USE_HIP) &&
-     !defined(GGML_USE_MUSA) -- matching how the fork already guards the
-     same API for the sibling 2D copy -- and fall through to the existing
-     cudaMemcpyAsync staging fallback below (functionally identical,
-     slightly slower on multi-GPU ROCm).
+     cudaMemcpy3DPeerAsync, which HIP does not support. As of turboquant
+     f27268914 the fork HIP-ports this itself: the peer fast path is now
+     wrapped in #if !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA) with
+     an #else that does a plain cudaMemcpy2DAsync device-to-device copy. The
+     hunk we used to carry here is therefore now upstream and has been
+     DROPPED.
 
-  2. ggml_backend_cuda_device_event_new() creates its event with plain
+  2. ggml_backend_cuda_device_event_new() still creates its event with plain
      cudaEventCreate, which ggml's HIP shim does not alias (it only aliases
-     cudaEventCreateWithFlags). Use cudaEventCreateWithFlags(..., 
+     cudaEventCreateWithFlags). Use cudaEventCreateWithFlags(...,
      cudaEventDisableTiming) -- exactly what the rest of this file already
-     does (cf. lines ~1034, ~3461) and HIP-safe.
+     does (cf. lines ~1062, ~3523) and HIP-safe.
 
-CUDA builds are unaffected. Drop the relevant hunk once the fork HIP-ports
-these; apply-patches.sh fails fast if an anchor goes stale.
+CUDA builds are unaffected. Drop the remaining hunk once the fork HIP-ports
+it too; apply-patches.sh fails fast if an anchor goes stale.
 
 diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
-index 0427e6b..6352e6a 100644
+index 5d009f6d..622d28b9 100644
 --- a/ggml/src/ggml-cuda/ggml-cuda.cu
 +++ b/ggml/src/ggml-cuda/ggml-cuda.cu
-@@ -1933,6 +1933,7 @@ static cudaError_t ggml_cuda_copy2d_across_devices(
-     size_t width, size_t height, cudaStream_t dst_stream, cudaStream_t src_stream) {
- 
-     const auto & info = ggml_cuda_info();
-+#if !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)  // 3D-peer copy types unmapped by ggml's HIP/MUSA shim; use staging fallback below
-     if (info.peer_access[src_device][dst_device]) {
-         cudaMemcpy3DPeerParms p = {};
-         p.dstDevice = dst_device;
-@@ -1942,6 +1943,7 @@ static cudaError_t ggml_cuda_copy2d_across_devices(
-         p.extent = make_cudaExtent(width, height, 1);
-         return cudaMemcpy3DPeerAsync(&p, dst_stream);
-     }
-+#endif // !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)
- 
-     // Fallback: stage all rows through a single contiguous pinned buffer
-     int prev_device = ggml_cuda_get_device();
-@@ -5714,7 +5716,7 @@ static ggml_backend_event_t ggml_backend_cuda_device_event_new(ggml_backend_dev_
+@@ -5795,7 +5795,7 @@ static ggml_backend_event_t ggml_backend_cuda_device_event_new(ggml_backend_dev_
      ggml_cuda_set_device(dev_ctx->device);
- 
+
      cudaEvent_t event;
 -    CUDA_CHECK(cudaEventCreate(&event));
 +    CUDA_CHECK(cudaEventCreateWithFlags(&event, cudaEventDisableTiming));
- 
+
      return new ggml_backend_event {
          /* .device  = */ dev,