[gpu/NVDRV] Finalize, improve AllocObjCtx (#333)

Improves object allocation per channel, only allowing max amount of 6 objects contexts per channel.
Previously objects were stored in a heap allocated vector which is sub-optimal for performance reasons.
The new implementation instead uses a stack based array with a O(1) approach.
This should boost performance in games which heavily rely on object context creation.

Co-authored-by: MaranBr <maranbr@outlook.com>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/333
Reviewed-by: crueter <crueter@eden-emu.dev>
Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>
Co-authored-by: SDK-Chan <sdkchan@eden-emu.dev>
Co-committed-by: SDK-Chan <sdkchan@eden-emu.dev>

2 files changed, 45 insertions, 15 deletions

diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
index 95bf18dbf7..5f754650d9 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
@@ -219,28 +219,55 @@ NvResult nvhost_gpu::AllocGPFIFOEx2(IoctlAllocGpfifoEx& params, DeviceFD fd) {
     return NvResult::Success;
 }
 
+s32_le nvhost_gpu::GetObjectContextClassNumberIndex(CtxClasses class_number) {
+    constexpr s32_le invalid_class_number_index = -1;
+    switch (class_number) {
+    case CtxClasses::Ctx2D: return 0;
+    case CtxClasses::Ctx3D: return 1;
+    case CtxClasses::CtxCompute: return 2;
+    case CtxClasses::CtxKepler: return 3;
+    case CtxClasses::CtxDMA: return 4;
+    case CtxClasses::CtxChannelGPFIFO: return 5;
+    default: return invalid_class_number_index;
+    }
+}
+
 NvResult nvhost_gpu::AllocateObjectContext(IoctlAllocObjCtx& params) {
-    LOG_DEBUG(Service_NVDRV, "called, class_num={:X}, flags={:X}, obj_id={:X}", params.class_num,
-                params.flags, params.obj_id);
+    LOG_DEBUG(Service_NVDRV, "called, class_num={:#X}, flags={:#X}, obj_id={:#X}", params.class_num,
+              params.flags, params.obj_id);
 
-    if (!channel_state->initialized) {
+    if (!channel_state || !channel_state->initialized) {
         LOG_CRITICAL(Service_NVDRV, "No address space bound to allocate a object context!");
         return NvResult::NotInitialized;
     }
 
-    switch (static_cast<CtxClasses>(params.class_num)) { 
-    case CtxClasses::Ctx2D:
-    case CtxClasses::Ctx3D:
-    case CtxClasses::CtxCompute:
-    case CtxClasses::CtxKepler:
-    case CtxClasses::CtxDMA:
-    case CtxClasses::CtxChannelGPFIFO:
-        ctxObj_params.push_back(params);
-        return NvResult::Success;
-    default:
-        LOG_ERROR(Service_NVDRV, "Invalid class number for object context: {:X}", params.class_num);
+    std::scoped_lock lk(channel_mutex);
+
+    if (params.flags) {
+        LOG_WARNING(Service_NVDRV, "non-zero flags={:#X} for class={:#X}", params.flags,
+                    params.class_num);
+
+        constexpr u32 allowed_mask{};
+        params.flags = allowed_mask;
+    }
+
+    s32_le ctx_class_number_index = 
+        GetObjectContextClassNumberIndex(static_cast<CtxClasses>(params.class_num));
+    if (ctx_class_number_index < 0) {
+        LOG_ERROR(Service_NVDRV, "Invalid class number for object context: {:#X}",
+                  params.class_num);
         return NvResult::BadParameter;
     }
+
+    if (ctxObjs[ctx_class_number_index].has_value()) {
+        LOG_ERROR(Service_NVDRV, "Object context for class {:#X} already allocated on this channel",
+                  params.class_num);
+        return NvResult::AlreadyAllocated;
+    }
+
+    ctxObjs[ctx_class_number_index] = params;
+
+    return NvResult::Success;
 }
 
 static boost::container::small_vector<Tegra::CommandHeader, 512> BuildWaitCommandList(
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
index a017cc50d0..fb0a5be959 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h
@@ -172,7 +172,7 @@ private:
     s32_le nvmap_fd{};
     u64_le user_data{};
     IoctlZCullBind zcull_params{};
-    std::vector<IoctlAllocObjCtx> ctxObj_params{};
+    std::array<std::optional<IoctlAllocObjCtx>, 6> ctxObjs{};
     u32_le channel_priority{};
     u32_le channel_timeslice{};
 
@@ -184,9 +184,12 @@ private:
     NvResult SetChannelPriority(IoctlChannelSetPriority& params);
     NvResult AllocGPFIFOEx(IoctlAllocGpfifoEx& params, DeviceFD fd);
     NvResult AllocGPFIFOEx2(IoctlAllocGpfifoEx& params, DeviceFD fd);
+
+    s32_le GetObjectContextClassNumberIndex(CtxClasses class_number);
     NvResult AllocateObjectContext(IoctlAllocObjCtx& params);
 
     NvResult SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, Tegra::CommandList&& entries);
+
     NvResult SubmitGPFIFOBase1(IoctlSubmitGpfifo& params,
                                std::span<Tegra::CommandListHeader> commands, bool kickoff = false);
     NvResult SubmitGPFIFOBase2(IoctlSubmitGpfifo& params,