From abea6fa90c901d0b47487ed38d44511b18f0addf Mon Sep 17 00:00:00 2001
From: bunnei
Date: Fri, 23 Nov 2018 23:20:56 -0500
Subject: gpu: Rewrite GPU command list processing with DmaPusher class.

- More accurate impl., fixes Undertale (among other games).
---
 src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

(limited to 'src/core/hle')

diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
index 874d5e1c3..39a58b685 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
@@ -8,7 +8,6 @@
 #include "core/core.h"
 #include "core/hle/service/nvdrv/devices/nvhost_gpu.h"
 #include "core/memory.h"
-#include "video_core/command_processor.h"
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
 
@@ -129,6 +128,14 @@ u32 nvhost_gpu::AllocateObjectContext(const std::vector<u8>& input, std::vector<
     return 0;
 }
 
+static void PushGPUEntries(const std::vector<Tegra::CommandListHeader>& entries) {
+    auto& dma_pusher{Core::System::GetInstance().GPU().DmaPusher()};
+    for (const auto& entry : entries) {
+        dma_pusher.Push(entry);
+    }
+    dma_pusher.DispatchCalls();
+}
+
 u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& output) {
     if (input.size() < sizeof(IoctlSubmitGpfifo)) {
         UNIMPLEMENTED();
@@ -146,7 +153,7 @@ u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& outp
     std::memcpy(entries.data(), &input[sizeof(IoctlSubmitGpfifo)],
                 params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    Core::System::GetInstance().GPU().ProcessCommandLists(entries);
+    PushGPUEntries(entries);
 
     params.fence_out.id = 0;
     params.fence_out.value = 0;
@@ -167,7 +174,7 @@ u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output)
     Memory::ReadBlock(params.address, entries.data(),
                       params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    Core::System::GetInstance().GPU().ProcessCommandLists(entries);
+    PushGPUEntries(entries);
 
     params.fence_out.id = 0;
     params.fence_out.value = 0;
-- 
cgit v1.2.3


From ac74b71d7530452126792c5fa0bf01fe7378ba00 Mon Sep 17 00:00:00 2001
From: bunnei
Date: Tue, 27 Nov 2018 19:17:33 -0500
Subject: dma_pushbuffer: Optimize to avoid loop and copy on Push.

---
 src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp | 14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)

(limited to 'src/core/hle')

diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
index 39a58b685..2e2b0ae1c 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp
@@ -128,11 +128,9 @@ u32 nvhost_gpu::AllocateObjectContext(const std::vector<u8>& input, std::vector<
     return 0;
 }
 
-static void PushGPUEntries(const std::vector<Tegra::CommandListHeader>& entries) {
+static void PushGPUEntries(Tegra::CommandList&& entries) {
     auto& dma_pusher{Core::System::GetInstance().GPU().DmaPusher()};
-    for (const auto& entry : entries) {
-        dma_pusher.Push(entry);
-    }
+    dma_pusher.Push(std::move(entries));
     dma_pusher.DispatchCalls();
 }
 
@@ -149,11 +147,11 @@ u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& outp
                                    params.num_entries * sizeof(Tegra::CommandListHeader),
                "Incorrect input size");
 
-    std::vector<Tegra::CommandListHeader> entries(params.num_entries);
+    Tegra::CommandList entries(params.num_entries);
     std::memcpy(entries.data(), &input[sizeof(IoctlSubmitGpfifo)],
                 params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    PushGPUEntries(entries);
+    PushGPUEntries(std::move(entries));
 
     params.fence_out.id = 0;
     params.fence_out.value = 0;
@@ -170,11 +168,11 @@ u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output)
     LOG_WARNING(Service_NVDRV, "(STUBBED) called, gpfifo={:X}, num_entries={:X}, flags={:X}",
                 params.address, params.num_entries, params.flags);
 
-    std::vector<Tegra::CommandListHeader> entries(params.num_entries);
+    Tegra::CommandList entries(params.num_entries);
     Memory::ReadBlock(params.address, entries.data(),
                       params.num_entries * sizeof(Tegra::CommandListHeader));
 
-    PushGPUEntries(entries);
+    PushGPUEntries(std::move(entries));
 
     params.fence_out.id = 0;
     params.fence_out.value = 0;
-- 
cgit v1.2.3