video_core/GPU Implemented the GPU PFIFO puller semaphore operations. (#1908)

* Implemented the puller semaphore operations.

* Nit: Fix 2 style issues

* Nit: Add Break to default case.

* Fix style.

* Update for comments. Added ReferenceCount method

* Forgot to remove GpuSmaphoreAddress union.

* Fix the clang-format issues.

* More clang formatting.

* two more white spaces for the Clang formatting.

* Move puller members into the regs union

* Updated to use Memory::WriteBlock instead of Memory::Write*

* Fix clang style issues

* White space clang error

* Removing unused funcitons and other pr comment

* Removing unused funcitons and other pr comment

* More union magic for setting regs value.

* union magic refcnt as well

*  Remove local var

* Set up the regs and regs_assert_positions up properly

* Fix clang error

src/video_core/gpu.cpp

@@ -3,6 +3,8 @@
 // Refer to the license.txt file included.
 
 #include "common/assert.h"
+#include "core/core_timing.h"
+#include "core/memory.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/kepler_memory.h"
 #include "video_core/engines/maxwell_3d.h"
@@ -124,9 +126,36 @@ u32 DepthFormatBytesPerPixel(DepthFormat format) {
     }
 }
 
+// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence
+// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4.
+// So the values you see in docs might be multiplied by 4.
 enum class BufferMethods {
-    BindObject = 0,
-    CountBufferMethods = 0x40,
+    BindObject = 0x0,
+    Nop = 0x2,
+    SemaphoreAddressHigh = 0x4,
+    SemaphoreAddressLow = 0x5,
+    SemaphoreSequence = 0x6,
+    SemaphoreTrigger = 0x7,
+    NotifyIntr = 0x8,
+    WrcacheFlush = 0x9,
+    Unk28 = 0xA,
+    Unk2c = 0xB,
+    RefCnt = 0x14,
+    SemaphoreAcquire = 0x1A,
+    SemaphoreRelease = 0x1B,
+    Unk70 = 0x1C,
+    Unk74 = 0x1D,
+    Unk78 = 0x1E,
+    Unk7c = 0x1F,
+    Yield = 0x20,
+    NonPullerMethods = 0x40,
+};
+
+enum class GpuSemaphoreOperation {
+    AcquireEqual = 0x1,
+    WriteLong = 0x2,
+    AcquireGequal = 0x4,
+    AcquireMask = 0x8,
 };
 
 void GPU::CallMethod(const MethodCall& method_call) {
@@ -135,20 +164,78 @@ void GPU::CallMethod(const MethodCall& method_call) {
 
     ASSERT(method_call.subchannel < bound_engines.size());
 
-    if (method_call.method == static_cast<u32>(BufferMethods::BindObject)) {
-        // Bind the current subchannel to the desired engine id.
-        LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel,
-                  method_call.argument);
-        bound_engines[method_call.subchannel] = static_cast<EngineID>(method_call.argument);
-        return;
+    if (ExecuteMethodOnEngine(method_call)) {
+        CallEngineMethod(method_call);
+    } else {
+        CallPullerMethod(method_call);
     }
+}
+
+bool GPU::ExecuteMethodOnEngine(const MethodCall& method_call) {
+    const auto method = static_cast<BufferMethods>(method_call.method);
+    return method >= BufferMethods::NonPullerMethods;
+}
 
-    if (method_call.method < static_cast<u32>(BufferMethods::CountBufferMethods)) {
-        // TODO(Subv): Research and implement these methods.
-        LOG_ERROR(HW_GPU, "Special buffer methods other than Bind are not implemented");
-        return;
+void GPU::CallPullerMethod(const MethodCall& method_call) {
+    regs.reg_array[method_call.method] = method_call.argument;
+    const auto method = static_cast<BufferMethods>(method_call.method);
+
+    switch (method) {
+    case BufferMethods::BindObject: {
+        ProcessBindMethod(method_call);
+        break;
+    }
+    case BufferMethods::Nop:
+    case BufferMethods::SemaphoreAddressHigh:
+    case BufferMethods::SemaphoreAddressLow:
+    case BufferMethods::SemaphoreSequence:
+    case BufferMethods::RefCnt:
+        break;
+    case BufferMethods::SemaphoreTrigger: {
+        ProcessSemaphoreTriggerMethod();
+        break;
+    }
+    case BufferMethods::NotifyIntr: {
+        // TODO(Kmather73): Research and implement this method.
+        LOG_ERROR(HW_GPU, "Special puller engine method NotifyIntr not implemented");
+        break;
+    }
+    case BufferMethods::WrcacheFlush: {
+        // TODO(Kmather73): Research and implement this method.
+        LOG_ERROR(HW_GPU, "Special puller engine method WrcacheFlush not implemented");
+        break;
+    }
+    case BufferMethods::Unk28: {
+        // TODO(Kmather73): Research and implement this method.
+        LOG_ERROR(HW_GPU, "Special puller engine method Unk28 not implemented");
+        break;
+    }
+    case BufferMethods::Unk2c: {
+        // TODO(Kmather73): Research and implement this method.
+        LOG_ERROR(HW_GPU, "Special puller engine method Unk2c not implemented");
+        break;
+    }
+    case BufferMethods::SemaphoreAcquire: {
+        ProcessSemaphoreAcquire();
+        break;
     }
+    case BufferMethods::SemaphoreRelease: {
+        ProcessSemaphoreRelease();
+        break;
+    }
+    case BufferMethods::Yield: {
+        // TODO(Kmather73): Research and implement this method.
+        LOG_ERROR(HW_GPU, "Special puller engine method Yield not implemented");
+        break;
+    }
+    default:
+        LOG_ERROR(HW_GPU, "Special puller engine method {:X} not implemented",
+                  static_cast<u32>(method));
+        break;
+    }
+}
 
+void GPU::CallEngineMethod(const MethodCall& method_call) {
     const EngineID engine = bound_engines[method_call.subchannel];
 
     switch (engine) {
@@ -172,4 +259,76 @@ void GPU::CallMethod(const MethodCall& method_call) {
     }
 }
 
+void GPU::ProcessBindMethod(const MethodCall& method_call) {
+    // Bind the current subchannel to the desired engine id.
+    LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel,
+              method_call.argument);
+    bound_engines[method_call.subchannel] = static_cast<EngineID>(method_call.argument);
+}
+
+void GPU::ProcessSemaphoreTriggerMethod() {
+    const auto semaphoreOperationMask = 0xF;
+    const auto op =
+        static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask);
+    if (op == GpuSemaphoreOperation::WriteLong) {
+        auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
+        struct Block {
+            u32 sequence;
+            u32 zeros = 0;
+            u64 timestamp;
+        };
+
+        Block block{};
+        block.sequence = regs.semaphore_sequence;
+        // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of
+        // CoreTiming
+        block.timestamp = CoreTiming::GetTicks();
+        Memory::WriteBlock(*address, &block, sizeof(block));
+    } else {
+        const auto address =
+            memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
+        const u32 word = Memory::Read32(*address);
+        if ((op == GpuSemaphoreOperation::AcquireEqual && word == regs.semaphore_sequence) ||
+            (op == GpuSemaphoreOperation::AcquireGequal &&
+             static_cast<s32>(word - regs.semaphore_sequence) > 0) ||
+            (op == GpuSemaphoreOperation::AcquireMask && (word & regs.semaphore_sequence))) {
+            // Nothing to do in this case
+        } else {
+            regs.acquire_source = true;
+            regs.acquire_value = regs.semaphore_sequence;
+            if (op == GpuSemaphoreOperation::AcquireEqual) {
+                regs.acquire_active = true;
+                regs.acquire_mode = false;
+            } else if (op == GpuSemaphoreOperation::AcquireGequal) {
+                regs.acquire_active = true;
+                regs.acquire_mode = true;
+            } else if (op == GpuSemaphoreOperation::AcquireMask) {
+                // TODO(kemathe) The acquire mask operation waits for a value that, ANDed with
+                // semaphore_sequence, gives a non-0 result
+                LOG_ERROR(HW_GPU, "Invalid semaphore operation AcquireMask not implemented");
+            } else {
+                LOG_ERROR(HW_GPU, "Invalid semaphore operation");
+            }
+        }
+    }
+}
+
+void GPU::ProcessSemaphoreRelease() {
+    const auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
+    Memory::Write32(*address, regs.semaphore_release);
+}
+
+void GPU::ProcessSemaphoreAcquire() {
+    const auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
+    const u32 word = Memory::Read32(*address);
+    const auto value = regs.semaphore_acquire;
+    if (word != value) {
+        regs.acquire_active = true;
+        regs.acquire_value = value;
+        // TODO(kemathe73) figure out how to do the acquire_timeout
+        regs.acquire_mode = false;
+        regs.acquire_source = false;
+    }
+}
+
 } // namespace Tegra

src/video_core/gpu.h

@@ -156,6 +156,46 @@ public:
     /// Returns a const reference to the GPU DMA pusher.
     const Tegra::DmaPusher& DmaPusher() const;
 
+    struct Regs {
+        static constexpr size_t NUM_REGS = 0x100;
+
+        union {
+            struct {
+                INSERT_PADDING_WORDS(0x4);
+                struct {
+                    u32 address_high;
+                    u32 address_low;
+
+                    GPUVAddr SmaphoreAddress() const {
+                        return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
+                                                     address_low);
+                    }
+                } smaphore_address;
+
+                u32 semaphore_sequence;
+                u32 semaphore_trigger;
+                INSERT_PADDING_WORDS(0xC);
+
+                // The puser and the puller share the reference counter, the pusher only has read
+                // access
+                u32 reference_count;
+                INSERT_PADDING_WORDS(0x5);
+
+                u32 semaphore_acquire;
+                u32 semaphore_release;
+                INSERT_PADDING_WORDS(0xE4);
+
+                // Puller state
+                u32 acquire_mode;
+                u32 acquire_source;
+                u32 acquire_active;
+                u32 acquire_timeout;
+                u32 acquire_value;
+            };
+            std::array<u32, NUM_REGS> reg_array;
+        };
+    } regs{};
+
 private:
     std::unique_ptr<Tegra::DmaPusher> dma_pusher;
     std::unique_ptr<Tegra::MemoryManager> memory_manager;
@@ -173,6 +213,37 @@ private:
     std::unique_ptr<Engines::MaxwellDMA> maxwell_dma;
     /// Inline memory engine
     std::unique_ptr<Engines::KeplerMemory> kepler_memory;
+
+    void ProcessBindMethod(const MethodCall& method_call);
+    void ProcessSemaphoreTriggerMethod();
+    void ProcessSemaphoreRelease();
+    void ProcessSemaphoreAcquire();
+
+    // Calls a GPU puller method.
+    void CallPullerMethod(const MethodCall& method_call);
+    // Calls a GPU engine method.
+    void CallEngineMethod(const MethodCall& method_call);
+    // Determines where the method should be executed.
+    bool ExecuteMethodOnEngine(const MethodCall& method_call);
 };
 
+#define ASSERT_REG_POSITION(field_name, position)                                                  \
+    static_assert(offsetof(GPU::Regs, field_name) == position * 4,                                 \
+                  "Field " #field_name " has invalid position")
+
+ASSERT_REG_POSITION(smaphore_address, 0x4);
+ASSERT_REG_POSITION(semaphore_sequence, 0x6);
+ASSERT_REG_POSITION(semaphore_trigger, 0x7);
+ASSERT_REG_POSITION(reference_count, 0x14);
+ASSERT_REG_POSITION(semaphore_acquire, 0x1A);
+ASSERT_REG_POSITION(semaphore_release, 0x1B);
+
+ASSERT_REG_POSITION(acquire_mode, 0x100);
+ASSERT_REG_POSITION(acquire_source, 0x101);
+ASSERT_REG_POSITION(acquire_active, 0x102);
+ASSERT_REG_POSITION(acquire_timeout, 0x103);
+ASSERT_REG_POSITION(acquire_value, 0x104);
+
+#undef ASSERT_REG_POSITION
+
 } // namespace Tegra