pointer: add drm dumb buffers for cursors (#8399)

---------

Co-authored-by: Mihai Fufezan <[email protected]>

6 files changed, 137 insertions, 74 deletions

diff --git a/src/config/ConfigManager.cpp b/src/config/ConfigManager.cpp
index e3d59874..10299277 100644
--- a/src/config/ConfigManager.cpp
+++ b/src/config/ConfigManager.cpp
@@ -571,7 +571,7 @@ CConfigManager::CConfigManager() {
     m_pConfig->addConfigValue("cursor:sync_gsettings_theme", Hyprlang::INT{1});
     m_pConfig->addConfigValue("cursor:hide_on_key_press", Hyprlang::INT{0});
     m_pConfig->addConfigValue("cursor:hide_on_touch", Hyprlang::INT{1});
-    m_pConfig->addConfigValue("cursor:allow_dumb_copy", Hyprlang::INT{0});
+    m_pConfig->addConfigValue("cursor:use_cpu_buffer", Hyprlang::INT{0});
 
     m_pConfig->addConfigValue("autogenerated", Hyprlang::INT{0});
 
diff --git a/src/managers/PointerManager.cpp b/src/managers/PointerManager.cpp
index 75be235f..0580c9be 100644
--- a/src/managers/PointerManager.cpp
+++ b/src/managers/PointerManager.cpp
@@ -375,6 +375,8 @@ SP<Aquamarine::IBuffer> CPointerManager::renderHWCursorBuffer(SP<CPointerManager
     auto        maxSize    = state->monitor->output->cursorPlaneSize();
     auto const& cursorSize = currentCursorImage.size;
 
+    static auto PDUMB = CConfigValue<Hyprlang::INT>("cursor:use_cpu_buffer");
+
     if (maxSize == Vector2D{})
         return nullptr;
 
@@ -386,10 +388,23 @@ SP<Aquamarine::IBuffer> CPointerManager::renderHWCursorBuffer(SP<CPointerManager
     } else
         maxSize = cursorSize;
 
-    if (!state->monitor->cursorSwapchain || maxSize != state->monitor->cursorSwapchain->currentOptions().size) {
+    if (!state->monitor->cursorSwapchain || maxSize != state->monitor->cursorSwapchain->currentOptions().size ||
+        *PDUMB != (state->monitor->cursorSwapchain->getAllocator()->type() != Aquamarine::AQ_ALLOCATOR_TYPE_GBM)) {
+
+        if (!state->monitor->cursorSwapchain || *PDUMB != (state->monitor->cursorSwapchain->getAllocator()->type() != Aquamarine::AQ_ALLOCATOR_TYPE_GBM)) {
+
+            auto allocator = state->monitor->output->getBackend()->preferredAllocator();
+            if (*PDUMB) {
+                for (const auto& a : state->monitor->output->getBackend()->getAllocators()) {
+                    if (a->type() == Aquamarine::AQ_ALLOCATOR_TYPE_DRM_DUMB) {
+                        allocator = a;
+                        break;
+                    }
+                }
+            }
 
-        if (!state->monitor->cursorSwapchain)
-            state->monitor->cursorSwapchain = Aquamarine::CSwapchain::create(state->monitor->output->getBackend()->preferredAllocator(), state->monitor->output->getBackend());
+            state->monitor->cursorSwapchain = Aquamarine::CSwapchain::create(allocator, state->monitor->output->getBackend());
+        }
 
         auto options     = state->monitor->cursorSwapchain->currentOptions();
         options.size     = maxSize;
@@ -397,8 +412,10 @@ SP<Aquamarine::IBuffer> CPointerManager::renderHWCursorBuffer(SP<CPointerManager
         options.scanout  = true;
         options.cursor   = true;
         options.multigpu = state->monitor->output->getBackend()->preferredAllocator()->drmFD() != g_pCompositor->m_iDRMFD;
-        // We do not set the format. If it's unset (DRM_FORMAT_INVALID) then the swapchain will pick for us,
+        // We do not set the format (unless shm). If it's unset (DRM_FORMAT_INVALID) then the swapchain will pick for us,
         // but if it's set, we don't wanna change it.
+        if (*PDUMB)
+            options.format = DRM_FORMAT_ARGB8888;
 
         if (!state->monitor->cursorSwapchain->reconfigure(options)) {
             Debug::log(TRACE, "Failed to reconfigure cursor swapchain");
@@ -420,62 +437,71 @@ SP<Aquamarine::IBuffer> CPointerManager::renderHWCursorBuffer(SP<CPointerManager
         return nullptr;
     }
 
-    g_pHyprRenderer->makeEGLCurrent();
-    g_pHyprOpenGL->m_RenderData.pMonitor = state->monitor;
-
-    auto RBO = g_pHyprRenderer->getOrCreateRenderbuffer(buf, state->monitor->cursorSwapchain->currentOptions().format);
-    if (!RBO) {
-        Debug::log(TRACE, "Failed to create cursor RB with format {}, mod {}", buf->dmabuf().format, buf->dmabuf().modifier);
-        static auto PDUMB = CConfigValue<Hyprlang::INT>("cursor:allow_dumb_copy");
-        if (!*PDUMB)
-            return nullptr;
-
-        auto bufData = buf->beginDataPtr(0);
-        auto bufPtr  = std::get<0>(bufData);
-
-        // clear buffer
-        memset(bufPtr, 0, std::get<2>(bufData));
-
-        if (currentCursorImage.pBuffer) {
-            auto texAttrs = currentCursorImage.pBuffer->shm();
-
-            if (!texAttrs.success) {
+    if (*PDUMB) {
+        // get the texture data if available.
+        auto texData = texture->dataCopy();
+        if (texData.empty()) {
+            if (currentCursorImage.surface && currentCursorImage.surface->resource()->role->role() == SURFACE_ROLE_CURSOR) {
+                const auto SURFACE   = currentCursorImage.surface->resource();
+                auto&      shmBuffer = CCursorSurfaceRole::cursorPixelData(SURFACE);
+
+                bool       flipRB = false;
+
+                if (SURFACE->current.texture) {
+                    Debug::log(TRACE, "Cursor CPU surface: format {}, expecting AR24", FormatUtils::drmFormatName(SURFACE->current.texture->m_iDrmFormat));
+                    if (SURFACE->current.texture->m_iDrmFormat == DRM_FORMAT_ABGR8888) {
+                        Debug::log(TRACE, "Cursor CPU surface format AB24, will flip. WARNING: this will break on big endian!");
+                        flipRB = true;
+                    } else if (SURFACE->current.texture->m_iDrmFormat != DRM_FORMAT_ARGB8888) {
+                        Debug::log(TRACE, "Cursor CPU surface format rejected, falling back to sw");
+                        return nullptr;
+                    }
+                }
+
+                if (shmBuffer.data())
+                    texData = shmBuffer;
+                else {
+                    texData.resize(texture->m_vSize.x * 4 * texture->m_vSize.y);
+                    memset(texData.data(), 0x00, texData.size());
+                }
+
+                if (flipRB) {
+                    for (size_t i = 0; i < shmBuffer.size(); i += 4) {
+                        std::swap(shmBuffer.at(i), shmBuffer.at(i + 2)); // little-endian!!!!!!
+                    }
+                }
+            } else {
                 Debug::log(TRACE, "Cannot use dumb copy on dmabuf cursor buffers");
                 return nullptr;
             }
+        }
 
-            auto texData = currentCursorImage.pBuffer->beginDataPtr(GBM_BO_TRANSFER_WRITE);
-            auto texPtr  = std::get<0>(texData);
-            Debug::log(TRACE, "cursor texture {}x{} {} {} {}", texAttrs.size.x, texAttrs.size.y, (void*)texPtr, texAttrs.format, texAttrs.stride);
-            // copy cursor texture
-            for (int i = 0; i < texAttrs.size.y; i++)
-                memcpy(bufPtr + i * buf->dmabuf().strides[0], texPtr + i * texAttrs.stride, texAttrs.stride);
-        } else if (currentCursorImage.surface && currentCursorImage.surface->resource()->role->role() == SURFACE_ROLE_CURSOR) {
-            const auto SURFACE   = currentCursorImage.surface->resource();
-            auto&      shmBuffer = CCursorSurfaceRole::cursorPixelData(SURFACE);
-            Debug::log(TRACE, "cursor texture pixel data length: {}B", shmBuffer.size());
-
-            if (shmBuffer.data()) {
-                // copy cursor texture
-                // assume format is 32bpp
-                size_t STRIDE = 4 * SURFACE->current.bufferSize.x;
-                for (int i = 0; i < SURFACE->current.bufferSize.y; i++)
-                    memcpy(bufPtr + i * buf->dmabuf().strides[0], shmBuffer.data() + i * STRIDE, STRIDE);
-            } else {
-                // if there is no data, hide the cursor
-                memset(bufPtr, '\0', buf->size.x * buf->size.y * 4 /* assume 32bpp */);
-            }
+        // then, we just yeet it into the dumb buffer
 
-        } else {
-            Debug::log(TRACE, "Unsupported cursor buffer/surface, falling back to sw (can't dumb copy)");
-            return nullptr;
-        }
+        auto [data, fmt, size] = buf->beginDataPtr(0);
+
+        memset(data, 0, size);
+        if (buf->dmabuf().size.x > texture->m_vSize.x) {
+            size_t STRIDE = 4 * texture->m_vSize.x;
+            for (int i = 0; i < texture->m_vSize.y; i++)
+                memcpy(data + i * buf->dmabuf().strides[0], texData.data() + i * STRIDE, STRIDE);
+        } else
+            memcpy(data, texData.data(), std::min(size, texData.size()));
 
         buf->endDataPtr();
 
         return buf;
     }
 
+    g_pHyprRenderer->makeEGLCurrent();
+    g_pHyprOpenGL->m_RenderData.pMonitor = state->monitor;
+
+    auto RBO = g_pHyprRenderer->getOrCreateRenderbuffer(buf, state->monitor->cursorSwapchain->currentOptions().format);
+    if (!RBO) {
+        Debug::log(TRACE, "Failed to create cursor RB with format {}, mod {}", buf->dmabuf().format, buf->dmabuf().modifier);
+        return nullptr;
+    }
+
     RBO->bind();
 
     g_pHyprOpenGL->beginSimple(state->monitor.lock(), {0, 0, INT16_MAX, INT16_MAX}, RBO);
@@ -773,7 +799,7 @@ SP<CTexture> CPointerManager::getCurrentCursorTexture() {
 
     if (currentCursorImage.pBuffer) {
         if (!currentCursorImage.bufferTex)
-            currentCursorImage.bufferTex = makeShared<CTexture>(currentCursorImage.pBuffer);
+            currentCursorImage.bufferTex = makeShared<CTexture>(currentCursorImage.pBuffer, true);
         return currentCursorImage.bufferTex;
     }
 
diff --git a/src/protocols/core/Compositor.cpp b/src/protocols/core/Compositor.cpp
index 57f61f87..89f2a4cb 100644
--- a/src/protocols/core/Compositor.cpp
+++ b/src/protocols/core/Compositor.cpp
@@ -438,12 +438,13 @@ void CWLSurfaceResource::commitPendingState() {
         current.texture->m_eTransform = wlTransformToHyprutils(current.transform);
 
     if (current.buffer && current.buffer->buffer) {
-        current.buffer->buffer->update(accumulateCurrentBufferDamage());
+        const auto DAMAGE = accumulateCurrentBufferDamage();
+        current.buffer->buffer->update(DAMAGE);
 
         // if the surface is a cursor, update the shm buffer
         // TODO: don't update the entire texture
-        if (role->role() == SURFACE_ROLE_CURSOR)
-            updateCursorShm();
+        if (role->role() == SURFACE_ROLE_CURSOR && !DAMAGE.empty())
+            updateCursorShm(DAMAGE);
 
         // release the buffer if it's synchronous as update() has done everything thats needed
         // so we can let the app know we're done.
@@ -486,13 +487,12 @@ void CWLSurfaceResource::commitPendingState() {
     lastBuffer = current.buffer ? current.buffer->buffer : WP<IHLBuffer>{};
 }
 
-void CWLSurfaceResource::updateCursorShm() {
+void CWLSurfaceResource::updateCursorShm(CRegion damage) {
     auto buf = current.buffer ? current.buffer->buffer : lastBuffer;
 
     if (!buf)
         return;
 
-    // TODO: actually use damage
     auto& shmData  = CCursorSurfaceRole::cursorPixelData(self.lock());
     auto  shmAttrs = buf->shm();
 
@@ -501,11 +501,25 @@ void CWLSurfaceResource::updateCursorShm() {
         return;
     }
 
+    damage.intersect(CBox{0, 0, buf->size.x, buf->size.y});
+
     // no need to end, shm.
     auto [pixelData, fmt, bufLen] = buf->beginDataPtr(0);
 
     shmData.resize(bufLen);
-    memcpy(shmData.data(), pixelData, bufLen);
+
+    if (const auto RECTS = damage.getRects(); RECTS.size() == 1 && RECTS.at(0).x2 == buf->size.x && RECTS.at(0).y2 == buf->size.y)
+        memcpy(shmData.data(), pixelData, bufLen);
+    else {
+        for (auto& box : damage.getRects()) {
+            for (auto y = box.y1; y < box.y2; ++y) {
+                // bpp is 32 INSALLAH
+                auto begin = 4 * box.y1 * (box.x2 - box.x1) + box.x1;
+                auto len   = 4 * (box.x2 - box.x1);
+                memcpy((uint8_t*)shmData.data() + begin, (uint8_t*)pixelData + begin, len);
+            }
+        }
+    }
 }
 
 void CWLSurfaceResource::presentFeedback(timespec* when, PHLMONITOR pMonitor) {
diff --git a/src/protocols/core/Compositor.hpp b/src/protocols/core/Compositor.hpp
index e5bdf082..c036041a 100644
--- a/src/protocols/core/Compositor.hpp
+++ b/src/protocols/core/Compositor.hpp
@@ -148,7 +148,7 @@ class CWLSurfaceResource {
     void          dropCurrentBuffer();
     void          commitPendingState();
     void          bfHelper(std::vector<SP<CWLSurfaceResource>> const& nodes, std::function<void(SP<CWLSurfaceResource>, const Vector2D&, void*)> fn, void* data);
-    void          updateCursorShm();
+    void          updateCursorShm(CRegion damage = CBox{0, 0, INT16_MAX, INT16_MAX});
 
     friend class CWLPointerResource;
 };
diff --git a/src/render/Texture.cpp b/src/render/Texture.cpp
index 91e70afa..633f0212 100644
--- a/src/render/Texture.cpp
+++ b/src/render/Texture.cpp
@@ -3,6 +3,7 @@
 #include "../Compositor.hpp"
 #include "../protocols/types/Buffer.hpp"
 #include "../helpers/Format.hpp"
+#include <cstring>
 
 CTexture::CTexture() {
     // naffin'
@@ -16,7 +17,7 @@ CTexture::~CTexture() {
     destroyTexture();
 }
 
-CTexture::CTexture(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const Vector2D& size_) {
+CTexture::CTexture(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const Vector2D& size_, bool keepDataCopy) : m_iDrmFormat(drmFormat), m_bKeepDataCopy(keepDataCopy) {
     createFromShm(drmFormat, pixels, stride, size_);
 }
 
@@ -24,7 +25,7 @@ CTexture::CTexture(const Aquamarine::SDMABUFAttrs& attrs, void* image) {
     createFromDma(attrs, image);
 }
 
-CTexture::CTexture(const SP<Aquamarine::IBuffer> buffer) {
+CTexture::CTexture(const SP<Aquamarine::IBuffer> buffer, bool keepDataCopy) : m_bKeepDataCopy(keepDataCopy) {
     if (!buffer)
         return;
 
@@ -43,6 +44,8 @@ CTexture::CTexture(const SP<Aquamarine::IBuffer> buffer) {
 
         auto [pixelData, fmt, bufLen] = buffer->beginDataPtr(0);
 
+        m_iDrmFormat = fmt;
+
         createFromShm(fmt, pixelData, bufLen, shm.size);
         return;
     }
@@ -80,6 +83,11 @@ void CTexture::createFromShm(uint32_t drmFormat, uint8_t* pixels, uint32_t strid
     GLCALL(glTexImage2D(GL_TEXTURE_2D, 0, format->glInternalFormat ? format->glInternalFormat : format->glFormat, size_.x, size_.y, 0, format->glFormat, format->glType, pixels));
     GLCALL(glPixelStorei(GL_UNPACK_ROW_LENGTH_EXT, 0));
     GLCALL(glBindTexture(GL_TEXTURE_2D, 0));
+
+    if (m_bKeepDataCopy) {
+        m_vDataCopy.resize(stride * size_.y);
+        memcpy(m_vDataCopy.data(), pixels, stride * size_.y);
+    }
 }
 
 void CTexture::createFromDma(const Aquamarine::SDMABUFAttrs& attrs, void* image) {
@@ -135,6 +143,11 @@ void CTexture::update(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, cons
     GLCALL(glPixelStorei(GL_UNPACK_SKIP_ROWS_EXT, 0));
 
     glBindTexture(GL_TEXTURE_2D, 0);
+
+    if (m_bKeepDataCopy) {
+        m_vDataCopy.resize(stride * m_vSize.y);
+        memcpy(m_vDataCopy.data(), pixels, stride * m_vSize.y);
+    }
 }
 
 void CTexture::destroyTexture() {
@@ -152,3 +165,7 @@ void CTexture::allocate() {
     if (!m_iTexID)
         GLCALL(glGenTextures(1, &m_iTexID));
 }
+
+const std::vector<uint8_t>& CTexture::dataCopy() {
+    return m_vDataCopy;
+}
diff --git a/src/render/Texture.hpp b/src/render/Texture.hpp
index e0ef5503..ae949d5f 100644
--- a/src/render/Texture.hpp
+++ b/src/render/Texture.hpp
@@ -24,26 +24,32 @@ class CTexture {
     CTexture(const CTexture&&) = delete;
     CTexture(const CTexture&)  = delete;
 
-    CTexture(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const Vector2D& size);
+    CTexture(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const Vector2D& size, bool keepDataCopy = false);
 
-    CTexture(const SP<Aquamarine::IBuffer> buffer);
+    CTexture(const SP<Aquamarine::IBuffer> buffer, bool keepDataCopy = false);
     // this ctor takes ownership of the eglImage.
     CTexture(const Aquamarine::SDMABUFAttrs&, void* image);
     ~CTexture();
 
-    void        destroyTexture();
-    void        allocate();
-    void        update(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const CRegion& damage);
+    void                        destroyTexture();
+    void                        allocate();
+    void                        update(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const CRegion& damage);
+    const std::vector<uint8_t>& dataCopy();
 
-    TEXTURETYPE m_iType      = TEXTURE_RGBA;
-    GLenum      m_iTarget    = GL_TEXTURE_2D;
-    GLuint      m_iTexID     = 0;
-    Vector2D    m_vSize      = {};
-    void*       m_pEglImage  = nullptr;
-    eTransform  m_eTransform = HYPRUTILS_TRANSFORM_NORMAL;
-    bool        m_bOpaque    = false;
+    TEXTURETYPE                 m_iType      = TEXTURE_RGBA;
+    GLenum                      m_iTarget    = GL_TEXTURE_2D;
+    GLuint                      m_iTexID     = 0;
+    Vector2D                    m_vSize      = {};
+    void*                       m_pEglImage  = nullptr;
+    eTransform                  m_eTransform = HYPRUTILS_TRANSFORM_NORMAL;
+    bool                        m_bOpaque    = false;
+    uint32_t                    m_iDrmFormat = 0; // for shm
 
   private:
-    void createFromShm(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const Vector2D& size);
-    void createFromDma(const Aquamarine::SDMABUFAttrs&, void* image);
+    void                 createFromShm(uint32_t drmFormat, uint8_t* pixels, uint32_t stride, const Vector2D& size);
+    void                 createFromDma(const Aquamarine::SDMABUFAttrs&, void* image);
+
+    bool                 m_bKeepDataCopy = false;
+
+    std::vector<uint8_t> m_vDataCopy;
 };
\ No newline at end of file