Optimize mapper unlock()

- copying to dma from dma: skip if ptr is same
- Always go down the DMA path for unlock() via never
  doing sub-region updates
- skip unlock write if we didnt lock for write

Bug: 189960317
Test: ./cts-tradefed run cts -m CtsViewTestCases -t
android.view.cts.ASurfaceControlBackPressureTest#testSurfaceTransaction_setEnableBackPressure
fails fewer frames
Test: ./cts-tradefed run cts -m CtsNativeHardwareTestCases 100%

Change-Id: I96fa3637c3b03d4177aa9edd2d1ff94845fc7a7b
Merged-In: I96fa3637c3b03d4177aa9edd2d1ff94845fc7a7b

4 files changed, 61 insertions, 40 deletions

diff --git a/system/OpenglSystemCommon/HostConnection.h b/system/OpenglSystemCommon/HostConnection.h
index 6d087148..31ca84cf 100644
--- a/system/OpenglSystemCommon/HostConnection.h
+++ b/system/OpenglSystemCommon/HostConnection.h
@@ -85,7 +85,9 @@ public:
     }
     virtual uint64_t lockAndWriteDma(void* data, uint32_t size) {
         if (m_dmaPtr && m_dmaPhysAddr) {
-            memcpy(m_dmaPtr, data, size);
+            if (data != m_dmaPtr) {
+                memcpy(m_dmaPtr, data, size);
+            }
             return m_dmaPhysAddr;
         } else if (m_dmaCxt) {
             return writeGoldfishDma(data, size, m_dmaCxt);
diff --git a/system/hals/Android.mk b/system/hals/Android.mk
index 72ac6ceb..06fdc213 100644
--- a/system/hals/Android.mk
+++ b/system/hals/Android.mk
@@ -35,7 +35,7 @@ LOCAL_SHARED_LIBRARIES += \
     libcutils \
     libhidlbase \
     liblog \
-    libutils
+    libutils \
 
 LOCAL_STATIC_LIBRARIES += libqemupipe.ranchu libGoldfishAddressSpace$(GOLDFISH_OPENGL_LIB_SUFFIX)
 LOCAL_HEADER_LIBRARIES += libgralloc_cb.ranchu
@@ -75,7 +75,8 @@ LOCAL_SHARED_LIBRARIES += \
     libhidlbase \
     liblog \
     libutils \
-    libsync
+    libsync \
+	libandroidemu \
 
 LOCAL_STATIC_LIBRARIES += libqemupipe.ranchu libGoldfishAddressSpace$(GOLDFISH_OPENGL_LIB_SUFFIX)
 LOCAL_HEADER_LIBRARIES += libgralloc_cb.ranchu
diff --git a/system/hals/cb_handle_30.h b/system/hals/cb_handle_30.h
index 374008eb..7e5e73dd 100644
--- a/system/hals/cb_handle_30.h
+++ b/system/hals/cb_handle_30.h
@@ -53,7 +53,9 @@ struct cb_handle_30_t : public cb_handle_t {
                           p_mmapedOffset),
               mmapedSize(p_mmapedSize),
               bytesPerPixel(p_bytesPerPixel),
-              stride(p_stride) {
+              stride(p_stride),
+              locked(0),
+              lockedUsage(0) {
         numInts = CB_HANDLE_NUM_INTS(numFds);
     }
 
@@ -76,6 +78,8 @@ struct cb_handle_30_t : public cb_handle_t {
     uint32_t mmapedSize;            // real allocation side
     uint32_t bytesPerPixel;
     uint32_t stride;
+    uint32_t locked;
+    uint32_t lockedUsage;
 };
 
 #endif // SYSTEM_HALS_CB_HANDLE_30_H
diff --git a/system/hals/mapper3.cpp b/system/hals/mapper3.cpp
index e74e76cd..0788a0a6 100644
--- a/system/hals/mapper3.cpp
+++ b/system/hals/mapper3.cpp
@@ -23,6 +23,10 @@
 #include "FormatConversions.h"
 #include "debug.h"
 
+#include "android/base/Tracing.h"
+
+#define ATRACE_TAG ATRACE_TAG_GRAPHICS
+
 const int kOMX_COLOR_FormatYUV420Planar = 19;
 
 using ::android::hardware::hidl_handle;
@@ -465,10 +469,14 @@ private:  // **** impl ****
             cb.lockedHeight = cb.height;
         }
 
+        cb.locked = 1;
+        cb.lockedUsage = usage;
+
         RETURN(Error3::NONE);
     }
 
     Error3 unlockImpl(void* raw) {
+        AEMU_SCOPED_TRACE("unlockImpl body");
         if (!raw) {
             RETURN_ERROR(Error3::BAD_BUFFER);
         }
@@ -492,54 +500,60 @@ private:  // **** impl ****
     }
 
     void unlockHostImpl(cb_handle_30_t& cb, char* const bufferBits) {
+        AEMU_SCOPED_TRACE("unlockHostImpl body");
         const int bpp = glUtilsPixelBitSize(cb.glFormat, cb.glType) >> 3;
-        const int left = cb.lockedLeft;
-        const int top = cb.lockedTop;
-        const int width = cb.lockedWidth;
-        const int height = cb.lockedHeight;
-        const uint32_t rgbSize = width * height * bpp;
-        std::vector<char> convertedBuf;
+        const uint32_t lockedUsage = cb.lockedUsage;
+        const uint32_t rgbSize = cb.width * cb.height * bpp;
         const char* bitsToSend;
         uint32_t sizeToSend;
 
-        if (gralloc_is_yuv_format(cb.format)) {
-            bitsToSend = bufferBits;
-            switch (static_cast<PixelFormat>(cb.format)) {
-            case PixelFormat::YV12:
-                get_yv12_offsets(width, height, nullptr, nullptr, &sizeToSend);
-                break;
-            case PixelFormat::YCBCR_420_888:
-                get_yuv420p_offsets(width, height, nullptr, nullptr, &sizeToSend);
-                break;
-            default:
-                CRASH("Unexpected format, switch is out of sync with gralloc_is_yuv_format");
-                break;
+        const uint32_t usageSwWrite = (uint32_t)BufferUsage::CPU_WRITE_MASK;
+        uint32_t readOnly = (!(lockedUsage & usageSwWrite));
+
+        if (!readOnly) {
+            if (gralloc_is_yuv_format(cb.format)) {
+                bitsToSend = bufferBits;
+                switch (static_cast<PixelFormat>(cb.format)) {
+                    case PixelFormat::YV12:
+                        get_yv12_offsets(cb.width, cb.height, nullptr, nullptr, &sizeToSend);
+                        break;
+                    case PixelFormat::YCBCR_420_888:
+                        get_yuv420p_offsets(cb.width, cb.height, nullptr, nullptr, &sizeToSend);
+                        break;
+                    default:
+                        CRASH("Unexpected format, switch is out of sync with gralloc_is_yuv_format");
+                        break;
+                }
+            } else {
+                bitsToSend = bufferBits;
+                sizeToSend = rgbSize;
             }
-        } else {
-            convertedBuf.resize(rgbSize);
-            copy_rgb_buffer_from_unlocked(
-                convertedBuf.data(), bufferBits,
-                cb.width,
-                width, height, top, left, bpp);
-            bitsToSend = convertedBuf.data();
-            sizeToSend = rgbSize;
-        }
 
-        {
-            const HostConnectionSession conn = getHostConnectionSession();
-            ExtendedRCEncoderContext *const rcEnc = conn.getRcEncoder();
-            rcEnc->bindDmaDirectly(bufferBits,
-                                   getMmapedPhysAddr(cb.getMmapedOffset()));
-            rcEnc->rcUpdateColorBufferDMA(rcEnc, cb.hostHandle,
-                    left, top, width, height,
-                    cb.glFormat, cb.glType,
-                    const_cast<char*>(bitsToSend), sizeToSend);
+            {
+                const HostConnectionSession conn = getHostConnectionSession();
+                ExtendedRCEncoderContext *const rcEnc = conn.getRcEncoder();
+                {
+                    AEMU_SCOPED_TRACE("bindDmaDirectly");
+                    rcEnc->bindDmaDirectly(bufferBits,
+                            getMmapedPhysAddr(cb.getMmapedOffset()));
+                }
+                {
+                    AEMU_SCOPED_TRACE("updateColorBuffer");
+                    rcEnc->rcUpdateColorBufferDMA(rcEnc, cb.hostHandle,
+                            0, 0, cb.width, cb.height,
+                            cb.glFormat, cb.glType,
+                            const_cast<char*>(bitsToSend),
+                            sizeToSend);
+                }
+            }
         }
 
         cb.lockedLeft = 0;
         cb.lockedTop = 0;
         cb.lockedWidth = 0;
         cb.lockedHeight = 0;
+        cb.locked = 0;
+        cb.lockedUsage = 0;
     }
 
     bool isSupportedImpl(const IMapper::BufferDescriptorInfo& descriptor) const {