Making memory API less error prone; fixes buffer/constant uploads.

src/xenia/base/memory.cc

@@ -22,109 +22,99 @@ void copy_128_aligned(void* dest, const void* src, size_t count) {
   std::memcpy(dest, src, count * 16);
 }
 
-void copy_and_swap_16_aligned(uint16_t* dest, const uint16_t* src,
+void copy_and_swap_16_aligned(void* dest, const void* src, size_t count) {
-                              size_t count) {
   return copy_and_swap_16_unaligned(dest, src, count);
 }
 
-void copy_and_swap_16_unaligned(uint16_t* dest, const uint16_t* src,
+void copy_and_swap_16_unaligned(void* dest_ptr, const void* src_ptr,
                                 size_t count) {
+  auto dest = reinterpret_cast<uint16_t*>(dest_ptr);
+  auto src = reinterpret_cast<const uint16_t*>(src_ptr);
   size_t i;
-  __m128i input, output;
-
   for (i = 0; i + 8 <= count; i += 8) {
-    input = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&src[i]));
+    __m128i input = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&src[i]));
-    output = _mm_or_si128(_mm_slli_epi16(input, 8), _mm_srli_epi16(input, 8));
+    __m128i output =
+        _mm_or_si128(_mm_slli_epi16(input, 8), _mm_srli_epi16(input, 8));
     _mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
   }
-
   for (; i < count; ++i) {  // handle residual elements
     dest[i] = byte_swap(src[i]);
   }
 }
 
-void copy_and_swap_32_aligned(uint32_t* dest, const uint32_t* src,
+void copy_and_swap_32_aligned(void* dest, const void* src, size_t count) {
-                              size_t count) {
   return copy_and_swap_32_unaligned(dest, src, count);
 }
 
-void copy_and_swap_32_unaligned(uint32_t* dest, const uint32_t* src,
+void copy_and_swap_32_unaligned(void* dest_ptr, const void* src_ptr,
                                 size_t count) {
-  size_t i;
+  auto dest = reinterpret_cast<uint32_t*>(dest_ptr);
-  __m128i input, byte1, byte2, byte3, byte4, output;
+  auto src = reinterpret_cast<const uint32_t*>(src_ptr);
   __m128i byte2mask = _mm_set1_epi32(0x00FF0000);
   __m128i byte3mask = _mm_set1_epi32(0x0000FF00);
-
+  size_t i;
   for (i = 0; i + 4 <= count; i += 4) {
-    input = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&src[i]));
+    __m128i input = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&src[i]));
-
+    // Do the four shifts.
-    // Do the four shifts
+    __m128i byte1 = _mm_slli_epi32(input, 24);
-    byte1 = _mm_slli_epi32(input, 24);
+    __m128i byte2 = _mm_slli_epi32(input, 8);
-    byte2 = _mm_slli_epi32(input, 8);
+    __m128i byte3 = _mm_srli_epi32(input, 8);
-    byte3 = _mm_srli_epi32(input, 8);
+    __m128i byte4 = _mm_srli_epi32(input, 24);
-    byte4 = _mm_srli_epi32(input, 24);
+    // OR bytes together.
-
+    __m128i output = _mm_or_si128(byte1, byte4);
-    // Or bytes together
-    output = _mm_or_si128(byte1, byte4);
     byte2 = _mm_and_si128(byte2, byte2mask);
     output = _mm_or_si128(output, byte2);
     byte3 = _mm_and_si128(byte3, byte3mask);
     output = _mm_or_si128(output, byte3);
-
     _mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
   }
-
   for (; i < count; ++i) {  // handle residual elements
     dest[i] = byte_swap(src[i]);
   }
 }
 
-void copy_and_swap_64_aligned(uint64_t* dest, const uint64_t* src,
+void copy_and_swap_64_aligned(void* dest, const void* src, size_t count) {
-                              size_t count) {
   return copy_and_swap_64_unaligned(dest, src, count);
 }
 
-void copy_and_swap_64_unaligned(uint64_t* dest, const uint64_t* src,
+void copy_and_swap_64_unaligned(void* dest_ptr, const void* src_ptr,
                                 size_t count) {
-  size_t i;
+  auto dest = reinterpret_cast<uint64_t*>(dest_ptr);
-  __m128i input, byte1, byte2, byte3, byte4, output;
+  auto src = reinterpret_cast<const uint64_t*>(src_ptr);
   __m128i byte2mask = _mm_set1_epi32(0x00FF0000);
   __m128i byte3mask = _mm_set1_epi32(0x0000FF00);
-
+  size_t i;
   for (i = 0; i + 2 <= count; i += 2) {
-    input = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&src[i]));
+    __m128i input = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&src[i]));
-
+    // Do the four shifts.
-    // Do the four shifts
+    __m128i byte1 = _mm_slli_epi32(input, 24);
-    byte1 = _mm_slli_epi32(input, 24);
+    __m128i byte2 = _mm_slli_epi32(input, 8);
-    byte2 = _mm_slli_epi32(input, 8);
+    __m128i byte3 = _mm_srli_epi32(input, 8);
-    byte3 = _mm_srli_epi32(input, 8);
+    __m128i byte4 = _mm_srli_epi32(input, 24);
-    byte4 = _mm_srli_epi32(input, 24);
+    // OR bytes together.
-
+    __m128i output = _mm_or_si128(byte1, byte4);
-    // Or bytes together
-    output = _mm_or_si128(byte1, byte4);
     byte2 = _mm_and_si128(byte2, byte2mask);
     output = _mm_or_si128(output, byte2);
     byte3 = _mm_and_si128(byte3, byte3mask);
     output = _mm_or_si128(output, byte3);
-
+    // Reorder the two words.
-    // Reorder the two words
     output = _mm_shuffle_epi32(output, _MM_SHUFFLE(2, 3, 0, 1));
-
     _mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
   }
-
   for (; i < count; ++i) {  // handle residual elements
     dest[i] = byte_swap(src[i]);
   }
 }
 
-void copy_and_swap_16_in_32_aligned(uint32_t* dest, const uint32_t* src,
+void copy_and_swap_16_in_32_aligned(void* dest_ptr, const void* src_ptr,
                                     size_t count) {
+  auto dest = reinterpret_cast<uint64_t*>(dest_ptr);
+  auto src = reinterpret_cast<const uint64_t*>(src_ptr);
   size_t i;
-  __m128i input, output;
   for (i = 0; i + 4 <= count; i += 4) {
-    input = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&src[i]));
+    __m128i input = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&src[i]));
-    output = _mm_or_si128(_mm_slli_epi32(input, 16), _mm_srli_epi32(input, 16));
+    __m128i output =
+        _mm_or_si128(_mm_slli_epi32(input, 16), _mm_srli_epi32(input, 16));
     _mm_storeu_si128(reinterpret_cast<__m128i*>(&dest[i]), output);
   }
   for (; i < count; ++i) {  // handle residual elements

src/xenia/base/memory.h

@@ -123,20 +123,13 @@ inline void* low_address(void* address) {
 
 void copy_128_aligned(void* dest, const void* src, size_t count);
 
-void copy_and_swap_16_aligned(uint16_t* dest, const uint16_t* src,
+void copy_and_swap_16_aligned(void* dest, const void* src, size_t count);
-                              size_t count);
+void copy_and_swap_16_unaligned(void* dest, const void* src, size_t count);
-void copy_and_swap_16_unaligned(uint16_t* dest, const uint16_t* src,
+void copy_and_swap_32_aligned(void* dest, const void* src, size_t count);
-                                size_t count);
+void copy_and_swap_32_unaligned(void* dest, const void* src, size_t count);
-void copy_and_swap_32_aligned(uint32_t* dest, const uint32_t* src,
+void copy_and_swap_64_aligned(void* dest, const void* src, size_t count);
-                              size_t count);
+void copy_and_swap_64_unaligned(void* dest, const void* src, size_t count);
-void copy_and_swap_32_unaligned(uint32_t* dest, const uint32_t* src,
+void copy_and_swap_16_in_32_aligned(void* dest, const void* src, size_t count);
-                                size_t count);
-void copy_and_swap_64_aligned(uint64_t* dest, const uint64_t* src,
-                              size_t count);
-void copy_and_swap_64_unaligned(uint64_t* dest, const uint64_t* src,
-                                size_t count);
-void copy_and_swap_16_in_32_aligned(uint32_t* dest, const uint32_t* src,
-                                    size_t count);
 
 template <typename T>
 void copy_and_swap(T* dest, const T* src, size_t count) {

src/xenia/gpu/command_processor.cc

@@ -1019,9 +1019,8 @@ bool CommandProcessor::ExecutePacketType3_EVENT_WRITE_EXT(RingBuffer* reader,
       1,          // max z
   };
   assert_true(endianness == Endian::k8in16);
-  xe::copy_and_swap_16_aligned(
+  xe::copy_and_swap_16_aligned(memory_->TranslatePhysical(address), extents,
-      reinterpret_cast<uint16_t*>(memory_->TranslatePhysical(address)), extents,
+                               xe::countof(extents));
-      xe::countof(extents));
   trace_writer_.WriteMemoryWrite(CpuToGpu(address), sizeof(extents));
   return true;
 }

src/xenia/gpu/gl4/gl4_command_processor.cc

@@ -1410,7 +1410,7 @@ GL4CommandProcessor::UpdateStatus GL4CommandProcessor::PopulateVertexBuffers() {
       // as we copy and only if it differs from the previous value committing
       // it (and if it matches just discard and reuse).
       xe::copy_and_swap_32_aligned(
-          reinterpret_cast<uint32_t*>(allocation.host_ptr),
+          allocation.host_ptr,
           memory_->TranslatePhysical<const uint32_t*>(fetch->address << 2),
           valid_range / 4);
 

src/xenia/gpu/gl4/texture_cache.cc

@@ -662,19 +662,13 @@ void TextureSwap(Endian endianness, void* dest, const void* src,
                  size_t length) {
   switch (endianness) {
     case Endian::k8in16:
-      xe::copy_and_swap_16_aligned(reinterpret_cast<uint16_t*>(dest),
+      xe::copy_and_swap_16_aligned(dest, src, length / 2);
-                                   reinterpret_cast<const uint16_t*>(src),
-                                   length / 2);
       break;
     case Endian::k8in32:
-      xe::copy_and_swap_32_aligned(reinterpret_cast<uint32_t*>(dest),
+      xe::copy_and_swap_32_aligned(dest, src, length / 4);
-                                   reinterpret_cast<const uint32_t*>(src),
-                                   length / 4);
       break;
     case Endian::k16in32:  // Swap high and low 16 bits within a 32 bit word
-      xe::copy_and_swap_16_in_32_aligned(reinterpret_cast<uint32_t*>(dest),
+      xe::copy_and_swap_16_in_32_aligned(dest, src, length);
-                                         reinterpret_cast<const uint32_t*>(src),
-                                         length);
       break;
     default:
     case Endian::kUnspecified:

src/xenia/gpu/vulkan/buffer_cache.cc

@@ -290,13 +290,13 @@ std::pair<VkBuffer, VkDeviceSize> BufferCache::UploadIndexBuffer(
   if (format == IndexFormat::kInt16) {
     // Endian::k8in16, swap half-words.
     xe::copy_and_swap_16_aligned(
-        reinterpret_cast<uint16_t*>(transient_buffer_data_) + offset,
+        reinterpret_cast<uint8_t*>(transient_buffer_data_) + offset, source_ptr,
-        reinterpret_cast<const uint16_t*>(source_ptr), source_length / 2);
+        source_length / 2);
   } else if (format == IndexFormat::kInt32) {
     // Endian::k8in32, swap words.
     xe::copy_and_swap_32_aligned(
-        reinterpret_cast<uint32_t*>(transient_buffer_data_) + offset,
+        reinterpret_cast<uint8_t*>(transient_buffer_data_) + offset, source_ptr,
-        reinterpret_cast<const uint32_t*>(source_ptr), source_length / 4);
+        source_length / 4);
   }
 
   return {transient_index_buffer_, offset};
@@ -317,8 +317,8 @@ std::pair<VkBuffer, VkDeviceSize> BufferCache::UploadVertexBuffer(
   // TODO(benvanik): memcpy then use compute shaders to swap?
   // Endian::k8in32, swap words.
   xe::copy_and_swap_32_aligned(
-      reinterpret_cast<uint32_t*>(transient_buffer_data_) + offset,
+      reinterpret_cast<uint8_t*>(transient_buffer_data_) + offset, source_ptr,
-      reinterpret_cast<const uint32_t*>(source_ptr), source_length / 4);
+      source_length / 4);
 
   return {transient_vertex_buffer_, offset};
 }

src/xenia/kernel/xboxkrnl/xboxkrnl_video.cc

@@ -343,8 +343,7 @@ void VdSwap(lpvoid_t buffer_ptr,  // ptr into primary ringbuffer
             lpunknown_t unk8, unknown_t unk9) {
   gpu::xenos::xe_gpu_texture_fetch_t fetch;
   xe::copy_and_swap_32_unaligned(
-      reinterpret_cast<uint32_t*>(&fetch),
+      &fetch, reinterpret_cast<uint32_t*>(fetch_ptr.host_address()), 6);
-      reinterpret_cast<uint32_t*>(fetch_ptr.host_address()), 6);
 
   auto color_format = gpu::ColorFormat(color_format_ptr.value());
   auto color_space = *color_space_ptr;