VertexLoader: Convert count register to remaining register

This more accurately represents what's going on, and also ends at 0 instead of 1, making some indexing operations easier.  This also changes it so that position_matrix_index_cache actually starts from index 0 instead of index 1.

Source/Core/VideoCommon/VertexLoader.cpp

@@ -22,8 +22,8 @@ u8* g_vertex_manager_write_ptr;
 static void PosMtx_ReadDirect_UByte(VertexLoader* loader)
 {
   u32 posmtx = DataRead<u8>() & 0x3f;
-  if (loader->m_counter < 3)
+  if (loader->m_remaining < 3)
-    VertexLoaderManager::position_matrix_index_cache[loader->m_counter + 1] = posmtx;
+    VertexLoaderManager::position_matrix_index_cache[loader->m_remaining] = posmtx;
   DataWrite<u32>(posmtx);
   PRIM_LOG("posmtx: {}, ", posmtx);
 }
@@ -257,7 +257,7 @@ int VertexLoader::RunVertices(DataReader src, DataReader dst, int count)
   m_numLoadedVertices += count;
   m_skippedVertices = 0;
 
-  for (m_counter = count - 1; m_counter >= 0; m_counter--)
+  for (m_remaining = count - 1; m_remaining >= 0; m_remaining--)
   {
     m_tcIndex = 0;
     m_colIndex = 0;

Source/Core/VideoCommon/VertexLoader.h

@@ -35,7 +35,7 @@ public:
   int m_texmtxread;
   bool m_vertexSkip;
   int m_skippedVertices;
-  int m_counter;
+  int m_remaining;
 
 private:
   // Pipeline.

Source/Core/VideoCommon/VertexLoaderARM64.cpp

@@ -14,7 +14,7 @@ using namespace Arm64Gen;
 
 constexpr ARM64Reg src_reg = ARM64Reg::X0;
 constexpr ARM64Reg dst_reg = ARM64Reg::X1;
-constexpr ARM64Reg count_reg = ARM64Reg::W2;
+constexpr ARM64Reg remaining_reg = ARM64Reg::W2;
 constexpr ARM64Reg skipped_reg = ARM64Reg::W17;
 constexpr ARM64Reg scratch1_reg = ARM64Reg::W16;
 constexpr ARM64Reg scratch2_reg = ARM64Reg::W15;
@@ -209,13 +209,10 @@ int VertexLoaderARM64::ReadVertex(VertexComponentFormat attribute, ComponentForm
   // Z-Freeze
   if (native_format == &m_native_vtx_decl.position)
   {
-    CMP(count_reg, 3);
+    CMP(remaining_reg, 3);
-    FixupBranch dont_store = B(CC_GT);
+    FixupBranch dont_store = B(CC_GE);
     MOVP2R(EncodeRegTo64(scratch2_reg), VertexLoaderManager::position_cache.data());
-    ADD(EncodeRegTo64(scratch1_reg), EncodeRegTo64(scratch2_reg), EncodeRegTo64(count_reg),
+    m_float_emit.STR(128, coords, EncodeRegTo64(scratch2_reg), ArithOption(remaining_reg, true));
-        ArithOption(EncodeRegTo64(count_reg), ShiftType::LSL, 4));
-    m_float_emit.STUR(write_size, coords, EncodeRegTo64(scratch1_reg),
-                      -int(sizeof(decltype(VertexLoaderManager::position_cache[0]))));
     SetJumpTarget(dont_store);
   }
 
@@ -404,7 +401,7 @@ void VertexLoaderARM64::GenerateVertexLoader()
   AlignCode16();
   if (IsIndexed(m_VtxDesc.low.Position))
     MOV(skipped_reg, ARM64Reg::WZR);
-  MOV(saved_count, count_reg);
+  ADD(saved_count, remaining_reg, 1);
 
   MOVP2R(stride_reg, g_main_cp_state.array_strides.data());
   MOVP2R(arraybase_reg, VertexLoaderManager::cached_arraybases.data());
@@ -421,10 +418,10 @@ void VertexLoaderARM64::GenerateVertexLoader()
     STR(IndexType::Unsigned, scratch1_reg, dst_reg, m_dst_ofs);
 
     // Z-Freeze
-    CMP(count_reg, 3);
+    CMP(remaining_reg, 3);
-    FixupBranch dont_store = B(CC_GT);
+    FixupBranch dont_store = B(CC_GE);
     MOVP2R(EncodeRegTo64(scratch2_reg), VertexLoaderManager::position_matrix_index_cache.data());
-    STR(scratch1_reg, EncodeRegTo64(scratch2_reg), ArithOption(count_reg, true));
+    STR(scratch1_reg, EncodeRegTo64(scratch2_reg), ArithOption(remaining_reg, true));
     SetJumpTarget(dont_store);
 
     m_native_vtx_decl.posmtx.components = 4;
@@ -584,8 +581,8 @@ void VertexLoaderARM64::GenerateVertexLoader()
   const u8* cont = GetCodePtr();
   ADD(src_reg, src_reg, m_src_ofs);
 
-  SUB(count_reg, count_reg, 1);
+  SUBS(remaining_reg, remaining_reg, 1);
-  CBNZ(count_reg, loop_start);
+  B(CCFlags::CC_GE, loop_start);
 
   if (IsIndexed(m_VtxDesc.low.Position))
   {
@@ -612,5 +609,5 @@ int VertexLoaderARM64::RunVertices(DataReader src, DataReader dst, int count)
 {
   m_numLoadedVertices += count;
   return ((int (*)(u8 * src, u8 * dst, int count)) region)(src.GetPointer(), dst.GetPointer(),
-                                                           count);
+                                                           count - 1);
 }

Source/Core/VideoCommon/VertexLoaderManager.cpp

@@ -32,10 +32,9 @@
 namespace VertexLoaderManager
 {
 // Used by zfreeze
-std::array<std::array<float, 4>, 3> position_cache;
+std::array<u32, 3> position_matrix_index_cache;
-// The counter added to the address of the array is 1, 2, or 3, but never zero.
+// 3 vertices, 4 floats each to allow SIMD overwrite
-// So only index 1 - 3 are used.
+alignas(sizeof(std::array<float, 4>)) std::array<std::array<float, 4>, 3> position_cache;
-std::array<u32, 4> position_matrix_index_cache;
 
 static NativeVertexFormatMap s_native_vertex_map;
 static NativeVertexFormat* s_current_vtx_fmt;
@@ -251,8 +250,9 @@ static VertexLoaderBase* RefreshLoader(int vtx_attr_group, bool preprocess = fal
 int RunVertices(int vtx_attr_group, OpcodeDecoder::Primitive primitive, int count, DataReader src,
                 bool is_preprocess)
 {
-  if (!count)
+  if (count == 0)
     return 0;
+  ASSERT(count > 0);
 
   VertexLoaderBase* loader = RefreshLoader(vtx_attr_group, is_preprocess);
 

Source/Core/VideoCommon/VertexLoaderManager.h

@@ -54,7 +54,7 @@ void UpdateVertexArrayPointers();
 // Position cache for zfreeze (3 vertices, 4 floats each to allow SIMD overwrite).
 // These arrays are in reverse order.
 extern std::array<std::array<float, 4>, 3> position_cache;
-extern std::array<u32, 4> position_matrix_index_cache;
+extern std::array<u32, 3> position_matrix_index_cache;
 
 // VB_HAS_X. Bitmask telling what vertex components are present.
 extern u32 g_current_components;

Source/Core/VideoCommon/VertexLoaderX64.cpp

@@ -26,7 +26,9 @@ static const X64Reg dst_reg = ABI_PARAM2;
 static const X64Reg scratch1 = RAX;
 static const X64Reg scratch2 = ABI_PARAM3;
 static const X64Reg scratch3 = ABI_PARAM4;
-static const X64Reg count_reg = R10;
+// The remaining number of vertices to be processed.  Starts at count - 1, and the final loop has it
+// at 0.
+static const X64Reg remaining_reg = R10;
 static const X64Reg skipped_reg = R11;
 static const X64Reg base_reg = RBX;
 
@@ -117,10 +119,11 @@ int VertexLoaderX64::ReadVertex(OpArg data, VertexComponentFormat attribute, Com
   const auto write_zfreeze = [&]() {  // zfreeze
     if (native_format == &m_native_vtx_decl.position)
     {
-      CMP(32, R(count_reg), Imm8(3));
+      CMP(32, R(remaining_reg), Imm8(3));
-      FixupBranch dont_store = J_CC(CC_A);
+      FixupBranch dont_store = J_CC(CC_AE);
-      LEA(32, scratch3,
+      // The position cache is composed of 3 rows of 4 floats each; since each float is 4 bytes,
-          MScaled(count_reg, SCALE_4, -int(VertexLoaderManager::position_cache[0].size())));
+      // we need to scale by 4 twice to cover the 4 floats.
+      LEA(32, scratch3, MScaled(remaining_reg, SCALE_4, 0));
       MOVUPS(MPIC(VertexLoaderManager::position_cache.data(), scratch3, SCALE_4), coords);
       SetJumpTarget(dont_store);
     }
@@ -381,7 +384,7 @@ void VertexLoaderX64::ReadColor(OpArg data, VertexComponentFormat attribute, Col
 void VertexLoaderX64::GenerateVertexLoader()
 {
   BitSet32 regs = {src_reg,  dst_reg,       scratch1,    scratch2,
-                   scratch3, count_reg, skipped_reg, base_reg};
+                   scratch3, remaining_reg, skipped_reg, base_reg};
   regs &= ABI_ALL_CALLEE_SAVED;
   ABI_PushRegistersAndAdjustStack(regs, 0);
 
@@ -389,7 +392,9 @@ void VertexLoaderX64::GenerateVertexLoader()
   PUSH(32, R(ABI_PARAM3));
 
   // ABI_PARAM3 is one of the lower registers, so free it for scratch2.
-  MOV(32, R(count_reg), R(ABI_PARAM3));
+  // We also have it end at a value of 0, to simplify indexing for zfreeze;
+  // this requires subtracting 1 at the start.
+  LEA(32, remaining_reg, MDisp(ABI_PARAM3, -1));
 
   MOV(64, R(base_reg), R(ABI_PARAM4));
 
@@ -407,9 +412,9 @@ void VertexLoaderX64::GenerateVertexLoader()
     MOV(32, MDisp(dst_reg, m_dst_ofs), R(scratch1));
 
     // zfreeze
-    CMP(32, R(count_reg), Imm8(3));
+    CMP(32, R(remaining_reg), Imm8(3));
-    FixupBranch dont_store = J_CC(CC_A);
+    FixupBranch dont_store = J_CC(CC_AE);
-    MOV(32, MPIC(VertexLoaderManager::position_matrix_index_cache.data(), count_reg, SCALE_4),
+    MOV(32, MPIC(VertexLoaderManager::position_matrix_index_cache.data(), remaining_reg, SCALE_4),
         R(scratch1));
     SetJumpTarget(dont_store);
 
@@ -509,8 +514,8 @@ void VertexLoaderX64::GenerateVertexLoader()
   const u8* cont = GetCodePtr();
   ADD(64, R(src_reg), Imm32(m_src_ofs));
 
-  SUB(32, R(count_reg), Imm8(1));
+  SUB(32, R(remaining_reg), Imm8(1));
-  J_CC(CC_NZ, loop_start);
+  J_CC(CC_AE, loop_start);
 
   // Get the original count.
   POP(32, R(ABI_RETURN));

Source/Core/VideoCommon/VertexLoader_Position.cpp

@@ -41,8 +41,8 @@ void Pos_ReadDirect(VertexLoader* loader)
   for (int i = 0; i < N; ++i)
   {
     const float value = PosScale(src.Read<T>(), scale);
-    if (loader->m_counter < 3)
+    if (loader->m_remaining < 3)
-      VertexLoaderManager::position_cache[loader->m_counter][i] = value;
+      VertexLoaderManager::position_cache[loader->m_remaining][i] = value;
     dst.Write(value);
   }
 
@@ -68,8 +68,8 @@ void Pos_ReadIndex(VertexLoader* loader)
   for (int i = 0; i < N; ++i)
   {
     const float value = PosScale(Common::FromBigEndian(data[i]), scale);
-    if (loader->m_counter < 3)
+    if (loader->m_remaining < 3)
-      VertexLoaderManager::position_cache[loader->m_counter][i] = value;
+      VertexLoaderManager::position_cache[loader->m_remaining][i] = value;
     dst.Write(value);
   }
 

Source/Core/VideoCommon/VertexManagerBase.cpp

@@ -558,7 +558,7 @@ void VertexManagerBase::CalculateZSlope(NativeVertexFormat* format)
   {
     // If this vertex format has per-vertex position matrix IDs, look it up.
     if (vert_decl.posmtx.enable)
-      mtxIdx = VertexLoaderManager::position_matrix_index_cache[3 - i];
+      mtxIdx = VertexLoaderManager::position_matrix_index_cache[2 - i];
 
     if (vert_decl.position.components == 2)
       VertexLoaderManager::position_cache[2 - i][2] = 0;