Optimize display lists in J3DShapeDraw

This is a stop-gap until DL optimization
is upstreamed to Aurora.

libs/JSystem/include/JSystem/J3DGraphBase/J3DShapeDraw.h

@@ -1,6 +1,7 @@
 #ifndef J3DSHAPEDRAW_H
 #define J3DSHAPEDRAW_H
 
+#include <gx.h>
 #include <types.h>
 
 /**
@@ -12,6 +13,9 @@ public:
     u32 countVertex(u32);
     void addTexMtxIndexInDL(u32, u32, u32);
     J3DShapeDraw(u8 const*, u32);
+#if TARGET_PC
+    J3DShapeDraw(u8 const*, u32, const GXVtxDescList*);
+#endif
     void draw() const;
 
     virtual ~J3DShapeDraw();

libs/JSystem/src/J3DGraphBase/J3DShapeDraw.cpp

@@ -1,15 +1,310 @@
-#include "JSystem/JSystem.h" // IWYU pragma: keep
+#include "JSystem/JSystem.h"  // IWYU pragma: keep
 
+#include <cstring>
+#include <gx.h>
+#include <stdint.h>
 #include "JSystem/J3DGraphBase/J3DShapeDraw.h"
 #include "JSystem/JKernel/JKRHeap.h"
-#include <cstring>
-#include <stdint.h>
-#include <gx.h>
+
+#if TARGET_PC
+#include <algorithm>
+#include <tracy/Tracy.hpp>
+#include <vector>
+#include "dusk/logging.h"
+
+namespace {
+
+u16 read_be16(const u8* data) {
+    return (u16(data[0]) << 8) | data[1];
+}
+
+void append_be16(std::vector<u8>& out, u16 value) {
+    out.push_back(value >> 8);
+    out.push_back(value & 0xFF);
+}
+
+void append_bytes(std::vector<u8>& out, const u8* data, u32 size) {
+    out.insert(out.end(), data, data + size);
+}
+
+bool is_matrix_idx_attr(GXAttr attr) {
+    return attr >= GX_VA_PNMTXIDX && attr <= GX_VA_TEX7MTXIDX;
+}
+
+bool is_draw_opcode(u8 opcode) {
+    return opcode == GX_QUADS || opcode == GX_TRIANGLES || opcode == GX_TRIANGLESTRIP ||
+           opcode == GX_TRIANGLEFAN || opcode == GX_LINES || opcode == GX_LINESTRIP ||
+           opcode == GX_POINTS;
+}
+
+bool is_mergeable_draw_opcode(u8 opcode) {
+    return opcode == GX_QUADS || opcode == GX_TRIANGLES || opcode == GX_TRIANGLESTRIP ||
+           opcode == GX_TRIANGLEFAN;
+}
+
+bool calc_vtx_stride(const GXVtxDescList* vtxDesc, u32& stride) {
+    stride = 0;
+    for (; vtxDesc->attr != GX_VA_NULL; vtxDesc++) {
+        switch (vtxDesc->type) {
+        case GX_NONE:
+            break;
+        case GX_DIRECT:
+            if (!is_matrix_idx_attr(vtxDesc->attr)) {
+                return false;
+            }
+            stride += 1;
+            break;
+        case GX_INDEX8:
+            stride += 1;
+            break;
+        case GX_INDEX16:
+            stride += 2;
+            break;
+        default:
+            return false;
+        }
+    }
+    return stride != 0;
+}
+
+bool get_command_size(const u8* dlStart, u32 dlSize, u32 offset, u32 stride, u32& cmdSize) {
+    if (offset >= dlSize) {
+        return false;
+    }
+
+    const u8 cmd = dlStart[offset];
+    const u8 opcode = cmd & GX_OPCODE_MASK;
+    switch (opcode) {
+    case GX_NOP:
+    case GX_CMD_INVL_VC:
+        cmdSize = 1;
+        return true;
+    case (GX_LOAD_BP_REG & GX_OPCODE_MASK):
+        cmdSize = 5;
+        return offset + cmdSize <= dlSize;
+    case GX_LOAD_CP_REG:
+        cmdSize = 6;
+        return offset + cmdSize <= dlSize;
+    case GX_LOAD_XF_REG: {
+        if (offset + 5 > dlSize) {
+            return false;
+        }
+        const u16 count = read_be16(dlStart + offset + 1) + 1;
+        cmdSize = 5 + count * 4;
+        return offset + cmdSize <= dlSize;
+    }
+    case GX_LOAD_INDX_A:
+    case GX_LOAD_INDX_B:
+    case GX_LOAD_INDX_C:
+    case GX_LOAD_INDX_D:
+        cmdSize = 5;
+        return offset + cmdSize <= dlSize;
+    case GX_CMD_CALL_DL:
+        cmdSize = 9;
+        return offset + cmdSize <= dlSize;
+    default:
+        if (is_draw_opcode(opcode)) {
+            if (offset + 3 > dlSize) {
+                return false;
+            }
+            const u16 vtxCount = read_be16(dlStart + offset + 1);
+            cmdSize = 3 + vtxCount * stride;
+            return offset + cmdSize <= dlSize;
+        }
+        return false;
+    }
+}
+
+struct MergeRun {
+    u8 cmd = 0;
+    u16 vtxCount = 0;
+    std::vector<u8> vertices;
+};
+
+void flush_merge_run(std::vector<u8>& out, MergeRun& run) {
+    if (run.vtxCount == 0) {
+        return;
+    }
+
+    out.push_back(run.cmd);
+    append_be16(out, run.vtxCount);
+    append_bytes(out, run.vertices.data(), run.vertices.size());
+    run.vertices.clear();
+    run.vtxCount = 0;
+}
+
+void append_vertex(std::vector<u8>& out, const u8* vertices, u32 stride, u16 idx) {
+    append_bytes(out, vertices + idx * stride, stride);
+}
+
+bool triangulate_draw(
+    std::vector<u8>& out, u8 opcode, const u8* vertices, u32 stride, u16 vtxCount) {
+    switch (opcode) {
+    case GX_TRIANGLES:
+        append_bytes(out, vertices, vtxCount * stride);
+        return true;
+    case GX_TRIANGLEFAN:
+        if (vtxCount < 3) {
+            return false;
+        }
+        for (u16 v = 2; v < vtxCount; v++) {
+            append_vertex(out, vertices, stride, 0);
+            append_vertex(out, vertices, stride, v - 1);
+            append_vertex(out, vertices, stride, v);
+        }
+        return true;
+    case GX_TRIANGLESTRIP:
+        if (vtxCount < 3) {
+            return false;
+        }
+        for (u16 v = 2; v < vtxCount; v++) {
+            if ((v & 1) == 0) {
+                append_vertex(out, vertices, stride, v - 2);
+                append_vertex(out, vertices, stride, v - 1);
+            } else {
+                append_vertex(out, vertices, stride, v - 1);
+                append_vertex(out, vertices, stride, v - 2);
+            }
+            append_vertex(out, vertices, stride, v);
+        }
+        return true;
+    case GX_QUADS:
+        if ((vtxCount & 3) != 0) {
+            return false;
+        }
+        for (u16 v = 0; v < vtxCount; v += 4) {
+            append_vertex(out, vertices, stride, v);
+            append_vertex(out, vertices, stride, v + 1);
+            append_vertex(out, vertices, stride, v + 2);
+            append_vertex(out, vertices, stride, v + 2);
+            append_vertex(out, vertices, stride, v + 3);
+            append_vertex(out, vertices, stride, v);
+        }
+        return true;
+    default:
+        return false;
+    }
+}
+
+void append_triangles_to_run(
+    std::vector<u8>& out, MergeRun& run, u8 cmd, const std::vector<u8>& vertices, u32 stride) {
+    u32 offset = 0;
+    u32 remaining = vertices.size() / stride;
+    while (remaining != 0) {
+        if (run.vtxCount != 0 && run.cmd != cmd) {
+            flush_merge_run(out, run);
+        }
+
+        if (run.vtxCount == 0) {
+            run.cmd = cmd;
+        }
+
+        u32 available = 0xFFFF - run.vtxCount;
+        if (available == 0) {
+            flush_merge_run(out, run);
+            continue;
+        }
+
+        u32 toCopy = std::min(remaining, available);
+        append_bytes(run.vertices, vertices.data() + offset * stride, toCopy * stride);
+        run.vtxCount += toCopy;
+        offset += toCopy;
+        remaining -= toCopy;
+
+        if (run.vtxCount == 0xFFFF) {
+            flush_merge_run(out, run);
+        }
+    }
+}
+
+bool optimize_display_list(const u8* dlStart, u32 dlSize, u32 stride, std::vector<u8>& out) {
+    MergeRun run;
+    out.reserve(dlSize);
+
+    for (u32 offset = 0; offset < dlSize;) {
+        u32 cmdSize = 0;
+        if (!get_command_size(dlStart, dlSize, offset, stride, cmdSize)) {
+            return false;
+        }
+
+        const u8 cmd = dlStart[offset];
+        const u8 opcode = cmd & GX_OPCODE_MASK;
+        if (opcode == GX_NOP) {
+            offset += cmdSize;
+            continue;
+        }
+
+        if (!is_draw_opcode(opcode)) {
+            flush_merge_run(out, run);
+            append_bytes(out, dlStart + offset, cmdSize);
+            offset += cmdSize;
+            continue;
+        }
+
+        if (!is_mergeable_draw_opcode(opcode)) {
+            flush_merge_run(out, run);
+            append_bytes(out, dlStart + offset, cmdSize);
+            offset += cmdSize;
+            continue;
+        }
+
+        const u16 vtxCount = read_be16(dlStart + offset + 1);
+        const u8* vertices = dlStart + offset + 3;
+        std::vector<u8> triangles;
+        if (!triangulate_draw(triangles, opcode, vertices, stride, vtxCount)) {
+            flush_merge_run(out, run);
+            append_bytes(out, dlStart + offset, cmdSize);
+            offset += cmdSize;
+            continue;
+        }
+
+        append_triangles_to_run(out, run, (GX_TRIANGLES | (cmd & GX_VAT_MASK)), triangles, stride);
+        offset += cmdSize;
+    }
+
+    flush_merge_run(out, run);
+    return true;
+}
+
+void set_display_list_copy(void*& displayList, u32& displayListSize, const u8* data, u32 size) {
+    const u32 alignedSize = ALIGN_NEXT(size, 0x20);
+    u8* newDL = JKR_NEW_ARRAY_ARGS(u8, alignedSize, 0x20);
+    if (size != 0) {
+        std::memcpy(newDL, data, size);
+    }
+    for (u32 i = size; i < alignedSize; i++) {
+        newDL[i] = 0;
+    }
+
+    displayList = newDL;
+    displayListSize = alignedSize;
+    DCStoreRange(newDL, displayListSize);
+}
+
+}  // namespace
+#endif
 
 u32 J3DShapeDraw::countVertex(u32 stride) {
     u32 count = 0;
     u8* dlStart = (u8*)getDisplayList();
 
+#if TARGET_PC
+    for (u32 offset = 0; offset < getDisplayListSize();) {
+        u8 cmd = dlStart[offset];
+        u8 opcode = cmd & GX_OPCODE_MASK;
+        u32 cmdSize = 0;
+        if (!get_command_size(dlStart, getDisplayListSize(), offset, stride, cmdSize)) {
+            break;
+        }
+        if (!is_draw_opcode(opcode)) {
+            offset += cmdSize;
+            continue;
+        }
+        int vtxNum = be16(*reinterpret_cast<u16*>(dlStart + offset + 1));
+        count += vtxNum;
+        offset += 3 + stride * vtxNum;
+    }
+#else
     for (u8* dl = dlStart; (dl - dlStart) < getDisplayListSize();) {
         u8 cmd = *(u8*)dl;
         dl++;
@@ -20,6 +315,7 @@ u32 J3DShapeDraw::countVertex(u32 stride) {
         count += vtxNum;
         dl = (u8*)dl + stride * vtxNum;
     }
+#endif
 
     return count;
 }
@@ -34,13 +330,32 @@ void J3DShapeDraw::addTexMtxIndexInDL(u32 stride, u32 attrOffs, u32 valueBase) {
     u8* newDL = newDLStart;
 
     for (; (oldDL - oldDLStart) < mDisplayListSize;) {
+#if TARGET_PC
+        u32 oldOffset = oldDL - oldDLStart;
+        u32 cmdSize = 0;
+        if (!get_command_size(oldDLStart, mDisplayListSize, oldOffset, stride, cmdSize)) {
+            memcpy(newDL, oldDL, mDisplayListSize - oldOffset);
+            newDL += mDisplayListSize - oldOffset;
+            break;
+        }
+#endif
         // Copy command
         u8 cmd = *(u8*)oldDL;
         oldDL++;
         *newDL++ = cmd;
 
+#if TARGET_PC
+        u8 opcode = cmd & GX_OPCODE_MASK;
+        if (!is_draw_opcode(opcode)) {
+            memcpy(newDL, oldDL, cmdSize - 1);
+            oldDL += cmdSize - 1;
+            newDL += cmdSize - 1;
+            continue;
+        }
+#else
         if (cmd != GX_TRIANGLEFAN && cmd != GX_TRIANGLESTRIP)
             break;
+#endif
 
         // Copy count
         int vtxNum = *(u16*)oldDL;
@@ -71,11 +386,31 @@ void J3DShapeDraw::addTexMtxIndexInDL(u32 stride, u32 attrOffs, u32 valueBase) {
 }
 
 J3DShapeDraw::J3DShapeDraw(const u8* displayList, u32 displayListSize) {
+#if TARGET_PC
+    set_display_list_copy(mDisplayList, mDisplayListSize, displayList, displayListSize);
+#else
     mDisplayList = (void*)displayList;
     mDisplayListSize = displayListSize;
+#endif
 }
 
+#if TARGET_PC
+J3DShapeDraw::J3DShapeDraw(
+    const u8* displayList, u32 displayListSize, const GXVtxDescList* vtxDesc) {
+    u32 stride = 0;
+    std::vector<u8> optimized;
+    if (calc_vtx_stride(vtxDesc, stride) &&
+        optimize_display_list(displayList, displayListSize, stride, optimized))
+    {
+        set_display_list_copy(mDisplayList, mDisplayListSize, optimized.data(), optimized.size());
+    } else {
+        set_display_list_copy(mDisplayList, mDisplayListSize, displayList, displayListSize);
+    }
+}
+#endif
+
 void J3DShapeDraw::draw() const {
+    ZoneScoped;
     GXCallDisplayList(mDisplayList, mDisplayListSize);
 }
 

libs/JSystem/src/J3DGraphLoader/J3DShapeFactory.cpp

@@ -132,7 +132,12 @@ J3DShapeDraw* J3DShapeFactory::newShapeDraw(int shapeNo, int mtxGroupNo) const {
     const J3DShapeInitData& shapeInitData = mShapeInitData[mIndexTable[shapeNo]];
     const J3DShapeDrawInitData& drawInitData =
         (&mDrawInitData[shapeInitData.mDrawInitDataIndex])[mtxGroupNo];
+#if TARGET_PC
+    shapeDraw = JKR_NEW J3DShapeDraw(&mDisplayListData[drawInitData.mDisplayListIndex], drawInitData.mDisplayListSize,
+                                     getVtxDescList(shapeNo));
+#else
     shapeDraw = JKR_NEW J3DShapeDraw(&mDisplayListData[drawInitData.mDisplayListIndex], drawInitData.mDisplayListSize);
+#endif
     J3D_ASSERT_ALLOCMEM(193, shapeDraw);
     return shapeDraw;
 }
@@ -154,7 +159,7 @@ s32 J3DShapeFactory::calcSize(int shapeNo, u32 flag) {
 
     for (u32 i = 0; i < mtxGroupNo; i++) {
         size += calcSizeShapeMtx(flag, shapeNo, i);
-        size += 0x0C;
+        size += sizeof(J3DShapeDraw);
     }
 
     return size;