[GPU] Add log_draws: dump per-draw vertex declaration for RE

Reverse-engineering aid for decoding game mesh formats against GPU ground
truth. When the `log_draws` cvar is set, each distinct draw's primitive
type, index buffer (guest base / count / format / endianness), and full
per-stream vertex declaration (fetch-constant base + stride, and every
element's format + offset) is written to xenia_re_draws.log.

- command_processor.{h,cc}: CommandProcessor::LogDrawForRE(), no-op unless
  the cvar is set. De-dups by the vertex-declaration fingerprint (shader +
  primitive + element formats/offsets), so animated UI that redraws the
  same format into fresh buffers every frame collapses to one record --
  keeping it near-free (an earlier address-keyed de-dup flooded the log and
  stalled the GPU thread). Capped at 4096 distinct formats.
- pm4_command_processor_implement.h: call it from ExecutePacketType3Draw
  after IssueDraw (so the vertex shader has been analyzed). Backend-agnostic
  base path -- works for the Vulkan build.

Used to confirm Project Sylpheed's XBG7 mesh layout (triangle list, pos
f32x3 / normal f16x4 / uv f16x2, variable stride).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/xenia/gpu/command_processor.cc

@@ -9,6 +9,9 @@
 
 #include "xenia/gpu/command_processor.h"
 
+#include <fstream>
+#include <unordered_set>
+
 #include "third_party/fmt/include/fmt/format.h"
 #include "xenia/base/byte_stream.h"
 #include "xenia/base/clock.h"
@@ -18,6 +21,8 @@
 #include "xenia/gpu/gpu_flags.h"
 #include "xenia/gpu/graphics_system.h"
 #include "xenia/gpu/packet_disassembler.h"
+#include "xenia/gpu/registers.h"
+#include "xenia/gpu/shader.h"
 #include "xenia/gpu/sampler_info.h"
 #include "xenia/gpu/texture_info.h"
 #include "xenia/gpu/xenos_zpd_report.h"
@@ -77,6 +82,14 @@ DEFINE_string(
 
 UPDATE_from_string(readback_resolve, 2025, 12, 4, 21, "fast");
 
+DEFINE_bool(
+    log_draws, false,
+    "Reverse-engineering aid: write each distinct draw's primitive type, index "
+    "buffer, and per-stream vertex declaration (stream base/stride + per-element "
+    "format/offset) to xenia_re_draws.log in the working directory. For decoding "
+    "game mesh formats against GPU ground truth.",
+    "GPU");
+
 DEFINE_bool(
     readback_memexport, false,
     "Read data written by memory export in shaders on the CPU. "
@@ -88,6 +101,126 @@ DEFINE_bool(
 namespace xe {
 namespace gpu {
 
+namespace {
+// Short readable name for a guest vertex element format (RE logging only).
+const char* ReVertexFormatName(xenos::VertexFormat f) {
+  switch (f) {
+    case xenos::VertexFormat::k_32_FLOAT:          return "f32";
+    case xenos::VertexFormat::k_32_32_FLOAT:       return "f32x2";
+    case xenos::VertexFormat::k_32_32_32_FLOAT:    return "f32x3";
+    case xenos::VertexFormat::k_32_32_32_32_FLOAT: return "f32x4";
+    case xenos::VertexFormat::k_16_16_FLOAT:       return "f16x2";
+    case xenos::VertexFormat::k_16_16_16_16_FLOAT: return "f16x4";
+    case xenos::VertexFormat::k_16_16:             return "s16x2";
+    case xenos::VertexFormat::k_16_16_16_16:       return "s16x4";
+    case xenos::VertexFormat::k_8_8_8_8:           return "8888";
+    case xenos::VertexFormat::k_2_10_10_10:        return "2_10_10_10";
+    case xenos::VertexFormat::k_10_11_11:          return "10_11_11";
+    case xenos::VertexFormat::k_11_11_10:          return "11_11_10";
+    case xenos::VertexFormat::k_32:                return "u32";
+    case xenos::VertexFormat::k_32_32:             return "u32x2";
+    case xenos::VertexFormat::k_32_32_32_32:       return "u32x4";
+    default:                                       return "?";
+  }
+}
+}  // namespace
+
+void CommandProcessor::LogDrawForRE(uint32_t vgt_draw_initiator_value,
+                                    const IndexBufferInfo* index_buffer_info) {
+  if (!cvars::log_draws) {
+    return;
+  }
+  static std::mutex re_mutex;
+  static std::ofstream re_out;
+  static std::unordered_set<uint64_t> re_seen;
+  std::lock_guard<std::mutex> lock(re_mutex);
+  if (!re_out.is_open()) {
+    re_out.open("xenia_re_draws.log", std::ios::out | std::ios::trunc);
+    XELOGI("[RE-DRAW] logging distinct draws to xenia_re_draws.log");
+  }
+  if (!re_out.is_open()) {
+    return;
+  }
+
+  reg::VGT_DRAW_INITIATOR init;
+  init.value = vgt_draw_initiator_value;
+  Shader* vs = active_vertex_shader_;
+
+  // De-dup by the VERTEX-DECLARATION FINGERPRINT (shader + primitive type +
+  // per-stream element formats/offsets), NOT by buffer address. Animated UI
+  // that redraws the same mesh format into fresh buffers every frame therefore
+  // collapses to a single record, keeping the logging near-free — while every
+  // distinct mesh format (the player plane, each weapon) is still captured once.
+  uint64_t sig = 1469598103934665603ull;  // FNV-ish seed
+  auto mix = [&sig](uint64_t v) { sig = (sig ^ v) * 1099511628211ull; };
+  mix(uint64_t(init.prim_type));
+  if (vs) {
+    mix(vs->ucode_data_hash());
+  }
+  bool analyzed = vs && vs->is_ucode_analyzed();
+  if (analyzed) {
+    for (const auto& binding : vs->vertex_bindings()) {
+      mix(binding.fetch_constant);
+      mix(binding.stride_words);
+      for (const auto& attr : binding.attributes) {
+        mix(uint64_t(attr.fetch_instr.attributes.data_format));
+        mix(uint64_t(uint32_t(attr.fetch_instr.attributes.offset)));
+      }
+    }
+  } else {
+    mix(uint64_t(init.num_indices));
+  }
+  if (!re_seen.insert(sig).second) {
+    return;
+  }
+  // Safety cap on distinct formats, so a pathological title can't grow the log
+  // (and the working set) without bound.
+  if (re_seen.size() > 4096) {
+    return;
+  }
+
+  re_out << fmt::format("DRAW prim={} indices={} src={} ",
+                        uint32_t(init.prim_type), uint32_t(init.num_indices),
+                        uint32_t(init.source_select));
+  if (index_buffer_info) {
+    re_out << fmt::format(
+        "index[base=0x{:08X} count={} fmt={} endian={} len={}] ",
+        index_buffer_info->guest_base, index_buffer_info->count,
+        index_buffer_info->format == xenos::IndexFormat::kInt16 ? "u16" : "u32",
+        uint32_t(index_buffer_info->endianness), index_buffer_info->length);
+  } else {
+    re_out << "index[auto] ";
+  }
+  if (vs) {
+    re_out << fmt::format("vs=0x{:016X}", vs->ucode_data_hash());
+  }
+  re_out << "\n";
+
+  if (analyzed) {
+    for (const auto& binding : vs->vertex_bindings()) {
+      xenos::xe_gpu_vertex_fetch_t fetch =
+          register_file_->GetVertexFetch(binding.fetch_constant);
+      re_out << fmt::format(
+          "  stream fc={} base=0x{:08X} stride_words={} size_words={} "
+          "endian={} type={}\n",
+          binding.fetch_constant, uint32_t(fetch.address) << 2,
+          binding.stride_words, uint32_t(fetch.size), uint32_t(fetch.endian),
+          uint32_t(fetch.type));
+      for (const auto& attr : binding.attributes) {
+        const auto& a = attr.fetch_instr.attributes;
+        re_out << fmt::format(
+            "    attr fmt={}({}) offset_words={} stride_words={} signed={} "
+            "int={} exp_adjust={}\n",
+            uint32_t(a.data_format), ReVertexFormatName(a.data_format), a.offset,
+            a.stride, a.is_signed ? 1 : 0, a.is_integer ? 1 : 0, a.exp_adjust);
+      }
+    }
+  } else {
+    re_out << "  (vertex shader not analyzed yet)\n";
+  }
+  re_out.flush();
+}
+
 // This should be written completely differently with support for different
 // types.
 void SaveGPUSetting(GPUSetting setting, uint64_t value) {

src/xenia/gpu/command_processor.h

@@ -446,6 +446,13 @@ class CommandProcessor {
   }
   virtual bool IssueCopy() { return false; }
 
+  // Reverse-engineering aid (cvar `log_draws`): dump the guest's exact
+  // primitive type + index buffer + per-stream vertex declaration for each
+  // distinct draw to a dedicated file, for decoding game mesh formats.
+  // No-op unless the cvar is enabled. Defined in command_processor.cc.
+  void LogDrawForRE(uint32_t vgt_draw_initiator_value,
+                    const IndexBufferInfo* index_buffer_info);
+
   // "Actual" is for the command processor thread, to be read by the
   // implementations.
   SwapPostEffect GetActualSwapPostEffect() const {

src/xenia/gpu/pm4_command_processor_implement.h

@@ -1152,6 +1152,11 @@ bool COMMAND_PROCESSOR::ExecutePacketType3Draw(
                uint32_t(vgt_draw_initiator.prim_type),
                uint32_t(vgt_draw_initiator.source_select));
       }
+      // Reverse-engineering aid (no-op unless the `log_draws` cvar is set):
+      // record the guest's primitive type + index buffer + vertex declaration.
+      // Placed after IssueDraw so the vertex shader has been analyzed.
+      COMMAND_PROCESSOR::LogDrawForRE(
+          vgt_draw_initiator.value, is_indexed ? &index_buffer_info : nullptr);
     }
   }