nv2a: Bump vmstate version for new NV2A_MAX_BATCH_LENGTH

build.sh

@@ -66,8 +66,14 @@ package_macos() {
 
     cp Info.plist dist/xemu.app/Contents/
 
-    plutil -replace CFBundleShortVersionString -string $(cat ${project_source_dir}/XEMU_VERSION | cut -f1 -d-) dist/xemu.app/Contents/Info.plist
-    plutil -replace CFBundleVersion            -string $(cat ${project_source_dir}/XEMU_VERSION | cut -f1 -d-) dist/xemu.app/Contents/Info.plist
+    if [[ -e "${project_source_dir}/XEMU_VERSION" ]]; then
+      xemu_version="$(cat ${project_source_dir}/XEMU_VERSION | cut -f1 -d-)"
+    else
+      xemu_version="0.0.0"
+    fi
+
+    plutil -replace CFBundleShortVersionString -string "${xemu_version}" dist/xemu.app/Contents/Info.plist
+    plutil -replace CFBundleVersion            -string "${xemu_version}" dist/xemu.app/Contents/Info.plist
 
     codesign --force --deep --preserve-metadata=entitlements,requirements,flags,runtime --sign - "${exe_path}"
     python3 ./scripts/gen-license.py --version-file=macos-libs/$target_arch/INSTALLED > dist/LICENSE.txt

debian/control

@@ -6,6 +6,8 @@ Build-Depends: debhelper (>= 11),
  cmake,
  git,
  python3:any,
+ python3-pip,
+ python3-tomli,
  python3-yaml,
  python3-venv,
  ninja-build,

hw/xbox/nv2a/nv2a.c

@@ -423,7 +423,7 @@ const VMStateDescription vmstate_nv2a_pgraph_vertex_attributes = {
 
 static const VMStateDescription vmstate_nv2a = {
     .name = "nv2a",
-    .version_id = 2,
+    .version_id = 3,
     .minimum_version_id = 1,
     .post_save = nv2a_post_save,
     .post_load = nv2a_post_load,
@@ -507,9 +507,11 @@ static const VMStateDescription vmstate_nv2a = {
         VMSTATE_BOOL_ARRAY(pgraph.ltc1_dirty, NV2AState, NV2A_LTC1_COUNT),
         VMSTATE_STRUCT_ARRAY(pgraph.vertex_attributes, NV2AState, NV2A_VERTEXSHADER_ATTRIBUTES, 1, vmstate_nv2a_pgraph_vertex_attributes, VertexAttribute),
         VMSTATE_UINT32(pgraph.inline_array_length, NV2AState),
-        VMSTATE_UINT32_ARRAY(pgraph.inline_array, NV2AState, NV2A_MAX_BATCH_LENGTH),
+        VMSTATE_UINT32_SUB_ARRAY(pgraph.inline_array, NV2AState, 0, NV2A_MAX_BATCH_LENGTH_V2),
+        VMSTATE_UINT32_SUB_ARRAY_V(pgraph.inline_array, NV2AState, NV2A_MAX_BATCH_LENGTH_V2, NV2A_MAX_BATCH_LENGTH - NV2A_MAX_BATCH_LENGTH_V2, 3),
         VMSTATE_UINT32(pgraph.inline_elements_length, NV2AState), // fixme
-        VMSTATE_UINT32_ARRAY(pgraph.inline_elements, NV2AState, NV2A_MAX_BATCH_LENGTH),
+        VMSTATE_UINT32_SUB_ARRAY(pgraph.inline_elements, NV2AState, 0, NV2A_MAX_BATCH_LENGTH_V2),
+        VMSTATE_UINT32_SUB_ARRAY_V(pgraph.inline_elements, NV2AState, NV2A_MAX_BATCH_LENGTH_V2, NV2A_MAX_BATCH_LENGTH - NV2A_MAX_BATCH_LENGTH_V2, 3),
         VMSTATE_UINT32(pgraph.inline_buffer_length, NV2AState), // fixme
         VMSTATE_UINT32(pgraph.draw_arrays_length, NV2AState),
         VMSTATE_UINT32(pgraph.draw_arrays_max_count, NV2AState),

hw/xbox/nv2a/nv2a_regs.h

@@ -315,11 +315,14 @@
 #define NV_PGRAPH_CSV0_C                                 0x00000FB8
 #   define NV_PGRAPH_CSV0_C_CHEOPS_PROGRAM_START                0x0000FF00
 #   define NV_PGRAPH_CSV0_C_SPECULAR_ENABLE                     (1 << 16)
+#   define NV_PGRAPH_CSV0_C_ALPHA_FROM_MATERIAL_SPECULAR        (1 << 17)
+#   define NV_PGRAPH_CSV0_C_SEPARATE_SPECULAR                   (1 << 18)
 #   define NV_PGRAPH_CSV0_C_SPECULAR                            (3 << 19)
 #   define NV_PGRAPH_CSV0_C_DIFFUSE                             (3 << 21)
 #   define NV_PGRAPH_CSV0_C_AMBIENT                             (3 << 23)
 #   define NV_PGRAPH_CSV0_C_EMISSION                            (3 << 25)
 #   define NV_PGRAPH_CSV0_C_NORMALIZATION_ENABLE                (1 << 27)
+#   define NV_PGRAPH_CSV0_C_LOCALEYE                            (1 << 30)
 #   define NV_PGRAPH_CSV0_C_LIGHTING                            (1 << 31)
 #define NV_PGRAPH_CSV1_B                                 0x00000FBC
 #define NV_PGRAPH_CSV1_A                                 0x00000FC0
@@ -882,6 +885,10 @@
 #       define NV097_SET_CONTROL0_STENCIL_WRITE_ENABLE            (1 << 0)
 #       define NV097_SET_CONTROL0_Z_FORMAT                        (1 << 12)
 #       define NV097_SET_CONTROL0_Z_PERSPECTIVE_ENABLE            (1 << 16)
+#   define NV097_SET_LIGHT_CONTROL                            0x00000294
+#       define NV097_SET_LIGHT_CONTROL_SEPARATE_SPECULAR          1
+#       define NV097_SET_LIGHT_CONTROL_LOCALEYE                   (1 << 16)
+#       define NV097_SET_LIGHT_CONTROL_ALPHA_FROM_MATERIAL_SPECULAR (1 << 17)
 #   define NV097_SET_COLOR_MATERIAL                           0x00000298
 #   define NV097_SET_FOG_MODE                                 0x0000029C
 #       define NV097_SET_FOG_MODE_V_LINEAR                        0x2601
@@ -1049,6 +1056,7 @@
 #       define NV097_SET_TEXGEN_VIEW_MODEL_LOCAL_VIEWER           0
 #       define NV097_SET_TEXGEN_VIEW_MODEL_INFINITE_VIEWER        1
 #   define NV097_SET_FOG_PLANE                                0x000009D0
+#   define NV097_SET_SPECULAR_PARAMS                          0x000009E0
 #   define NV097_SET_SCENE_AMBIENT_COLOR                      0x00000A10
 #   define NV097_SET_VIEWPORT_OFFSET                          0x00000A20
 #   define NV097_SET_POINT_PARAMS                             0x00000A30
@@ -1099,6 +1107,11 @@
 #   define NV097_SET_TEXCOORD3_4F                             0x00001620
 #   define NV097_SET_TEXCOORD3_2S                             0x00001610
 #   define NV097_SET_TEXCOORD3_4S                             0x00001630
+#   define NV097_SET_FOG_COORD                                0x00001698
+#   define NV097_SET_WEIGHT1F                                 0x0000169C
+#   define NV097_SET_WEIGHT2F                                 0x000016A0
+#   define NV097_SET_WEIGHT3F                                 0x000016B0
+#   define NV097_SET_WEIGHT4F                                 0x000016C0
 #   define NV097_SET_VERTEX_DATA_ARRAY_OFFSET                 0x00001720
 #   define NV097_SET_VERTEX_DATA_ARRAY_FORMAT                 0x00001760
 #       define NV097_SET_VERTEX_DATA_ARRAY_FORMAT_TYPE            0x0000000F
@@ -1252,6 +1265,7 @@
 #   define NV097_SET_CLEAR_RECT_HORIZONTAL                    0x00001D98
 #   define NV097_SET_CLEAR_RECT_VERTICAL                      0x00001D9C
 #   define NV097_SET_SPECULAR_FOG_FACTOR                      0x00001E20
+#   define NV097_SET_SPECULAR_PARAMS_BACK                     0x00001E28
 #   define NV097_SET_COMBINER_COLOR_OCW                       0x00001E40
 #   define NV097_SET_COMBINER_CONTROL                         0x00001E60
 #   define NV097_SET_SHADOW_ZSLOPE_THRESHOLD                  0x00001E68
@@ -1453,7 +1467,22 @@
 #define NV2A_NUM_SUBCHANNELS 8
 #define NV2A_CACHE1_SIZE 128
 
-#define NV2A_MAX_BATCH_LENGTH 0x1FFFF
+/* This is a multi-use limit. Testing on an Xbox 1.0, it is possible to send
+ * arrays of at least 0x0FFFFF elements without issue, however sending
+ * NV097_DRAW_ARRAYS with a start value > 0xFFFF raises an exception implying
+ * that there may be a vertex limit. Since xemu uses batch length for vertex
+ * elements in NV097_INLINE_ARRAY the size should ideally be high enough to
+ * accommodate 0xFFFF vertices with maximum attributes specified.
+ *
+ * Retail games are known to send at least 0x410FA elements in a single draw, so
+ * a somewhat larger value is selected to balance memory use with real-world
+ * limits.
+ *
+ * NV2A_MAX_BATCH_LENGTH_V2 is the previous limit, for migration.
+ * FIXME: Remove NV2A_MAX_BATCH_LENGTH_V2 at some point in the future.
+ */
+#define NV2A_MAX_BATCH_LENGTH 0x07FFFF
+#define NV2A_MAX_BATCH_LENGTH_V2 0x1FFFF
 #define NV2A_VERTEXSHADER_ATTRIBUTES 16
 #define NV2A_MAX_TEXTURES 4
 

hw/xbox/nv2a/pgraph/gl/constants.h

@@ -298,7 +298,7 @@ static const SurfaceFormatInfo kelvin_surface_color_format_gl_map[] = {
     [NV097_SET_SURFACE_FORMAT_COLOR_LE_B8] =
         {1, GL_R8, GL_RED, GL_UNSIGNED_BYTE, GL_COLOR_ATTACHMENT0},
     [NV097_SET_SURFACE_FORMAT_COLOR_LE_G8B8] =
-        {2, GL_RG8, GL_RG, GL_UNSIGNED_SHORT, GL_COLOR_ATTACHMENT0},
+        {2, GL_RG8, GL_RG, GL_UNSIGNED_BYTE, GL_COLOR_ATTACHMENT0},
 };
 
 static const SurfaceFormatInfo kelvin_surface_zeta_float_format_gl_map[] = {

hw/xbox/nv2a/pgraph/gl/renderer.h

@@ -122,6 +122,7 @@ typedef struct ShaderBinding {
     GLint light_infinite_direction_loc[NV2A_MAX_LIGHTS];
     GLint light_local_position_loc[NV2A_MAX_LIGHTS];
     GLint light_local_attenuation_loc[NV2A_MAX_LIGHTS];
+    int specular_power_loc;
 
     GLint clip_region_loc[8];
 

hw/xbox/nv2a/pgraph/gl/shaders.c

@@ -193,8 +193,11 @@ static void update_shader_constant_locations(ShaderBinding *binding)
     if (binding->state.fixed_function) {
         binding->material_alpha_loc =
             glGetUniformLocation(binding->gl_program, "material_alpha");
+        binding->specular_power_loc =
+            glGetUniformLocation(binding->gl_program, "specularPower");
     } else {
         binding->material_alpha_loc = -1;
+        binding->specular_power_loc = -1;
     }
 }
 
@@ -836,6 +839,10 @@ static void shader_update_constants(PGRAPHState *pg, ShaderBinding *binding,
             }
         }
 
+        if (binding->specular_power_loc != -1) {
+    	    glUniform1f(binding->specular_power_loc, pg->specular_power);
+	    }
+
         /* estimate the viewport by assuming it matches the surface ... */
         unsigned int aa_width = 1, aa_height = 1;
         pgraph_apply_anti_aliasing_factor(pg, &aa_width, &aa_height);

hw/xbox/nv2a/pgraph/glsl/vsh-ff.c

@@ -230,10 +230,16 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
     }
 
     /* Lighting */
-    if (state->lighting) {
-
+    if (!state->lighting) {
+        mstring_append(body, "  oD0 = diffuse;\n");
+        mstring_append(body, "  oD1 = specular;\n");
+        mstring_append(body, "  oB0 = backDiffuse;\n");
+        mstring_append(body, "  oB1 = backSpecular;\n");
+    } else {
         //FIXME: Do 2 passes if we want 2 sided-lighting?
 
+        mstring_append_fmt(uniforms, "%sfloat specularPower;\n", u);
+
         static char alpha_source_diffuse[] = "diffuse.a";
         static char alpha_source_specular[] = "specular.a";
         static char alpha_source_material[] = "material_alpha";
@@ -264,17 +270,17 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
 
         mstring_append(body, "oD1 = vec4(0.0, 0.0, 0.0, specular.a);\n");
 
+        if (state->local_eye) {
+            mstring_append(body,
+                "vec3 VPeye = normalize(eyePosition.xyz / eyePosition.w - tPosition.xyz / tPosition.w);\n"
+            );
+        }
+
         for (i = 0; i < NV2A_MAX_LIGHTS; i++) {
             if (state->light[i] == LIGHT_OFF) {
                 continue;
             }
 
-            /* FIXME: It seems that we only have to handle the surface colors if
-             *        they are not part of the material [= vertex colors].
-             *        If they are material the cpu will premultiply light
-             *        colors
-             */
-
             mstring_append_fmt(body, "/* Light %d */ {\n", i);
 
             if (state->light[i] == LIGHT_LOCAL
@@ -285,18 +291,20 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
                     "%svec3 lightLocalAttenuation%d;\n",
                     u, i, u, i);
                 mstring_append_fmt(body,
-                    "  vec3 VP = lightLocalPosition%d - tPosition.xyz/tPosition.w;\n"
+                    "  vec3 tPos = tPosition.xyz/tPosition.w;\n"
+                    "  vec3 VP = lightLocalPosition%d - tPos;\n"
                     "  float d = length(VP);\n"
-//FIXME: if (d > lightLocalRange) { .. don't process this light .. } /* inclusive?! */ - what about directional lights?
-                    "  VP = normalize(VP);\n"
-                    "  float attenuation = 1.0 / (lightLocalAttenuation%d.x\n"
-                    "                               + lightLocalAttenuation%d.y * d\n"
-                    "                               + lightLocalAttenuation%d.z * d * d);\n"
-                    "  vec3 halfVector = normalize(VP + eyePosition.xyz / eyePosition.w);\n" /* FIXME: Not sure if eyePosition is correct */
-                    "  float nDotVP = max(0.0, dot(tNormal, VP));\n"
-                    "  float nDotHV = max(0.0, dot(tNormal, halfVector));\n",
-                    i, i, i, i);
-
+                    "  if (d <= lightLocalRange(%d)) {\n"  /* FIXME: Double check that range is inclusive */
+                    "    VP = normalize(VP);\n"
+                    "    float attenuation = 1.0 / (lightLocalAttenuation%d.x\n"
+                    "                                 + lightLocalAttenuation%d.y * d\n"
+                    "                                 + lightLocalAttenuation%d.z * d * d);\n"
+                    "    vec3 halfVector = normalize(VP + %s);\n"
+                    "    float nDotVP = max(0.0, dot(tNormal, VP));\n"
+                    "    float nDotHV = max(0.0, dot(tNormal, halfVector));\n",
+                    i, i, i, i, i,
+                    state->local_eye ? "VPeye" : "vec3(0.0, 0.0, 0.0)"
+                );
             }
 
             switch(state->light[i]) {
@@ -309,15 +317,21 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
                     "%svec3 lightInfiniteDirection%d;\n",
                     u, i, u, i);
                 mstring_append_fmt(body,
-                    "  float attenuation = 1.0;\n"
-                    "  float nDotVP = max(0.0, dot(tNormal, normalize(vec3(lightInfiniteDirection%d))));\n"
-                    "  float nDotHV = max(0.0, dot(tNormal, vec3(lightInfiniteHalfVector%d)));\n",
-                    i, i);
-
-                /* FIXME: Do specular */
-
-                /* FIXME: tBackDiffuse */
-
+                    "  {\n"
+                    "    float attenuation = 1.0;\n"
+                    "    vec3 lightDirection = normalize(lightInfiniteDirection%d);\n"
+                    "    float nDotVP = max(0.0, dot(tNormal, lightDirection));\n",
+                    i);
+                if (state->local_eye) {
+                    mstring_append(body,
+                        "    float nDotHV = max(0.0, dot(tNormal, normalize(lightDirection + VPeye)));\n"
+                    );
+                } else {
+                    mstring_append_fmt(body,
+                        "    float nDotHV = max(0.0, dot(tNormal, lightInfiniteHalfVector%d));\n",
+                        i
+                    );
+                }
                 break;
             case LIGHT_LOCAL:
                 /* Everything done already */
@@ -325,18 +339,18 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
             case LIGHT_SPOT:
                 /* https://docs.microsoft.com/en-us/windows/win32/direct3d9/attenuation-and-spotlight-factor#spotlight-factor */
                 mstring_append_fmt(body,
-                    "  vec4 spotDir = lightSpotDirection(%d);\n"
-                    "  float invScale = 1/length(spotDir.xyz);\n"
-                    "  float cosHalfPhi = -invScale*spotDir.w;\n"
-                    "  float cosHalfTheta = invScale + cosHalfPhi;\n"
-                    "  float spotDirDotVP = dot(spotDir.xyz, VP);\n"
-                    "  float rho = invScale*spotDirDotVP;\n"
-                    "  if (rho > cosHalfTheta) {\n"
-                    "  } else if (rho <= cosHalfPhi) {\n"
-                    "    attenuation = 0.0;\n"
-                    "  } else {\n"
-                    "    attenuation *= spotDirDotVP + spotDir.w;\n" /* FIXME: lightSpotFalloff */
-                    "  }\n",
+                    "    vec4 spotDir = lightSpotDirection(%d);\n"
+                    "    float invScale = 1/length(spotDir.xyz);\n"
+                    "    float cosHalfPhi = -invScale*spotDir.w;\n"
+                    "    float cosHalfTheta = invScale + cosHalfPhi;\n"
+                    "    float spotDirDotVP = dot(spotDir.xyz, VP);\n"
+                    "    float rho = invScale*spotDirDotVP;\n"
+                    "    if (rho > cosHalfTheta) {\n"
+                    "    } else if (rho <= cosHalfPhi) {\n"
+                    "      attenuation = 0.0;\n"
+                    "    } else {\n"
+                    "      attenuation *= spotDirDotVP + spotDir.w;\n" /* FIXME: lightSpotFalloff */
+                    "    }\n",
                     i);
                 break;
             default:
@@ -345,52 +359,83 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
             }
 
             mstring_append_fmt(body,
-                "  float pf;\n"
-                "  if (nDotVP == 0.0) {\n"
-                "    pf = 0.0;\n"
-                "  } else {\n"
-                "    pf = pow(nDotHV, /* specular(l, m, n, l1, m1, n1) */ 0.001);\n"
-                "  }\n"
-                "  vec3 lightAmbient = lightAmbientColor(%d) * attenuation;\n"
-                "  vec3 lightDiffuse = lightDiffuseColor(%d) * attenuation * nDotVP;\n"
-                "  vec3 lightSpecular = lightSpecularColor(%d) * pf;\n",
+                "    float pf;\n"
+                "    if (nDotVP == 0.0) {\n"
+                "      pf = 0.0;\n"
+                "    } else {\n"
+                "      pf = pow(nDotHV, specularPower);\n"
+                "    }\n"
+                "    vec3 lightAmbient = lightAmbientColor(%d) * attenuation;\n"
+                "    vec3 lightDiffuse = lightDiffuseColor(%d) * attenuation * nDotVP;\n"
+                "    vec3 lightSpecular = lightSpecularColor(%d) * attenuation * pf;\n",
                 i, i, i);
 
             mstring_append(body,
-                "  oD0.xyz += lightAmbient;\n");
+                "    oD0.xyz += lightAmbient;\n");
 
             switch (state->diffuse_src) {
             case MATERIAL_COLOR_SRC_MATERIAL:
                 mstring_append(body,
-                               "  oD0.xyz += lightDiffuse;\n");
+                               "    oD0.xyz += lightDiffuse;\n");
                 break;
             case MATERIAL_COLOR_SRC_DIFFUSE:
                 mstring_append(body,
-                               "  oD0.xyz += diffuse.xyz * lightDiffuse;\n");
+                               "    oD0.xyz += diffuse.xyz * lightDiffuse;\n");
                 break;
             case MATERIAL_COLOR_SRC_SPECULAR:
                 mstring_append(body,
-                               "  oD0.xyz += specular.xyz * lightDiffuse;\n");
+                               "    oD0.xyz += specular.xyz * lightDiffuse;\n");
                 break;
             }
 
-            mstring_append(body,
-                "  oD1.xyz += specular.xyz * lightSpecular;\n");
+            switch (state->specular_src) {
+            case MATERIAL_COLOR_SRC_MATERIAL:
+                mstring_append(body,
+                               "    oD1.xyz += lightSpecular;\n");
+                break;
+            case MATERIAL_COLOR_SRC_DIFFUSE:
+                mstring_append(body,
+                               "    oD1.xyz += diffuse.xyz * lightSpecular;\n");
+                break;
+            case MATERIAL_COLOR_SRC_SPECULAR:
+                mstring_append(body,
+                               "    oD1.xyz += specular.xyz * lightSpecular;\n");
+                break;
+            }
 
-            mstring_append(body, "}\n");
+            mstring_append(body, "  }\n"
+                                 "}\n");
         }
-    } else {
-        mstring_append(body, "  oD0 = diffuse;\n");
-        mstring_append(body, "  oD1 = specular;\n");
+
+        /* TODO: Implement two-sided lighting */
+        mstring_append(body, "  oB0 = backDiffuse;\n");
+        mstring_append(body, "  oB1 = backSpecular;\n");
     }
 
     if (!state->specular_enable) {
         mstring_append(body, "  oD1 = vec4(0.0, 0.0, 0.0, 1.0);\n");
+        mstring_append(body, "  oB1 = vec4(0.0, 0.0, 0.0, 1.0);\n");
+    } else {
+        if (!state->separate_specular) {
+            if (state->lighting) {
+				mstring_append(body,
+				               "  oD0.xyz += oD1.xyz;\n"
+				               "  oB0.xyz += oB1.xyz;\n"
+				);
+            }
+			mstring_append(body,
+				           "  oD1 = specular;\n"
+				           "  oB1 = backSpecular;\n"
+			);
+        }
+        if (state->ignore_specular_alpha) {
+            mstring_append(body,
+                           "  oD1.a = 1.0;\n"
+                           "  oB1.a = 1.0;\n"
+            );
+        }
     }
 
-    mstring_append(body, "  oB0 = backDiffuse;\n");
-    mstring_append(body, "  oB1 = backSpecular;\n");
-
     /* Fog */
     if (state->fog_enable) {
 
@@ -414,7 +459,7 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
             mstring_append(body, "  float fogDistance = fogCoord;\n");
             break;
         default:
-            assert(false);
+            assert(!"Invalid foggen mode");
             break;
         }
 

hw/xbox/nv2a/pgraph/glsl/vsh.c

@@ -256,6 +256,13 @@ MString *pgraph_gen_vsh_glsl(const ShaderState *state, bool prefix_outputs)
                        "  vtxD1 = clamp(oD1, 0.0, 1.0);\n"
                        "  vtxB1 = clamp(oB1, 0.0, 1.0);\n"
         );
+
+        if (state->ignore_specular_alpha) {
+            mstring_append(body,
+                           "  vtxD1.w = 1.0;\n"
+                           "  vtxB1.w = 1.0;\n"
+            );
+        }
     } else {
         mstring_append(body,
                        "  vtxD1 = vec4(0.0, 0.0, 0.0, 1.0);\n"

hw/xbox/nv2a/pgraph/methods.h.inc

@@ -27,6 +27,7 @@ DEF_METHOD(NV097, SET_COMBINER_SPECULAR_FOG_CW0)
 DEF_METHOD(NV097, SET_COMBINER_SPECULAR_FOG_CW1)
 DEF_METHOD_CASE_4(NV097, SET_TEXTURE_ADDRESS, 64)
 DEF_METHOD(NV097, SET_CONTROL0)
+DEF_METHOD(NV097, SET_LIGHT_CONTROL)
 DEF_METHOD(NV097, SET_COLOR_MATERIAL)
 DEF_METHOD(NV097, SET_FOG_MODE)
 DEF_METHOD(NV097, SET_FOG_GEN_MODE)
@@ -95,6 +96,7 @@ DEF_METHOD_RANGE(NV097, SET_FOG_PARAMS, 3)
 DEF_METHOD_RANGE(NV097, SET_TEXGEN_PLANE_S, 4*4*4)
 DEF_METHOD(NV097, SET_TEXGEN_VIEW_MODEL)
 DEF_METHOD_RANGE(NV097, SET_FOG_PLANE, 4)
+DEF_METHOD_RANGE(NV097, SET_SPECULAR_PARAMS, 6)
 DEF_METHOD_RANGE(NV097, SET_SCENE_AMBIENT_COLOR, 3)
 DEF_METHOD_RANGE(NV097, SET_VIEWPORT_OFFSET, 4)
 DEF_METHOD_RANGE(NV097, SET_POINT_PARAMS, 8)
@@ -134,6 +136,11 @@ DEF_METHOD_RANGE(NV097, SET_TEXCOORD3_2F, 2)
 DEF_METHOD_RANGE(NV097, SET_TEXCOORD3_4F, 4)
 DEF_METHOD_RANGE(NV097, SET_TEXCOORD3_2S, 1)
 DEF_METHOD_RANGE(NV097, SET_TEXCOORD3_4S, 2)
+DEF_METHOD(NV097, SET_FOG_COORD)
+DEF_METHOD(NV097, SET_WEIGHT1F)
+DEF_METHOD_RANGE(NV097, SET_WEIGHT2F, 2)
+DEF_METHOD_RANGE(NV097, SET_WEIGHT3F, 3)
+DEF_METHOD_RANGE(NV097, SET_WEIGHT4F, 4)
 DEF_METHOD_RANGE(NV097, SET_VERTEX_DATA_ARRAY_FORMAT, 16)
 DEF_METHOD_RANGE(NV097, SET_VERTEX_DATA_ARRAY_OFFSET, 16)
 DEF_METHOD(NV097, SET_LOGIC_OP_ENABLE)
@@ -177,6 +184,7 @@ DEF_METHOD(NV097, CLEAR_SURFACE)
 DEF_METHOD(NV097, SET_CLEAR_RECT_HORIZONTAL)
 DEF_METHOD(NV097, SET_CLEAR_RECT_VERTICAL)
 DEF_METHOD_RANGE(NV097, SET_SPECULAR_FOG_FACTOR, 2)
+DEF_METHOD_RANGE(NV097, SET_SPECULAR_PARAMS_BACK, 6)
 DEF_METHOD(NV097, SET_SHADER_CLIP_PLANE_MODE)
 DEF_METHOD_RANGE(NV097, SET_COMBINER_COLOR_OCW, 8)
 DEF_METHOD(NV097, SET_COMBINER_CONTROL)

hw/xbox/nv2a/pgraph/pgraph.c

@@ -19,6 +19,8 @@
  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  */
 
+#include <math.h>
+
 #include "hw/xbox/nv2a/nv2a_int.h"
 #include "ui/xemu-notifications.h"
 #include "ui/xemu-settings.h"
@@ -1075,6 +1077,18 @@ DEF_METHOD(NV097, SET_CONTROL0)
              z_perspective);
 }
 
+DEF_METHOD(NV097, SET_LIGHT_CONTROL)
+{
+    PG_SET_MASK(NV_PGRAPH_CSV0_C, NV_PGRAPH_CSV0_C_SEPARATE_SPECULAR,
+             (parameter & NV097_SET_LIGHT_CONTROL_SEPARATE_SPECULAR) != 0);
+
+    PG_SET_MASK(NV_PGRAPH_CSV0_C, NV_PGRAPH_CSV0_C_LOCALEYE,
+             (parameter & NV097_SET_LIGHT_CONTROL_LOCALEYE) != 0);
+
+    PG_SET_MASK(NV_PGRAPH_CSV0_C, NV_PGRAPH_CSV0_C_ALPHA_FROM_MATERIAL_SPECULAR,
+             (parameter & NV097_SET_LIGHT_CONTROL_ALPHA_FROM_MATERIAL_SPECULAR) != 0);
+}
+
 DEF_METHOD(NV097, SET_COLOR_MATERIAL)
 {
     PG_SET_MASK(NV_PGRAPH_CSV0_C, NV_PGRAPH_CSV0_C_EMISSION,
@@ -1791,6 +1805,113 @@ DEF_METHOD_INC(NV097, SET_FOG_PLANE)
     pg->vsh_constants_dirty[NV_IGRAPH_XF_XFCTX_FOG] = true;
 }
 
+struct CurveCoefficients {
+  float a;
+  float b;
+  float c;
+};
+
+static const struct CurveCoefficients curve_coefficients[] = {
+  {1.000108475163, -9.838607076280, 54.829089549713},
+  {1.199164441703, -3.292603784852, 7.799987995214},
+  {8.653441252033, 29.189473787191, 43.586027561823},
+  {-531.307758450301, 117.398468683934, 113.155490738338},
+  {-4.662713151292, 1.221108944572, 1.217360986939},
+  {-124.435242105211, 35.401219563514, 35.408114377045},
+  {10672560.259502287954, 21565843.555823743343, 10894794.336297152564},
+  {-51973801.463933646679, -104199997.554352939129, -52225454.356278456748},
+  {972270.324080004124, 2025882.096547174733, 1054898.052467488218},
+};
+
+static const float kCoefficient0StepPoints[] = {
+  -0.022553957999, // power = 1.25
+  -0.421539008617, // power = 4.00
+  -0.678715527058, // power = 9.00
+  -0.838916420937, // power = 20.00
+  -0.961754500866, // power = 90.00
+  -0.990773200989, // power = 375.00
+  -0.994858562946, // power = 650.00
+  -0.996561050415, // power = 1000.00
+  -0.999547004700, // power = 1250.00
+};
+
+static float reconstruct_quadratic(float c0, const struct CurveCoefficients *coefficients) {
+  return coefficients->a + coefficients->b * c0 + coefficients->c * c0 * c0;
+}
+
+static float reconstruct_saturation_growth_rate(float c0, const struct CurveCoefficients *coefficients) {
+  return (coefficients->a * c0) / (coefficients->b + coefficients->c * c0);
+}
+
+static float (* const reconstruct_func_map[])(float, const struct CurveCoefficients *) = {
+  reconstruct_quadratic, // 1.0..1.25 max error 0.01 %
+  reconstruct_quadratic, // 1.25..4.0 max error 2.2 %
+  reconstruct_quadratic, // 4.0..9.0 max error 2.3 %
+  reconstruct_saturation_growth_rate, // 9.0..20.0 max error 1.4 %
+  reconstruct_saturation_growth_rate, // 20.0..90.0 max error 2.1 %
+  reconstruct_saturation_growth_rate, // 90.0..375.0 max error 2.8%
+  reconstruct_quadratic, // 375..650 max error 1.0 %
+  reconstruct_quadratic, // 650..1000 max error 1.7%
+  reconstruct_quadratic, // 1000..1250 max error 1.0%
+};
+
+static float reconstruct_specular_power(const float *params) {
+  // See https://github.com/dracc/xgu/blob/db3172d8c983629f0dc971092981846da22438ae/xgux.h#L279
+
+  // Values < 1.0 will result in a positive c1 and (c2 - c0 * 2) will be very
+  // close to the original value.
+  if (params[1] > 0.0f && params[2] < 1.0f) {
+    return params[2] - (params[0] * 2.0f);
+  }
+
+  float c0 = params[0];
+  float c3 = params[3];
+  // FIXME: This handling is not correct, but is distinct without crashing.
+  // It does not appear possible for a DirectX-generated value to be positive,
+  // so while this differs from hardware behavior, it may be irrelevant in
+  // practice.
+  if (c0 > 0.0f || c3 > 0.0f) {
+    return 0.0001f;
+  }
+
+  float reconstructed_power = 0.f;
+  for (uint32_t i = 0; i < sizeof(kCoefficient0StepPoints) / sizeof(kCoefficient0StepPoints[0]); ++i) {
+    if (c0 > kCoefficient0StepPoints[i]) {
+      reconstructed_power = reconstruct_func_map[i](c0, &curve_coefficients[i]);
+      break;
+    }
+  }
+
+  float reconstructed_half_power = 0.f;
+  for (uint32_t i = 0; i < sizeof(kCoefficient0StepPoints) / sizeof(kCoefficient0StepPoints[0]); ++i) {
+    if (c3 > kCoefficient0StepPoints[i]) {
+      reconstructed_half_power = reconstruct_func_map[i](c3, &curve_coefficients[i]);
+      break;
+    }
+  }
+
+  // The range can be extended beyond 1250 by using the half power params. This
+  // will only work for DirectX generated values, arbitrary params could
+  // erroneously trigger this.
+  //
+  // There are some very low power (~1) values that have inverted powers, but
+  // they are easily identified by comparatively high c0 parameters.
+  if (reconstructed_power == 0.f || (reconstructed_half_power > reconstructed_power && c0 < -0.1f)) {
+    return reconstructed_half_power * 2.f;
+  }
+
+  return reconstructed_power;
+}
+
+DEF_METHOD_INC(NV097, SET_SPECULAR_PARAMS)
+{
+    int slot = (method - NV097_SET_SPECULAR_PARAMS) / 4;
+    pg->specular_params[slot] = *(float *)&parameter;
+    if (slot == 5) {
+        pg->specular_power = reconstruct_specular_power(pg->specular_params);
+    }
+}
+
 DEF_METHOD_INC(NV097, SET_SCENE_AMBIENT_COLOR)
 {
     int slot = (method - NV097_SET_SCENE_AMBIENT_COLOR) / 4;
@@ -2010,6 +2131,26 @@ DEF_METHOD_INC(NV097, SET_VERTEX4F)
     }
 }
 
+DEF_METHOD(NV097, SET_FOG_COORD)
+{
+    VertexAttribute *attribute = &pg->vertex_attributes[NV2A_VERTEX_ATTR_FOG];
+    pgraph_allocate_inline_buffer_vertices(pg, NV2A_VERTEX_ATTR_FOG);
+    attribute->inline_value[0] = *(float*)&parameter;
+    attribute->inline_value[1] = attribute->inline_value[0];
+    attribute->inline_value[2] = attribute->inline_value[0];
+    attribute->inline_value[3] = attribute->inline_value[0];
+}
+
+DEF_METHOD(NV097, SET_WEIGHT1F)
+{
+    VertexAttribute *attribute = &pg->vertex_attributes[NV2A_VERTEX_ATTR_WEIGHT];
+    pgraph_allocate_inline_buffer_vertices(pg, NV2A_VERTEX_ATTR_WEIGHT);
+    attribute->inline_value[0] = *(float*)&parameter;
+    attribute->inline_value[1] = 0.f;
+    attribute->inline_value[2] = 0.f;
+    attribute->inline_value[3] = 1.f;
+}
+
 DEF_METHOD_INC(NV097, SET_NORMAL3S)
 {
     int slot = (method - NV097_SET_NORMAL3S) / 4;
@@ -2144,7 +2285,6 @@ DEF_METHOD_INC(NV097, SET_TEXCOORD1_4F)
     SET_VERTEX_ATTRIBUTE_F(NV097_SET_TEXCOORD1_4F, NV2A_VERTEX_ATTR_TEXTURE1);
 }
 
-
 DEF_METHOD_INC(NV097, SET_TEXCOORD2_4F)
 {
     SET_VERTEX_ATTRIBUTE_F(NV097_SET_TEXCOORD2_4F, NV2A_VERTEX_ATTR_TEXTURE2);
@@ -2155,8 +2295,34 @@ DEF_METHOD_INC(NV097, SET_TEXCOORD3_4F)
     SET_VERTEX_ATTRIBUTE_F(NV097_SET_TEXCOORD3_4F, NV2A_VERTEX_ATTR_TEXTURE3);
 }
 
+DEF_METHOD_INC(NV097, SET_WEIGHT4F)
+{
+    SET_VERTEX_ATTRIBUTE_F(NV097_SET_WEIGHT4F, NV2A_VERTEX_ATTR_WEIGHT);
+}
+
 #undef SET_VERTEX_ATTRIBUTE_F
 
+DEF_METHOD_INC(NV097, SET_WEIGHT2F)
+{
+    int slot = (method - NV097_SET_WEIGHT2F) / 4;
+    VertexAttribute *attribute =
+        &pg->vertex_attributes[NV2A_VERTEX_ATTR_WEIGHT];
+    pgraph_allocate_inline_buffer_vertices(pg, NV2A_VERTEX_ATTR_WEIGHT);
+    attribute->inline_value[slot] = *(float*)&parameter;
+    attribute->inline_value[2] = 0.0f;
+    attribute->inline_value[3] = 1.0f;
+}
+
+DEF_METHOD_INC(NV097, SET_WEIGHT3F)
+{
+    int slot = (method - NV097_SET_WEIGHT3F) / 4;
+    VertexAttribute *attribute =
+        &pg->vertex_attributes[NV2A_VERTEX_ATTR_WEIGHT];
+    pgraph_allocate_inline_buffer_vertices(pg, NV2A_VERTEX_ATTR_WEIGHT);
+    attribute->inline_value[slot] = *(float*)&parameter;
+    attribute->inline_value[3] = 1.0f;
+}
+
 #define SET_VERTEX_ATRIBUTE_TEX_2F(command, attr_index)                    \
     do {                                                                   \
         int slot = (method - (command)) / 4;                               \
@@ -2526,7 +2692,11 @@ DEF_METHOD(NV097, DRAW_ARRAYS)
     int32_t count = GET_MASK(parameter, NV097_DRAW_ARRAYS_COUNT) + 1;
 
     if (pg->inline_elements_length) {
-        /* FIXME: Determine HW behavior for overflow case. */
+        /* FIXME: HW throws an exception if the start index is > 0xFFFF. This
+         * would prevent this assert from firing for any reasonable choice of
+         * NV2A_MAX_BATCH_LENGTH (which must be larger to accommodate
+         * NV097_INLINE_ARRAY anyway)
+         */
         assert((pg->inline_elements_length + count) < NV2A_MAX_BATCH_LENGTH);
         assert(!pg->draw_arrays_prevent_connect);
 
@@ -2728,6 +2898,15 @@ DEF_METHOD_INC(NV097, SET_SPECULAR_FOG_FACTOR)
     pgraph_reg_w(pg, NV_PGRAPH_SPECFOGFACTOR0 + slot*4, parameter);
 }
 
+DEF_METHOD_INC(NV097, SET_SPECULAR_PARAMS_BACK)
+{
+    int slot = (method - NV097_SET_SPECULAR_PARAMS_BACK) / 4;
+    pg->specular_params_back[slot] = *(float *)&parameter;
+    if (slot == 5) {
+        pg->specular_power_back = reconstruct_specular_power(pg->specular_params_back);
+    }
+}
+
 DEF_METHOD(NV097, SET_SHADER_CLIP_PLANE_MODE)
 {
     pgraph_reg_w(pg, NV_PGRAPH_SHADERCLIPMODE, parameter);

hw/xbox/nv2a/pgraph/pgraph.h

@@ -197,6 +197,11 @@ typedef struct PGRAPHState {
     float light_local_position[NV2A_MAX_LIGHTS][3];
     float light_local_attenuation[NV2A_MAX_LIGHTS][3];
 
+    float specular_params[6];
+    float specular_power;
+    float specular_params_back[6];
+    float specular_power_back;
+
     float point_params[8];
 
     VertexAttribute vertex_attributes[NV2A_VERTEXSHADER_ATTRIBUTES];

hw/xbox/nv2a/pgraph/shaders.c

@@ -92,6 +92,16 @@ ShaderState pgraph_get_shader_state(PGRAPHState *pg)
             pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_SPECULAR);
     }
 
+    state.separate_specular = GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_SEPARATE_SPECULAR);
+    state.ignore_specular_alpha = !GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_ALPHA_FROM_MATERIAL_SPECULAR);
+    state.local_eye = GET_MASK(
+        pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_LOCALEYE);
+
+    state.specular_power = pg->specular_power;
+    state.specular_power_back = pg->specular_power_back;
+
     /* vertex program stuff */
     state.vertex_program = vertex_program,
     state.z_perspective = pgraph_reg_r(pg, NV_PGRAPH_CONTROL_0) &

hw/xbox/nv2a/pgraph/shaders.h

@@ -79,6 +79,12 @@ typedef struct ShaderState {
     enum MaterialColorSource diffuse_src;
     enum MaterialColorSource specular_src;
 
+    bool separate_specular;
+    bool ignore_specular_alpha;
+    bool local_eye;
+    float specular_power;
+    float specular_power_back;
+
     bool lighting;
     enum VshLight light[NV2A_MAX_LIGHTS];
 

hw/xbox/nv2a/pgraph/vk/renderer.h

@@ -190,6 +190,7 @@ typedef struct ShaderBinding {
     int light_infinite_direction_loc[NV2A_MAX_LIGHTS];
     int light_local_position_loc[NV2A_MAX_LIGHTS];
     int light_local_attenuation_loc[NV2A_MAX_LIGHTS];
+    int specular_power_loc;
 
     int clip_region_loc;
 

hw/xbox/nv2a/pgraph/vk/shaders.c

@@ -313,6 +313,9 @@ static void update_shader_constant_locations(ShaderBinding *binding)
 
     binding->uniform_attrs_loc =
         uniform_index(&binding->vertex->uniforms, "inlineValue");
+
+    binding->specular_power_loc =
+        uniform_index(&binding->vertex->uniforms, "specularPower");
 }
 
 static void shader_cache_entry_init(Lru *lru, LruNode *node, void *state)
@@ -607,6 +610,11 @@ static void shader_update_constants(PGRAPHState *pg, ShaderBinding *binding,
             }
         }
 
+        if (binding->specular_power_loc != -1) {
+            uniform1f(&binding->vertex->uniforms, binding->specular_power_loc,
+                      pg->specular_power);
+        }
+
         /* estimate the viewport by assuming it matches the surface ... */
         unsigned int aa_width = 1, aa_height = 1;
         pgraph_apply_anti_aliasing_factor(pg, &aa_width, &aa_height);

hw/xbox/nv2a/pgraph/vsh.h

@@ -55,8 +55,6 @@ enum VshFoggen {
     FOGGEN_RADIAL,
     FOGGEN_PLANAR,
     FOGGEN_ABS_PLANAR,
-    FOGGEN_ERROR4,
-    FOGGEN_ERROR5,
     FOGGEN_FOG_X
 };
 

include/migration/vmstate.h

@@ -1085,6 +1085,9 @@ extern const VMStateInfo vmstate_info_qlist;
 #define VMSTATE_UINT32_SUB_ARRAY(_f, _s, _start, _num)                \
     VMSTATE_SUB_ARRAY(_f, _s, _start, _num, 0, vmstate_info_uint32, uint32_t)
 
+#define VMSTATE_UINT32_SUB_ARRAY_V(_f, _s, _start, _num, _v)          \
+    VMSTATE_SUB_ARRAY(_f, _s, _start, _num, _v, vmstate_info_uint32, uint32_t)
+
 #define VMSTATE_UINT32_2DARRAY(_f, _s, _n1, _n2)                      \
     VMSTATE_UINT32_2DARRAY_V(_f, _s, _n1, _n2, 0)