nv2a: Partial implementation of SET_SPECULAR_PARAMS

hw/xbox/nv2a/nv2a_regs.h

@@ -1056,6 +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
@@ -1264,6 +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

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";
@@ -269,12 +275,6 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
                 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
@@ -310,14 +310,10 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
                     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 nDotVP = max(0.0, dot(tNormal, normalize(lightInfiniteDirection%d)));\n"
-                    "  float nDotHV = max(0.0, dot(tNormal, vec3(lightInfiniteHalfVector%d)));\n",
+                    "  float nDotHV = max(0.0, dot(tNormal, lightInfiniteHalfVector%d));\n",
                     i, i);
 
-                /* FIXME: Do specular */
-
-                /* FIXME: tBackDiffuse */
-
                 break;
             case LIGHT_LOCAL:
                 /* Everything done already */
@@ -349,11 +345,11 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
                 "  if (nDotVP == 0.0) {\n"
                 "    pf = 0.0;\n"
                 "  } else {\n"
-                "    pf = pow(nDotHV, /* specular(l, m, n, l1, m1, n1) */ 0.001);\n"
+                "    pf = pow(nDotHV, specularPower);\n"
                 "  }\n"
                 "  vec3 lightAmbient = lightAmbientColor(%d) * attenuation;\n"
                 "  vec3 lightDiffuse = lightDiffuseColor(%d) * attenuation * nDotVP;\n"
-                "  vec3 lightSpecular = lightSpecularColor(%d) * pf;\n",
+                "  vec3 lightSpecular = lightSpecularColor(%d) * attenuation * pf;\n",
                 i, i, i);
 
             mstring_append(body,
@@ -374,26 +370,44 @@ GLSL_DEFINE(materialEmissionColor, GLSL_LTCTXA(NV_IGRAPH_XF_LTCTXA_CM_COL) ".xyz
                 break;
             }
 
-            mstring_append(body,
+            switch (state->specular_src) {
-                "  oD1.xyz += specular.xyz * lightSpecular;\n");
+            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");
         }
-    } else {
-        mstring_append(body, "  oD0 = diffuse;\n");
-        mstring_append(body, "  oD1 = specular;\n");
-    }
 
-    mstring_append(body, "  oB0 = backDiffuse;\n");
+        /* TODO: Implement two-sided lighting */
-    mstring_append(body, "  oB1 = backSpecular;\n");
+        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) {
-            mstring_append(body, "  oD1 = specular;\n");
+            if (state->lighting) {
-            mstring_append(body, "  oB1 = backSpecular;\n");
+				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,

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

@@ -96,6 +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)
@@ -183,6 +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"
@@ -1803,6 +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;
@@ -2785,6 +2894,15 @@ 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

@@ -97,6 +97,9 @@ ShaderState pgraph_get_shader_state(PGRAPHState *pg)
     state.ignore_specular_alpha = !GET_MASK(
         pgraph_reg_r(pg, NV_PGRAPH_CSV0_C), NV_PGRAPH_CSV0_C_ALPHA_FROM_MATERIAL_SPECULAR);
 
+    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

@@ -81,6 +81,8 @@ typedef struct ShaderState {
 
     bool separate_specular;
     bool ignore_specular_alpha;
+    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;