From b3fc80b3a83fa99aeb541a0c665cfb596cedd71c Mon Sep 17 00:00:00 2001
From: Matt Borgerson <contact@mborgerson.com>
Date: Sat, 21 Oct 2023 22:28:59 -0700
Subject: [PATCH] nv2a: Implement ZCLIP_MIN,MAX with gl_ClipDistance

---
 hw/xbox/nv2a/pgraph.c  | 9 +++++++--
 hw/xbox/nv2a/shaders.c | 9 +++++++--
 hw/xbox/nv2a/vsh.c     | 5 -----
 3 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/hw/xbox/nv2a/pgraph.c b/hw/xbox/nv2a/pgraph.c
index a134a2a661..fbf58fce9a 100644
--- a/hw/xbox/nv2a/pgraph.c
+++ b/hw/xbox/nv2a/pgraph.c
@@ -2945,6 +2945,10 @@ DEF_METHOD(NV097, SET_BEGIN_END)
             glDisable(GL_CULL_FACE);
         }
 
+        /* Clipping */
+        glEnable(GL_CLIP_DISTANCE0);
+        glEnable(GL_CLIP_DISTANCE1);
+
         /* Front-face select */
         glFrontFace(pg->regs[NV_PGRAPH_SETUPRASTER]
                         & NV_PGRAPH_SETUPRASTER_FRONTFACE
@@ -4179,7 +4183,6 @@ static void pgraph_shader_update_constants(PGRAPHState *pg,
                     *(float*)&pg->regs[NV_PGRAPH_FOGPARAM1]);
     }
 
-    /* FIXME: Handle NV_PGRAPH_ZCLIPMIN, NV_PGRAPH_ZCLIPMAX */
     float zmax;
     switch (pg->surface_shape.zeta_format) {
     case NV097_SET_SURFACE_FORMAT_ZETA_Z16:
@@ -4289,7 +4292,9 @@ static void pgraph_shader_update_constants(PGRAPHState *pg,
     }
 
     if (binding->clip_range_loc != -1) {
-        glUniform2f(binding->clip_range_loc, 0, zmax);
+        float zclip_min = *(float*)&pg->regs[NV_PGRAPH_ZCLIPMIN] / zmax * 2.0 - 1.0;
+        float zclip_max = *(float*)&pg->regs[NV_PGRAPH_ZCLIPMAX] / zmax * 2.0 - 1.0;
+        glUniform4f(binding->clip_range_loc, 0, zmax, zclip_min, zclip_max);
     }
 
     /* Clipping regions */
diff --git a/hw/xbox/nv2a/shaders.c b/hw/xbox/nv2a/shaders.c
index d0a1c0fa7c..cafe326e93 100644
--- a/hw/xbox/nv2a/shaders.c
+++ b/hw/xbox/nv2a/shaders.c
@@ -267,6 +267,8 @@ static MString* generate_geometry_shader(
                        "void emit_vertex(int index, int _unused) {\n"
                        "  gl_Position = gl_in[index].gl_Position;\n"
                        "  gl_PointSize = gl_in[index].gl_PointSize;\n"
+                       "  gl_ClipDistance[0] = gl_in[index].gl_ClipDistance[0];\n"
+                       "  gl_ClipDistance[1] = gl_in[index].gl_ClipDistance[1];\n"
                        "  vtx_inv_w = v_vtx_inv_w[index];\n"
                        "  vtx_inv_w_flat = v_vtx_inv_w[index];\n"
                        "  vtxD0 = v_vtxD0[index];\n"
@@ -289,6 +291,8 @@ static MString* generate_geometry_shader(
                        "void emit_vertex(int index, int provoking_index) {\n"
                        "  gl_Position = gl_in[index].gl_Position;\n"
                        "  gl_PointSize = gl_in[index].gl_PointSize;\n"
+                       "  gl_ClipDistance[0] = gl_in[index].gl_ClipDistance[0];\n"
+                       "  gl_ClipDistance[1] = gl_in[index].gl_ClipDistance[1];\n"
                        "  vtx_inv_w = v_vtx_inv_w[index];\n"
                        "  vtx_inv_w_flat = v_vtx_inv_w[provoking_index];\n"
                        "  vtxD0 = v_vtxD0[provoking_index];\n"
@@ -784,7 +788,7 @@ static MString *generate_vertex_shader(const ShaderState *state,
     MString *header = mstring_from_str(
 "#version 400\n"
 "\n"
-"uniform vec2 clipRange;\n"
+"uniform vec4 clipRange;\n"
 "uniform vec2 surfaceSize;\n"
 "\n"
 /* All constants in 1 array declaration */
@@ -864,7 +868,6 @@ GLSL_DEFINE(texMat3, GLSL_C_MAT4(NV_IGRAPH_XF_XFCTX_T3MAT))
 
     if (state->fixed_function) {
         generate_fixed_function(state, header, body);
-
     } else if (state->vertex_program) {
         vsh_translate(VSH_VERSION_XVS,
                       (uint32_t*)state->program_data,
@@ -973,6 +976,8 @@ GLSL_DEFINE(texMat3, GLSL_C_MAT4(NV_IGRAPH_XF_XFCTX_T3MAT))
                       "  vtxT3 = oT3 * vtx_inv_w;\n"
                       "  gl_Position = oPos;\n"
                       "  gl_PointSize = oPts.x;\n"
+                      "  gl_ClipDistance[0] = oPos.z - oPos.w*clipRange.z;\n" // Near
+                      "  gl_ClipDistance[1] = oPos.w*clipRange.w - oPos.z;\n" // Far
                       "\n"
                       "}\n",
                        shade_model_mult,
diff --git a/hw/xbox/nv2a/vsh.c b/hw/xbox/nv2a/vsh.c
index 5f338e7368..0e4cf314bc 100644
--- a/hw/xbox/nv2a/vsh.c
+++ b/hw/xbox/nv2a/vsh.c
@@ -849,11 +849,6 @@ void vsh_translate(uint16_t version,
         mstring_append(body, "  oPos.z = oPos.w;\n");
     }
     mstring_append(body,
-        /* Map the clip range into clip space so z is clipped correctly.
-         * Note this makes the values in the depth buffer wrong. This should be
-         * handled with gl_ClipDistance instead, but that has performance issues
-         * on OS X.
-         */
         "  if (clipRange.y != clipRange.x) {\n"
         "    oPos.z = (oPos.z - clipRange.x)/(0.5*(clipRange.y - clipRange.x)) - 1;\n"
         "  }\n"