Lighting: Take the specular viewer direction from the view matrix

The viewer is at infinity along view space +z, so in world space, where
lighting happens, it's the view matrix's third column rather than
(0,0,1): turning the camera moves the highlights (gpu/lighting/specular).
The shaders read it from u_view.

In a Need for Speed Carbon frame replayed on a PSP, this and the pow
bring the error of the cars from MSE 957 to 120 (Vulkan).

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

GPU/Common/TransformCommon.cpp

@@ -41,7 +41,7 @@ Lighter::Lighter(int vertType) {
 	materialSpecular.GetFromRGB(gstate.materialspecular);
 	materialSpecular.a = 1.0f;
 	specCoef_ = PSPSpecularCoef(getFloat24(gstate.materialspecularcoef));
-	// viewer_ = Vec3f(-gstate.viewMatrix[9], -gstate.viewMatrix[10], -gstate.viewMatrix[11]);
+	viewDir_ = PSPViewDirection(gstate.viewMatrix);
 	bool hasColor = (vertType & GE_VTYPE_COL_MASK) != 0;
 	materialUpdate_ = hasColor ? (gstate.materialupdate & 7) : 0;
 
@@ -148,13 +148,8 @@ void Lighter::Light(float colorOut0[4], float colorOut1[4], const float colorIn[
 		Color4 lightDiff(lcolor[1][l], 0.0f);
 		Color4 diff = (lightDiff * *diffuse) * dot;
 
-		// Real PSP specular
-		static const Vec3f toViewer(0, 0, 1);
-		// Better specular
-		// Vec3f toViewer = (viewer - pos).NormalizedOr001(cpu_info.bSSE4_1);
-
 		if (doSpecular && facingLight) {
-			Vec3f halfVec = (toLight + toViewer).NormalizedOr001(cpu_info.bSSE4_1);
+			Vec3f halfVec = (toLight + viewDir_).NormalizedOr001(cpu_info.bSSE4_1);
 
 			dot = Dot(halfVec, norm);
 			if (dot > 0.0f) {

GPU/Common/TransformCommon.h

@@ -85,6 +85,11 @@ inline float PSPSpecularCoef(float e) {
 	return e;
 }
 
+// The viewer is at infinity along view space +z, so in world space it's the view matrix's third column.
+inline Vec3f PSPViewDirection(const float viewMatrix[12]) {
+	return Vec3f(viewMatrix[2], viewMatrix[5], viewMatrix[8]).NormalizedOr001(false);
+}
+
 // Convenient way to do precomputation to save the parts of the lighting calculation
 // that's common between the many vertices of a draw call.
 class Lighter {
@@ -106,7 +111,7 @@ private:
 	Color4 materialDiffuse;
 	Color4 materialSpecular;
 	float specCoef_;
-	// Vec3f viewer_;
+	Vec3f viewDir_;
 	bool doShadeMapping_;
 	int materialUpdate_;
 

GPU/Common/VertexShaderGenerator.cpp

@@ -500,6 +500,14 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
 		} else {
 			WRITE(p, "  mediump vec3 worldnormal = normalizeOr001(mul(vec4(0.0, 0.0, %s1.0, 0.0), u_world).xyz);\n", flipNormal ? "-" : "");
 		}
+		if (enableLighting) {
+			// The viewer is at infinity along view space +z: in world space, the view matrix's third column.
+			if (compat.shaderLanguage == HLSL_D3D11) {
+				WRITE(p, "  mediump vec3 viewDir = normalizeOr001(vec3(u_view[0].z, u_view[1].z, u_view[2].z));\n");
+			} else {
+				WRITE(p, "  mediump vec3 viewDir = normalizeOr001(u_view[2].xyz);\n");
+			}
+		}
 
 		WRITE(p, "  vec4 viewPos = vec4(mul(vec4(worldpos, 1.0), u_view).xyz, 1.0);\n");
 		if (useSimpleStereo) {
@@ -668,7 +676,7 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
 				p.F("    diffuse = (u_lightdiffuse%s * diffuseColor) * max(ldot, 0.0);\n", iStr);
 				p.C("    if (comp == 0x1u && ldot >= 0.0) {\n");  // do specular. note - must allow for the >= case, since the u_matspecular.a <= 0.0 case relies on it.
 				p.C("      if (u_matspecular.a > 0.0) {\n");
-				p.C("        vec3 halfVec = toLight + vec3(0.0, 0.0, 1.0);\n");
+				p.C("        vec3 halfVec = toLight + viewDir;\n");
 				p.C("        float halfInvLen = inversesqrt(dot(halfVec, halfVec));\n");
 				p.C("        ldot = pspPow(dot(halfVec, worldnormal) * halfInvLen, u_matspecular.a);\n");
 				p.C("      } else {\n");
@@ -748,7 +756,7 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
 				if (doSpecular) {
 					p.C("  if (ldot >= 0.0) {\n");
 					p.C("    if (u_matspecular.a > 0.0) {\n");
-					p.C("      vec3 halfVec = toLight + vec3(0.0, 0.0, 1.0);\n");
+					p.C("      vec3 halfVec = toLight + viewDir;\n");
 					p.C("      float halfInvLen = inversesqrt(dot(halfVec, halfVec));\n");
 					p.C("      ldot = pspPow(dot(halfVec, worldnormal) * halfInvLen, u_matspecular.a);\n");
 					p.C("    } else {\n");

GPU/Software/Lighting.cpp

@@ -369,7 +369,7 @@ static void ProcessSIMD(VertexData &vertex, const WorldCoords &worldpos, const W
 		}
 
 		if (lstate.specular && diffuse_factor >= 0.0f) {
-			Vec3<float> H = L + Vec3<float>(0.f, 0.f, 1.f);
+			Vec3<float> H = L + state.viewDir;
 
 			float specular_factor = Dot33(H.NormalizedOr001(useSSE4), worldnormal);
 			specular_factor = pspLightPow(specular_factor, state.specularExp);

GPU/Software/Lighting.h

@@ -53,6 +53,7 @@ struct State {
 
 	Vec4<int> baseAmbientColorFactor;
 	float specularExp;
+	Vec3f viewDir;
 
 	struct {
 		bool colorForAmbient : 1;

GPU/Software/TransformUnit.cpp

@@ -28,6 +28,7 @@
 #include "GPU/Common/DrawEngineCommon.h"
 #include "GPU/Common/VertexDecoderCommon.h"
 #include "GPU/Common/SoftwareTransformCommon.h"
+#include "GPU/Common/TransformCommon.h"
 #include "GPU/Common/VertexReader.h"
 #include "GPU/GPUStateSIMDUtil.h"
 #include "Common/Math/SIMDHeaders.h"
@@ -259,6 +260,7 @@ void ComputeTransformState(TransformState *state, const VertexReader &vreader) {
 		} else {
 			state->lightingState.usesWorldNormal = state->uvGenMode == GE_TEXMAP_ENVIRONMENT_MAP;
 		}
+		state->lightingState.viewDir = PSPViewDirection(gstate.viewMatrix);
 
 		float world[16];
 		float view[16];