diff --git a/GPU/Software/Rasterizer.cpp b/GPU/Software/Rasterizer.cpp
index 2da86fb45b..8101e63f00 100644
--- a/GPU/Software/Rasterizer.cpp
+++ b/GPU/Software/Rasterizer.cpp
@@ -233,6 +233,7 @@ static inline void GetTextureCoordinates(const VertexData& v0, const VertexData&
 	float wq1 = (1.0f - p) * q1;
 
 	float q_recip = 1.0f / (wq0 + wq1);
+	// TODO: Handle projection.
 	s = (v0.texturecoords.s() * wq0 + v1.texturecoords.s() * wq1) * q_recip;
 	t = (v0.texturecoords.t() * wq0 + v1.texturecoords.t() * wq1) * q_recip;
 }
@@ -249,6 +250,7 @@ static inline void GetTextureCoordinates(const VertexData &v0, const VertexData
 	Vec4<float> wq2 = w2.Cast<float>() * q2;
 
 	Vec4<float> q_recip = (wq0 + wq1 + wq2).Reciprocal();
+	// TODO: Handle projection.
 	s = Interpolate(v0.texturecoords.s(), v1.texturecoords.s(), v2.texturecoords.s(), wq0, wq1, wq2, q_recip);
 	t = Interpolate(v0.texturecoords.t(), v1.texturecoords.t(), v2.texturecoords.t(), wq0, wq1, wq2, q_recip);
 }
@@ -770,8 +772,9 @@ void DrawRectangle(const VertexData &v0, const VertexData &v1, const BinCoords &
 	Vec2f stx(0.0f, 0.0f);
 	Vec2f sty(0.0f, 0.0f);
 	if (state.enableTextures) {
-		Vec2f tc0 = v0.texturecoords;
-		Vec2f tc1 = v1.texturecoords;
+		// TODO: Handle projection.
+		Vec2f tc0 = v0.texturecoords.uv();
+		Vec2f tc1 = v1.texturecoords.uv();
 		if (state.throughMode) {
 			// For levels > 0, mipmapping is always based on level 0.  Simpler to scale first.
 			tc0.s() *= 1.0f / (float)(1 << state.samplerID.width0Shift);
@@ -1268,8 +1271,8 @@ void DrawLine(const VertexData &v0, const VertexData &v1, const BinCoords &range
 				float s, s1;
 				float t, t1;
 				if (state.throughMode) {
-					Vec2<float> tc = (v0.texturecoords * (float)(steps - i) + v1.texturecoords * (float)i) / steps1;
-					Vec2<float> tc1 = (v0.texturecoords * (float)(steps - i - 1) + v1.texturecoords * (float)(i + 1)) / steps1;
+					Vec2<float> tc = (v0.texturecoords.uv() * (float)(steps - i) + v1.texturecoords.uv() * (float)i) / steps1;
+					Vec2<float> tc1 = (v0.texturecoords.uv() * (float)(steps - i - 1) + v1.texturecoords.uv() * (float)(i + 1)) / steps1;
 
 					s = tc.s() * (1.0f / (float)(1 << state.samplerID.width0Shift));
 					s1 = tc1.s() * (1.0f / (float)(1 << state.samplerID.width0Shift));
diff --git a/GPU/Software/RasterizerRectangle.cpp b/GPU/Software/RasterizerRectangle.cpp
index 3703678169..5cc40566d6 100644
--- a/GPU/Software/RasterizerRectangle.cpp
+++ b/GPU/Software/RasterizerRectangle.cpp
@@ -356,7 +356,7 @@ bool RectangleFastPath(const VertexData &v0, const VertexData &v1, BinManager &b
 	// Currently only works for TL/BR, which is the most common but not required.
 	bool orient_check = xdiff >= 0 && ydiff >= 0;
 	// We already have a fast path for clear in ClearRectangle.
-	bool state_check = state.throughMode && !state.pixelID.clearMode && !state.samplerID.hasAnyMips && NoClampOrWrap(state, v0.texturecoords) && NoClampOrWrap(state, v1.texturecoords);
+	bool state_check = state.throughMode && !state.pixelID.clearMode && !state.samplerID.hasAnyMips && NoClampOrWrap(state, v0.texturecoords.uv()) && NoClampOrWrap(state, v1.texturecoords.uv());
 	// This doesn't work well with offset drawing, see #15876.  Through never has a subpixel offset.
 	bool subpixel_check = ((v0.screenpos.x | v0.screenpos.y | v1.screenpos.x | v1.screenpos.y) & 0xF) == 0;
 	if ((coord_check || !state.enableTextures) && orient_check && state_check && subpixel_check) {
@@ -414,6 +414,9 @@ static bool AreCoordsRectangleCompatible(const RasterizerState &state, const Cli
 			if (data1.clippos.w - halftexel > data0.clippos.w || data1.clippos.w + halftexel < data0.clippos.w)
 				return false;
 		}
+		// If we're projecting textures, only allow an exact match for simplicity.
+		if (state.enableTextures && data1.v.texturecoords.q() != data0.v.texturecoords.q())
+			return false;
 		if (state.pixelID.applyFog && data1.v.fogdepth != data0.v.fogdepth) {
 			// Similar to w, this only matters if they're farther apart than 1/255.
 			static constexpr float foghalfstep = 0.5f / 255.0f;
diff --git a/GPU/Software/TransformUnit.cpp b/GPU/Software/TransformUnit.cpp
index 10a31c8bfb..9fd3540cf9 100644
--- a/GPU/Software/TransformUnit.cpp
+++ b/GPU/Software/TransformUnit.cpp
@@ -106,12 +106,14 @@ void SoftwareDrawEngine::DispatchSubmitImm(GEPrimitiveType prim, TransformedVert
 		vert.clippos = ClipCoords(buffer[i].pos);
 		vert.v.texturecoords.x = buffer[i].u;
 		vert.v.texturecoords.y = buffer[i].v;
+		vert.v.texturecoords.z = buffer[i].uv_w;
 		if (gstate.isModeThrough()) {
 			vert.v.texturecoords.x *= gstate.getTextureWidth(0);
 			vert.v.texturecoords.y *= gstate.getTextureHeight(0);
 		} else {
 			vert.clippos.z *= 1.0f / 65535.0f;
 		}
+		vert.v.clipw = buffer[i].pos_w;
 		vert.v.color0 = buffer[i].color0_32;
 		vert.v.color1 = gstate.isUsingSecondaryColor() && !gstate.isModeThrough() ? buffer[i].color1_32 : 0;
 		vert.v.fogdepth = buffer[i].fog;
@@ -324,9 +326,10 @@ ClipVertexData TransformUnit::ReadVertex(VertexReader &vreader, const TransformS
 	// VertexDecoder normally scales z, but we want it unscaled.
 	vreader.ReadPosThroughZ16(pos.AsArray());
 
-	static Vec2f lastTC;
+	static Vec3Packedf lastTC;
 	if (state.readUV) {
 		vreader.ReadUV(vertex.v.texturecoords.AsArray());
+		vertex.v.texturecoords.q() = 0.0f;
 		lastTC = vertex.v.texturecoords;
 	} else {
 		vertex.v.texturecoords = lastTC;
@@ -427,7 +430,7 @@ ClipVertexData TransformUnit::ReadVertex(VertexReader &vreader, const TransformS
 				break;
 
 			case GE_PROJMAP_UV:
-				source = Vec3f(vertex.v.texturecoords, 0.0f);
+				source = Vec3f(vertex.v.texturecoords.uv(), 0.0f);
 				break;
 
 			case GE_PROJMAP_NORMALIZED_NORMAL:
@@ -448,7 +451,7 @@ ClipVertexData TransformUnit::ReadVertex(VertexReader &vreader, const TransformS
 			// Note that UV scale/offset are not used in this mode.
 			Vec3<float> stq = Vec3ByMatrix43(source, gstate.tgenMatrix);
 			float z_recip = 1.0f / stq.z;
-			vertex.v.texturecoords = Vec2f(stq.x * z_recip, stq.y * z_recip);
+			vertex.v.texturecoords = Vec3Packedf(stq.x * z_recip, stq.y * z_recip, 1.0f);
 		} else if (state.uvGenMode == GE_TEXMAP_ENVIRONMENT_MAP) {
 			Lighting::GenerateLightST(vertex.v, worldnormal);
 		}
diff --git a/GPU/Software/TransformUnit.h b/GPU/Software/TransformUnit.h
index 9983bae101..b3d2e9b06d 100644
--- a/GPU/Software/TransformUnit.h
+++ b/GPU/Software/TransformUnit.h
@@ -79,12 +79,12 @@ struct DrawingCoords {
 };
 
 struct alignas(16) VertexData {
-	Vec2<float> texturecoords;
+	Vec3Packedf texturecoords;
+	float clipw;
 	uint32_t color0;
 	uint32_t color1;
 	ScreenCoords screenpos;
 	float fogdepth;
-	float clipw;
 };
 
 struct ClipVertexData {