diff --git a/include/mgba/internal/ds/gx.h b/include/mgba/internal/ds/gx.h
index 857c09ce3..004e13b18 100644
--- a/include/mgba/internal/ds/gx.h
+++ b/include/mgba/internal/ds/gx.h
@@ -97,8 +97,8 @@ struct DSGXVertex {
 	int32_t vz; // 16.16
 	int32_t vw; // 16.16
 
-	uint16_t color;
-	// Texcoords
+	// Color/Texcoords
+	uint16_t color; // 5.5.5
 	int16_t s; // 12.4
 	int16_t t; // 12.4
 };
diff --git a/include/mgba/internal/ds/gx/software.h b/include/mgba/internal/ds/gx/software.h
index 54aafc2c2..32edb07fa 100644
--- a/include/mgba/internal/ds/gx/software.h
+++ b/include/mgba/internal/ds/gx/software.h
@@ -12,6 +12,7 @@ CXX_GUARD_START
 
 #include <mgba/internal/ds/gx.h>
 #include <mgba/internal/ds/video.h>
+#include <mgba-util/table.h>
 #include <mgba-util/vector.h>
 
 struct DSGXSoftwarePolygon {
@@ -22,30 +23,55 @@ struct DSGXSoftwarePolygon {
 };
 
 struct DSGXSoftwareEdge {
-	struct DSGXPolygon* poly;
-	int32_t y0;
-	int32_t x0;
-	int32_t w0;
-	int32_t c0; // 6.6.6.6 ARGB
+	unsigned polyId;
+	int32_t y0; // 20.12
+	int32_t x0; // 20.12
+	int32_t w0; // 20.12
+	int8_t cr0;
+	int8_t cg0;
+	int8_t cb0;
 	int16_t s0;
 	int16_t t0;
 
-	int32_t y1;
-	int32_t x1;
-	int32_t w1;
-	int32_t c1; // 6.6.6.6 ARGB
+	int32_t y1; // 20.12
+	int32_t x1; // 20.12
+	int32_t w1; // 20.12
+	int8_t cr1;
+	int8_t cg1;
+	int8_t cb1;
+	int16_t s1;
+	int16_t t1;
+};
+
+struct DSGXSoftwareSpan {
+	int32_t x0; // 20.12
+	int32_t w0; // 20.12
+	int8_t cr0;
+	int8_t cg0;
+	int8_t cb0;
+	int16_t s0;
+	int16_t t0;
+
+	int32_t x1; // 20.12
+	int32_t w1; // 20.12
+	int8_t cr1;
+	int8_t cg1;
+	int8_t cb1;
 	int16_t s1;
 	int16_t t1;
 };
 
 DECLARE_VECTOR(DSGXSoftwarePolygonList, struct DSGXSoftwarePolygon);
 DECLARE_VECTOR(DSGXSoftwareEdgeList, struct DSGXSoftwareEdge);
+DECLARE_VECTOR(DSGXSoftwareSpanList, struct DSGXSoftwareSpan);
 
 struct DSGXSoftwareRenderer {
 	struct DSGXRenderer d;
 
 	struct DSGXSoftwarePolygonList activePolys;
 	struct DSGXSoftwareEdgeList activeEdges;
+	struct DSGXSoftwareSpanList activeSpans;
+	struct Table bucket;
 
 	uint16_t depthBuffer[DS_VIDEO_HORIZONTAL_PIXELS];
 	color_t* scanlineCache;
diff --git a/src/ds/gx.c b/src/ds/gx.c
index d04521bc4..24984f69e 100644
--- a/src/ds/gx.c
+++ b/src/ds/gx.c
@@ -171,8 +171,10 @@ static void _emitVertex(struct DSGX* gx, uint16_t x, uint16_t y, uint16_t z) {
 	gx->currentVertex.vz = _dotViewport(&gx->currentVertex, &gx->clipMatrix.m[2]);
 	gx->currentVertex.vw = _dotViewport(&gx->currentVertex, &gx->clipMatrix.m[3]);
 
-	gx->currentVertex.vx = (gx->currentVertex.vx + gx->currentVertex.vw) * gx->viewportWidth / (gx->currentVertex.vw * 2) + gx->viewportX1;
-	gx->currentVertex.vy = (gx->currentVertex.vy + gx->currentVertex.vw) * gx->viewportHeight / (gx->currentVertex.vw * 2) + gx->viewportY1;
+	// TODO: What to do if w is 0?
+
+	gx->currentVertex.vx = (gx->currentVertex.vx + gx->currentVertex.vw) * (int64_t) (gx->viewportWidth << 12) / (gx->currentVertex.vw * 2) + (gx->viewportX1 << 12);
+	gx->currentVertex.vy = (gx->currentVertex.vy + gx->currentVertex.vw) * (int64_t) (gx->viewportHeight << 12) / (gx->currentVertex.vw * 2) + (gx->viewportY1 << 12);
 
 	struct DSGXVertex* vbuf = gx->vertexBuffer[gx->bufferIndex];
 	vbuf[gx->vertexIndex] = gx->currentVertex;
diff --git a/src/ds/gx/software.c b/src/ds/gx/software.c
index cdb87443a..bdd71b698 100644
--- a/src/ds/gx/software.c
+++ b/src/ds/gx/software.c
@@ -7,8 +7,11 @@
 
 #include <mgba-util/memory.h>
 
+#define SCREEN_SIZE (DS_VIDEO_VERTICAL_PIXELS << 12)
+
 DEFINE_VECTOR(DSGXSoftwarePolygonList, struct DSGXSoftwarePolygon);
 DEFINE_VECTOR(DSGXSoftwareEdgeList, struct DSGXSoftwareEdge);
+DEFINE_VECTOR(DSGXSoftwareSpanList, struct DSGXSoftwareSpan);
 
 static void DSGXSoftwareRendererInit(struct DSGXRenderer* renderer);
 static void DSGXSoftwareRendererReset(struct DSGXRenderer* renderer);
@@ -17,20 +20,82 @@ static void DSGXSoftwareRendererSetRAM(struct DSGXRenderer* renderer, struct DSG
 static void DSGXSoftwareRendererDrawScanline(struct DSGXRenderer* renderer, int y);
 static void DSGXSoftwareRendererGetScanline(struct DSGXRenderer* renderer, int y, color_t** output);
 
+static void _expandColor(uint16_t c15, int8_t* r, int8_t* g, int8_t* b) {
+	*r = ((c15 << 1) & 0x3E) | 1;
+	*g = ((c15 >> 4) & 0x3E) | 1;
+	*b = ((c15 >> 9) & 0x3E) | 1;
+}
+
 static int _edgeSort(const void* a, const void* b) {
 	const struct DSGXSoftwareEdge* ea = a;
 	const struct DSGXSoftwareEdge* eb = b;
 
+	// Sort upside down
 	if (ea->y0 < eb->y0) {
-		return -1;
-	}
-	if (ea->y0 > eb->y0) {
 		return 1;
 	}
-	if (ea->y1 < eb->y1) {
+	if (ea->y0 > eb->y0) {
 		return -1;
 	}
+	if (ea->y1 < eb->y1) {
+		return 1;
+	}
 	if (ea->y1 > eb->y1) {
+		return -1;
+	}
+	return 0;
+}
+
+static bool _edgeToSpan(struct DSGXSoftwareSpan* span, const struct DSGXSoftwareEdge* edge, int index, int32_t y) {
+	// TODO: Perspective correction
+	int32_t height = edge->y1 - edge->y0;
+	int64_t yw = (y << 12) - edge->y0;
+	if (height) {
+		yw <<= 19;
+		yw /= height;
+	} else {
+		return false;
+	}
+	// Clamp to bounds
+	if (yw < 0) {
+		yw = 0;
+	} else if (yw > height) {
+		yw = height;
+	}
+	if (!index) {
+		span->x0 = ((int64_t) (edge->x1 - edge->x0) * yw) / height + edge->x0;
+		span->w0 = ((int64_t) (edge->w1 - edge->w0) * yw) / height + edge->w0;
+		span->cr0 = ((int32_t) (edge->cr1 - edge->cr0) * yw) / height + edge->cr0;
+		span->cg0 = ((int32_t) (edge->cg1 - edge->cg0) * yw) / height + edge->cg0;
+		span->cb0 = ((int32_t) (edge->cb1 - edge->cb0) * yw) / height + edge->cb0;
+		span->s0 = ((int32_t) (edge->s1 - edge->s0) * yw) / height + edge->s0;
+		span->t0 = ((int32_t) (edge->t1 - edge->t0) * yw) / height + edge->t0;
+	} else {
+		span->x1 = ((int64_t) (edge->x1 - edge->x0) * yw) / height + edge->x0;
+		span->w1 = ((int64_t) (edge->w1 - edge->w0) * yw) / height + edge->w0;
+		span->cr1 = ((int32_t) (edge->cr1 - edge->cr0) * yw) / height + edge->cr0;
+		span->cg1 = ((int32_t) (edge->cg1 - edge->cg0) * yw) / height + edge->cg0;
+		span->cb1 = ((int32_t) (edge->cb1 - edge->cb0) * yw) / height + edge->cb0;
+		span->s1 = ((int32_t) (edge->s1 - edge->s0) * yw) / height + edge->s0;
+		span->t1 = ((int32_t) (edge->t1 - edge->t0) * yw) / height + edge->t0;
+	}
+	return true;
+}
+
+static int _spanSort(const void* a, const void* b) {
+	const struct DSGXSoftwareSpan* sa = a;
+	const struct DSGXSoftwareSpan* sb = b;
+
+	if (sa->x0 < sb->x0) {
+		return -1;
+	}
+	if (sa->x0 > sb->x0) {
+		return 1;
+	}
+	if (sa->x1 < sb->x1) {
+		return -1;
+	}
+	if (sa->x1 > sb->x1) {
 		return 1;
 	}
 	return 0;
@@ -49,6 +114,8 @@ static void DSGXSoftwareRendererInit(struct DSGXRenderer* renderer) {
 	struct DSGXSoftwareRenderer* softwareRenderer = (struct DSGXSoftwareRenderer*) renderer;
 	DSGXSoftwarePolygonListInit(&softwareRenderer->activePolys, DS_GX_POLYGON_BUFFER_SIZE / 4);
 	DSGXSoftwareEdgeListInit(&softwareRenderer->activeEdges, DS_GX_POLYGON_BUFFER_SIZE);
+	DSGXSoftwareSpanListInit(&softwareRenderer->activeSpans, DS_GX_POLYGON_BUFFER_SIZE / 2);
+	TableInit(&softwareRenderer->bucket, DS_GX_POLYGON_BUFFER_SIZE / 8, NULL);
 	softwareRenderer->scanlineCache = anonymousMemoryMap(sizeof(color_t) * DS_VIDEO_HORIZONTAL_PIXELS * 48);
 }
 
@@ -61,6 +128,8 @@ static void DSGXSoftwareRendererDeinit(struct DSGXRenderer* renderer) {
 	struct DSGXSoftwareRenderer* softwareRenderer = (struct DSGXSoftwareRenderer*) renderer;
 	DSGXSoftwarePolygonListDeinit(&softwareRenderer->activePolys);
 	DSGXSoftwareEdgeListDeinit(&softwareRenderer->activeEdges);	
+	DSGXSoftwareSpanListDeinit(&softwareRenderer->activeSpans);
+	TableDeinit(&softwareRenderer->bucket);
 	mappedMemoryFree(softwareRenderer->scanlineCache, sizeof(color_t) * DS_VIDEO_HORIZONTAL_PIXELS * 48);
 }
 
@@ -75,6 +144,7 @@ static void DSGXSoftwareRendererSetRAM(struct DSGXRenderer* renderer, struct DSG
 		struct DSGXSoftwarePolygon* poly = DSGXSoftwarePolygonListAppend(&softwareRenderer->activePolys);
 		struct DSGXSoftwareEdge* edge = DSGXSoftwareEdgeListAppend(&softwareRenderer->activeEdges);
 		poly->poly = &polys[i];
+		edge->polyId = i;
 
 		struct DSGXVertex* v0 = &verts[poly->poly->vertIds[0]];
 		struct DSGXVertex* v1;
@@ -82,67 +152,68 @@ static void DSGXSoftwareRendererSetRAM(struct DSGXRenderer* renderer, struct DSG
 		int v;
 		for (v = 1; v < poly->poly->verts; ++v) {
 			v1 = &verts[poly->poly->vertIds[v]];
-			if (v0->vy <= v1->vy) {
-				edge->y0 = v0->vy;
+			if (v0->vy >= v1->vy) {
+				edge->y0 = SCREEN_SIZE - v0->vy;
 				edge->x0 = v0->vx;
 				edge->w0 = v0->vw;
-				edge->c0 = v0->color;
+				_expandColor(v0->color, &edge->cr0, &edge->cg0, &edge->cb0);
 				edge->s0 = v0->s;
 				edge->t0 = v0->t;
 
-				edge->y1 = v1->vy;
+				edge->y1 = SCREEN_SIZE - v1->vy;
 				edge->x1 = v1->vx;
 				edge->w1 = v1->vw;
-				edge->c1 = v1->color;
+				_expandColor(v1->color, &edge->cr1, &edge->cg1, &edge->cb1);
 				edge->s1 = v1->s;
 				edge->t1 = v1->t;
 			} else {
-				edge->y0 = v1->vy;
+				edge->y0 = SCREEN_SIZE - v1->vy;
 				edge->x0 = v1->vx;
 				edge->w0 = v1->vw;
-				edge->c0 = v1->color;
+				_expandColor(v1->color, &edge->cr0, &edge->cg0, &edge->cb0);
 				edge->s0 = v1->s;
 				edge->t0 = v1->t;
 
-				edge->y1 = v0->vy;
+				edge->y1 = SCREEN_SIZE - v0->vy;
 				edge->x1 = v0->vx;
 				edge->w1 = v0->vw;
-				edge->c1 = v0->color;
+				_expandColor(v0->color, &edge->cr1, &edge->cg1, &edge->cb1);
 				edge->s1 = v0->s;
 				edge->t1 = v0->t;
 			}
 
 			edge = DSGXSoftwareEdgeListAppend(&softwareRenderer->activeEdges);
+			edge->polyId = i;
 			v0 = v1;
 		}
 
 		v1 = &verts[poly->poly->vertIds[0]];
-		if (v0->vy <= v1->vy) {
-			edge->y0 = v0->vy;
+		if (v0->vy >= v1->vy) {
+			edge->y0 = SCREEN_SIZE - v0->vy;
 			edge->x0 = v0->vx;
 			edge->w0 = v0->vw;
-			edge->c0 = v0->color;
+			_expandColor(v0->color, &edge->cr0, &edge->cg0, &edge->cb0);
 			edge->s0 = v0->s;
 			edge->t0 = v0->t;
 
-			edge->y1 = v1->vy;
+			edge->y1 = SCREEN_SIZE - v1->vy;
 			edge->x1 = v1->vx;
 			edge->w1 = v1->vw;
-			edge->c1 = v1->color;
+			_expandColor(v1->color, &edge->cr1, &edge->cg1, &edge->cb1);
 			edge->s1 = v1->s;
 			edge->t1 = v1->t;
 		} else {
-			edge->y0 = v1->vy;
+			edge->y0 = SCREEN_SIZE - v1->vy;
 			edge->x0 = v1->vx;
 			edge->w0 = v1->vw;
-			edge->c0 = v1->color;
+			_expandColor(v1->color, &edge->cr0, &edge->cg0, &edge->cb0);
 			edge->s0 = v1->s;
 			edge->t0 = v1->t;
 
-			edge->y1 = v0->vy;
+			edge->y1 = SCREEN_SIZE - v0->vy;
 			edge->x1 = v0->vx;
 			edge->w1 = v0->vw;
-			edge->c1 = v0->color;
+			_expandColor(v0->color, &edge->cr1, &edge->cg1, &edge->cb1);
 			edge->s1 = v0->s;
 			edge->t1 = v0->t;
 		}
@@ -152,7 +223,35 @@ static void DSGXSoftwareRendererSetRAM(struct DSGXRenderer* renderer, struct DSG
 
 static void DSGXSoftwareRendererDrawScanline(struct DSGXRenderer* renderer, int y) {
 	struct DSGXSoftwareRenderer* softwareRenderer = (struct DSGXSoftwareRenderer*) renderer;
-	// TODO
+	DSGXSoftwareSpanListClear(&softwareRenderer->activeSpans);
+	TableClear(&softwareRenderer->bucket);
+	size_t i;
+	for (i = DSGXSoftwareEdgeListSize(&softwareRenderer->activeEdges); i; --i) {
+		size_t idx = i - 1;
+		struct DSGXSoftwareEdge* edge = DSGXSoftwareEdgeListGetPointer(&softwareRenderer->activeEdges, idx);
+		if (edge->y1 >> 12 < y) {
+			DSGXSoftwareEdgeListShift(&softwareRenderer->activeEdges, idx, 1);
+			continue;
+		} else if (edge->y0 >> 12 > y) {
+			continue;
+		}
+
+		unsigned poly = edge->polyId;
+		struct DSGXSoftwareSpan* span = TableLookup(&softwareRenderer->bucket, poly);
+		if (span) {
+			_edgeToSpan(span, edge, 1, y);
+			TableRemove(&softwareRenderer->bucket, poly);
+		} else {
+			span = DSGXSoftwareSpanListAppend(&softwareRenderer->activeSpans);
+			if (!_edgeToSpan(span, edge, 0, y)) {
+				// Horizontal line
+				DSGXSoftwareSpanListShift(&softwareRenderer->activeSpans, DSGXSoftwareSpanListSize(&softwareRenderer->activeSpans) - 1, 1);
+			} else {
+				TableInsert(&softwareRenderer->bucket, poly, span);
+			}
+		}
+	}
+	qsort(DSGXSoftwareSpanListGetPointer(&softwareRenderer->activeSpans, 0), DSGXSoftwareSpanListSize(&softwareRenderer->activeSpans), sizeof(struct DSGXSoftwareSpan), _spanSort);
 }
 
 static void DSGXSoftwareRendererGetScanline(struct DSGXRenderer* renderer, int y, color_t** output) {
diff --git a/src/ds/video.c b/src/ds/video.c
index 38ee90fdf..7075cc4d7 100644
--- a/src/ds/video.c
+++ b/src/ds/video.c
@@ -319,8 +319,14 @@ void _startHblank9(struct mTiming* timing, void* context, uint32_t cyclesLate) {
 
 	// Begin Hblank
 	dispstat = GBARegisterDISPSTATFillInHblank(dispstat);
-	if (video->vcount < DS_VIDEO_VERTICAL_PIXELS && video->frameskipCounter <= 0) {
-		video->renderer->drawScanline(video->renderer, video->vcount);
+	if (video->frameskipCounter <= 0) {
+		if (video->vcount < DS_VIDEO_VERTICAL_PIXELS) {
+			video->p->gx.renderer->drawScanline(video->p->gx.renderer, video->vcount + 48);
+			video->renderer->drawScanline(video->renderer, video->vcount);
+		}
+		if (video->vcount >= DS_VIDEO_VERTICAL_TOTAL_PIXELS - 48) {
+			video->p->gx.renderer->drawScanline(video->p->gx.renderer, video->vcount + 48 - DS_VIDEO_VERTICAL_TOTAL_PIXELS);
+		}
 	}
 
 	if (GBARegisterDISPSTATIsHblankIRQ(dispstat)) {