diff --git a/src/ds/gx/software.c b/src/ds/gx/software.c
index bdd71b698..a60d1ef54 100644
--- a/src/ds/gx/software.c
+++ b/src/ds/gx/software.c
@@ -6,6 +6,7 @@
 #include <mgba/internal/ds/gx/software.h>
 
 #include <mgba-util/memory.h>
+#include "gba/renderers/software-private.h"
 
 #define SCREEN_SIZE (DS_VIDEO_VERTICAL_PIXELS << 12)
 
@@ -101,6 +102,34 @@ static int _spanSort(const void* a, const void* b) {
 	return 0;
 }
 
+static color_t _lerpColor(const struct DSGXSoftwareSpan* span, unsigned x) {
+	int64_t width = span->x1 - span->x0;
+	int64_t xw = ((uint64_t) x << 12) - span->x0;
+	if (width) {
+		xw <<= 19;
+		xw /= width;
+	} else {
+		return 0; // TODO?
+	}
+	// Clamp to bounds
+	if (xw < 0) {
+		xw = 0;
+	} else if (xw > width) {
+		xw = width;
+	}
+	color_t r = ((int32_t) (span->cr1 - span->cr0) * xw) / width + span->cr0;
+	color_t g = ((int32_t) (span->cg1 - span->cg0) * xw) / width + span->cg0;
+	color_t b = ((int32_t) (span->cb1 - span->cb0) * xw) / width + span->cb0;
+#ifndef COLOR_16_BIT
+	color_t rgb = (r << 2) & 0xF8;
+	rgb |= (g << 10) & 0xF800;
+	rgb |= (b << 18) & 0xF80000;
+	return rgb;
+#else
+#error Unsupported color depth
+#endif
+}
+
 void DSGXSoftwareRendererCreate(struct DSGXSoftwareRenderer* renderer) {
 	renderer->d.init = DSGXSoftwareRendererInit;
 	renderer->d.reset = DSGXSoftwareRendererReset;
@@ -252,6 +281,38 @@ static void DSGXSoftwareRendererDrawScanline(struct DSGXRenderer* renderer, int
 		}
 	}
 	qsort(DSGXSoftwareSpanListGetPointer(&softwareRenderer->activeSpans, 0), DSGXSoftwareSpanListSize(&softwareRenderer->activeSpans), sizeof(struct DSGXSoftwareSpan), _spanSort);
+
+	y %= 48;
+	color_t* scanline = &softwareRenderer->scanlineCache[DS_VIDEO_HORIZONTAL_PIXELS * y];
+
+	size_t nextSpanX = DS_VIDEO_HORIZONTAL_PIXELS;
+	if (DSGXSoftwareSpanListSize(&softwareRenderer->activeSpans)) {
+		nextSpanX = DSGXSoftwareSpanListGetPointer(&softwareRenderer->activeSpans, DSGXSoftwareSpanListSize(&softwareRenderer->activeSpans) - 1)->x0;
+		nextSpanX >>= 12;
+	}
+	for (i = 0; i < DS_VIDEO_HORIZONTAL_PIXELS; ++i) {
+		struct DSGXSoftwareSpan* span = NULL;
+		if (i >= nextSpanX) {
+			size_t nextSpanId = DSGXSoftwareSpanListSize(&softwareRenderer->activeSpans);
+			span = DSGXSoftwareSpanListGetPointer(&softwareRenderer->activeSpans, nextSpanId - 1);
+			while (i > (uint32_t) (span->x1 >> 12)) {
+				DSGXSoftwareSpanListShift(&softwareRenderer->activeSpans, nextSpanId - 1, 1);
+				--nextSpanId;
+				if (!nextSpanId) {
+					nextSpanX = DS_VIDEO_HORIZONTAL_PIXELS;
+					span = NULL;
+					break;
+				}
+				span = DSGXSoftwareSpanListGetPointer(&softwareRenderer->activeSpans, nextSpanId - 1);
+				nextSpanX = span->x0 >> 12;
+			}
+		}
+		if (span) {
+			scanline[i] = _lerpColor(span, i);
+		} else {
+			scanline[i] = FLAG_UNWRITTEN; // TODO
+		}
+	}
 }
 
 static void DSGXSoftwareRendererGetScanline(struct DSGXRenderer* renderer, int y, color_t** output) {
diff --git a/src/ds/renderers/software.c b/src/ds/renderers/software.c
index 4bfab7017..a586ad4aa 100644
--- a/src/ds/renderers/software.c
+++ b/src/ds/renderers/software.c
@@ -385,7 +385,13 @@ static void DSVideoSoftwareRendererDrawGBAScanline(struct GBAVideoRenderer* rend
 				if (DSRegisterDISPCNTIs3D(softwareRenderer->dispcnt) && gx) {
 					color_t* scanline;
 					gx->renderer->getScanline(gx->renderer, y, &scanline);
-					// TODO
+					uint32_t flags = (softwareRenderer->bg[0].priority << OFFSET_PRIORITY) | FLAG_IS_BACKGROUND;
+					int x;
+					for (x = softwareRenderer->start; x < softwareRenderer->end; ++x) {
+						if ((scanline[x] & FLAG_UNWRITTEN) != FLAG_UNWRITTEN) {
+							_compositeNoBlendNoObjwin(softwareRenderer, &softwareRenderer->row[x], scanline[x] | flags, softwareRenderer->row[x]);
+						}
+					}
 				} else {
 					GBAVideoSoftwareRendererDrawBackgroundMode0(softwareRenderer, &softwareRenderer->bg[0], y);
 				}
diff --git a/src/ds/video.c b/src/ds/video.c
index 7075cc4d7..3141c0c88 100644
--- a/src/ds/video.c
+++ b/src/ds/video.c
@@ -201,6 +201,7 @@ void DSVideoAssociateRenderer(struct DSVideo* video, struct DSVideoRenderer* ren
 	memcpy(renderer->vramBBGExtPal, video->vramBBGExtPal, sizeof(renderer->vramBBGExtPal));
 	renderer->vramBOBJExtPal = video->vramBOBJExtPal;
 	renderer->oam = &video->oam;
+	renderer->gx = &video->p->gx;
 	video->renderer->init(video->renderer);
 }
 
@@ -321,8 +322,8 @@ void _startHblank9(struct mTiming* timing, void* context, uint32_t cyclesLate) {
 	dispstat = GBARegisterDISPSTATFillInHblank(dispstat);
 	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);
+			video->p->gx.renderer->drawScanline(video->p->gx.renderer, video->vcount + 48);
 		}
 		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);