I think I finally understood how origin-relative GE jumps are supposed to work.

GPU/GLES/DisplayListInterpreter.cpp

@@ -314,17 +314,14 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 	// Handle control and drawing commands here directly. The others we delegate.
 	switch (cmd) {
 	case GE_CMD_BASE:
-		//DEBUG_LOG(G3D,"DL BASE: %06x", data & 0xFFFFFF);
 		break;
 
 	case GE_CMD_VADDR:		/// <<8????
 		gstate_c.vertexAddr = ((gstate.base & 0x00FF0000) << 8)|data;
-		//DEBUG_LOG(G3D,"DL VADDR: %06x", gstate_c.vertexAddr);
 		break;
 
 	case GE_CMD_IADDR:
 		gstate_c.indexAddr	= ((gstate.base & 0x00FF0000) << 8)|data;
-		//DEBUG_LOG(G3D,"DL IADDR: %06x", gstate_c.indexAddr);
 		break;
 
 	case GE_CMD_PRIM:
@@ -388,23 +385,24 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 
 	case GE_CMD_JUMP:
 		{
-			u32 target = (((gstate.base & 0x00FF0000) << 8) | (op & 0xFFFFFC)) & 0x0FFFFFFF;
+			u32 target = gstate_c.getJumpAddress(data);
 			if (Memory::IsValidAddress(target)) {
 				currentList->pc = target - 4; // pc will be increased after we return, counteract that
 			} else {
-				ERROR_LOG(G3D, "JUMP to illegal address %08x - ignoring??", target);
+				ERROR_LOG(G3D, "JUMP to illegal address %08x - ignoring! data=%06x", target, data);
 			}
 		}
 		break;
 
 	case GE_CMD_CALL:
 		{
+			// Saint Seiya needs correct support for relative calls.
 			u32 retval = currentList->pc + 4;
-			u32 target = (((gstate.base & 0x00FF0000) << 8) | (op & 0xFFFFFC)) & 0xFFFFFFF;
+			u32 target = gstate_c.getJumpAddress(data);
 			if (stackptr == ARRAY_SIZE(stack)) {
 				ERROR_LOG(G3D, "CALL: Stack full!");
 			} else if (!Memory::IsValidAddress(target)) {
-				ERROR_LOG(G3D, "CALL to illegal address %08x - ignoring??", target);
+				ERROR_LOG(G3D, "CALL to illegal address %08x - ignoring! data=%06x", target, data);
 			} else {
 				stack[stackptr++] = retval;
 				currentList->pc = target - 4;	// pc will be increased after we return, counteract that
@@ -418,6 +416,7 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 				ERROR_LOG(G3D, "RET: Stack empty!");
 			} else {
 				u32 target = (currentList->pc & 0xF0000000) | (stack[--stackptr] & 0x0FFFFFFF);
+				//target = (target + gstate_c.originAddr) & 0xFFFFFFF;
 				currentList->pc = target - 4;
 				if (!Memory::IsValidAddress(currentList->pc)) {
 					ERROR_LOG(G3D, "Invalid DL PC %08x on return", currentList->pc);
@@ -427,6 +426,16 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 		}
 		break;
 
+	case GE_CMD_OFFSETADDR:
+		gstate_c.offsetAddr = data << 8;
+		// ???
+		break;
+
+	case GE_CMD_ORIGIN:
+		gstate_c.offsetAddr = currentList->pc;
+		break;
+
+
 	case GE_CMD_SIGNAL:
 		{
 			// Processed in GE_END. Has data.
@@ -495,10 +504,6 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 		// bounding box test. Let's do nothing.
 		break;
 
-	case GE_CMD_ORIGIN:
-		gstate.offsetAddr = currentList->pc & 0xFFFFFF;
-		break;
-
 	case GE_CMD_VERTEXTYPE:
 		if (diff & GE_VTYPE_THROUGH) {
 			// Throughmode changed, let's make the proj matrix dirty.
@@ -507,10 +512,6 @@ void GLES_GPU::ExecuteOp(u32 op, u32 diff) {
 		// This sets through-mode or not, as well.
 		break;
 
-	case GE_CMD_OFFSETADDR:
-		//			offsetAddr = data<<8;
-		break;
-
 	case GE_CMD_REGION1:
 		{
 			int x1 = data & 0x3ff;

GPU/GPUState.cpp

@@ -31,6 +31,7 @@ GPUStatistics gpuStats;
 void InitGfxState()
 {
 	memset(&gstate, 0, sizeof(gstate));
+	memset(&gstate_c, 0, sizeof(gstate_c));
 	for (int i = 0; i < 256; i++) {
 		gstate.cmdmem[i] = i << 24;
 	}

GPU/GPUState.h

@@ -231,6 +231,8 @@ struct GPUStateCache
 	u32 vertexAddr;
 	u32 indexAddr;
 
+	u32 offsetAddr;
+
 	bool textureChanged;
 
 	float uScale,vScale,zScale;
@@ -247,6 +249,9 @@ struct GPUStateCache
 
 	float vpWidth;
 	float vpHeight;
+
+
+	u32 getJumpAddress(u32 data) const;
 };
 
 // TODO: Implement support for these.
@@ -322,3 +327,8 @@ extern GPUgstate gstate;
 extern GPUStateCache gstate_c;
 extern GPUInterface *gpu;
 extern GPUStatistics gpuStats;
+
+inline u32 GPUStateCache::getJumpAddress(u32 data) const {
+	u32 baseExtended = ((gstate.base & 0x0F0000) << 8) | (data & 0xFFFFFF);
+	return (gstate_c.offsetAddr + baseExtended) & 0x0FFFFFFF;
+}

GPU/GeDisasm.cpp

@@ -85,7 +85,7 @@ void GeDisassembleOp(u32 pc, u32 op, u32 prev, char *buffer) {
 	case GE_CMD_CALL:
 		{
 			u32 retval = pc + 4;
-			u32 target = (((gstate.base & 0x00FF0000) << 8) | (op & 0xFFFFFC)) & 0xFFFFFFF;
+			u32 target = gstate_c.getJumpAddress(op & 0xFFFFFF);
 			sprintf(buffer, "CMD CALL - %08x to %08x, ret=%08x", pc, target, retval);
 		}
 		break;

GPU/Null/NullGpu.cpp

@@ -169,6 +169,7 @@ void NullGPU::ExecuteOp(u32 op, u32 diff)
 
 	case GE_CMD_ORIGIN:
 		gstate.offsetAddr = currentList->pc & 0xFFFFFF;
+		gstate_c.offsetAddr = currentList->pc;
 		break;
 
 	case GE_CMD_VERTEXTYPE: