diff --git a/desmume/src/OGLRender.cpp b/desmume/src/OGLRender.cpp
index d68ad57e0..47ac94427 100644
--- a/desmume/src/OGLRender.cpp
+++ b/desmume/src/OGLRender.cpp
@@ -821,13 +821,21 @@ static void OGLRender()
 
 	if(hasShaders)
 	{
+		//NOTE: this toon invalidation logic is hopelessly buggy.
+		//it may sometimes fail. it would be better to always recreate this data.
+		//but, that may be slow. since the cost of uploading that texture is huge in opengl (relative to rasterizer).
+		//someone please study it.
+		//here is a suggestion: it may make sense to memcmp the toon tables and upload only when it actually changes
 		if (gfx3d.renderState.invalidateToon)
 		{
 			glActiveTexture(GL_TEXTURE1);
 			glBindTexture(GL_TEXTURE_1D, oglToonTableTextureID);
 
-			glTexImage1D(GL_TEXTURE_1D, 0, GL_RGB, 32, 0, GL_RGBA, GL_UNSIGNED_BYTE, &gfx3d.renderState.rgbToonTable[0]);
-			gfx3d.renderState.invalidateToon = false;
+			u32 rgbToonTable[32];
+			for(int i=0;i<32;i++)
+				rgbToonTable[i] = RGB15TO32_NOALPHA(gfx3d.renderState.u16ToonTable[i]);
+			glTexImage1D(GL_TEXTURE_1D, 0, GL_RGB, 32, 0, GL_RGBA, GL_UNSIGNED_BYTE, &rgbToonTable[0]);
+			gfx3d.state.invalidateToon = false;
 		}
 	}
 
diff --git a/desmume/src/gfx3d.cpp b/desmume/src/gfx3d.cpp
index b68022314..e85254ad0 100644
--- a/desmume/src/gfx3d.cpp
+++ b/desmume/src/gfx3d.cpp
@@ -578,8 +578,6 @@ void gfx3d_reset()
 
 	GFX_PIPEclear();
 	GFX_FIFOclear();
-
-	gfx3d.state.invalidateToon = true;
 }
 
 
@@ -1611,16 +1609,16 @@ int gfx3d_GetNumVertex()
 
 void gfx3d_UpdateToonTable(u8 offset, u16 val)
 {
-	gfx3d.state.rgbToonTable[offset] = RGB15TO32(val, 255);
 	gfx3d.state.invalidateToon = true;
+	gfx3d.state.u16ToonTable[offset] = val;
 }
 
 void gfx3d_UpdateToonTable(u8 offset, u32 val)
 {
 	//C.O.P. sets toon table via this method
-	gfx3d.state.rgbToonTable[offset] = RGB15TO32(val & 0xFFFF, 255);
-	gfx3d.state.rgbToonTable[offset+1] = RGB15TO32(val >> 16, 255);
 	gfx3d.state.invalidateToon = true;
+	gfx3d.state.u16ToonTable[offset] = val & 0xFFFF;
+	gfx3d.state.u16ToonTable[offset+1] = val >> 16;
 }
 
 s32 gfx3d_GetClipMatrix (unsigned int index)
@@ -2329,7 +2327,7 @@ SFORMAT SF_GFX3D[]={
 	{ "GSCD", 4, 1, &gfx3d.state.clearDepth},
 	{ "GSFC", 4, 4, &gfx3d.state.fogColor},
 	{ "GSFO", 4, 1, &gfx3d.state.fogOffset},
-	{ "GST3", 4, 32, gfx3d.state.rgbToonTable},
+	{ "GST4", 2, 32, gfx3d.state.u16ToonTable},
 	{ "GSST", 4, 128, &gfx3d.state.shininessTable[0]},
 	{ "GSSI", 4, 1, &shininessInd},
 	{ "GSAF", 4, 1, &gfx3d.state.activeFlushCommand},
diff --git a/desmume/src/gfx3d.h b/desmume/src/gfx3d.h
index 90cc19c6e..044ca2ee0 100644
--- a/desmume/src/gfx3d.h
+++ b/desmume/src/gfx3d.h
@@ -103,6 +103,7 @@ inline u32 RGB15TO6665(u16 col, u8 alpha5)
 
 //produce a 15bpp color from individual 5bit components
 #define R5G5B5TORGB15(r,g,b) ((r)|((g)<<5)|((b)<<10))
+#define RGB15TO32_NOALPHA(col) ( color_15bit_to_24bit[col&0x7FFF] )
 
 //produce a 16bpp color from individual 5bit components
 #define R6G6B6TORGB15(r,g,b) ((r>>1)|((g&0x3E)<<4)|((b&0x3E)<<9))
@@ -317,8 +318,8 @@ struct GFX3D_State
 		for(u32 i=0;i<ARRAY_SIZE(shininessTable);i++)
 			shininessTable[i] = 0;
 
-		for(u32 i=0;i<ARRAY_SIZE(rgbToonTable);i++)
-			rgbToonTable[i] = 0;
+		for(u32 i=0;i<ARRAY_SIZE(u16ToonTable);i++)
+			u16ToonTable[i] = 0;
 	}
 
 	BOOL enableTexturing, enableAlphaTest, enableAlphaBlending, 
@@ -345,7 +346,7 @@ struct GFX3D_State
 	u32 fogShift;
 
 	bool invalidateToon;
-	u32 rgbToonTable[32];
+	u16 u16ToonTable[32];
 	float shininessTable[128];
 };
 
diff --git a/desmume/src/rasterize.cpp b/desmume/src/rasterize.cpp
index 529c93aca..aaa162960 100644
--- a/desmume/src/rasterize.cpp
+++ b/desmume/src/rasterize.cpp
@@ -1154,14 +1154,17 @@ void SoftRasterizerEngine::initFramebuffer(const int width, const int height, co
 
 void SoftRasterizerEngine::updateToonTable()
 {
-	if (!gfx3d.renderState.invalidateToon) return;
 	//convert the toon colors
 	for(int i=0;i<32;i++) {
-		toonTable[i].r = (gfx3d.renderState.rgbToonTable[i] >> 2) & 0x3F;
-		toonTable[i].g = (gfx3d.renderState.rgbToonTable[i] >> 10) & 0x3F;
-		toonTable[i].b = (gfx3d.renderState.rgbToonTable[i] >> 18) & 0x3F;
+		#ifdef WORDS_BIGENDIAN
+			u32 u32temp = RGB15TO32(gfx3d.renderState.u16ToonTable[i]);
+			toonTable[i].r = (u32temp >> 2) & 0x3F;
+			toonTable[i].g = (u32temp >> 10) & 0x3F;
+			toonTable[i].b = (u32temp >> 18) & 0x3F;
+		#else
+			toonTable[i].color = (RGB15TO32_NOALPHA(gfx3d.renderState.u16ToonTable[i])>>2)&0x3F3F3F3F;
+		#endif
 	}
-	gfx3d.renderState.invalidateToon = false;
 }
 
 void SoftRasterizerEngine::updateFogTable()