g3dvl: Use R16_SNORM instead of A8L8_UNORM for block rendering.

3 files changed, 153 insertions, 276 deletions

diff --git a/src/gallium/state_trackers/g3dvl/vl_context.c b/src/gallium/state_trackers/g3dvl/vl_context.c
index 638900b3f4..542ba996a7 100644
--- a/src/gallium/state_trackers/g3dvl/vl_context.c
+++ b/src/gallium/state_trackers/g3dvl/vl_context.c
@@ -365,7 +365,7 @@ static int vlCreateVertexShaderIMC(struct VL_CONTEXT *context)
 
 static int vlCreateFragmentShaderIMC(struct VL_CONTEXT *context)
 {
-	const unsigned int		max_tokens = 50;
+	const unsigned int		max_tokens = 100;
 	
 	struct pipe_context		*pipe;
 	struct pipe_shader_state	fs;
@@ -402,11 +402,19 @@ static int vlCreateFragmentShaderIMC(struct VL_CONTEXT *context)
 		decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
 		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
 	}
+	
+	/* decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
+	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
+	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
 
 	/* decl o0			; Fragment color */
 	decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
 	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
 	
+	/* decl t0, t1 */
+	decl = vl_decl_temps(0, 1);
+	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+	
 	/*
 	 * decl s0			; Sampler for luma texture
 	 * decl s1			; Sampler for chroma Cb texture
@@ -419,16 +427,30 @@ static int vlCreateFragmentShaderIMC(struct VL_CONTEXT *context)
 	}
 	
 	/*
-	 * tex2d o0.x, i0, s0		; Read texel from luma texture into .x channel
-	 * tex2d o0.y, i1, s1		; Read texel from chroma Cb texture into .y channel
-	 * tex2d o0.z, i1, s2		; Read texel from chroma Cr texture into .z channel
-	*/
+	 * tex2d t1, i0, s0		; Read texel from luma texture
+	 * mov t0.x, t1.x		; Move luma sample into .x component
+	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
+	 * mov t0.y, t1.x		; Move Cb sample into .y component
+	 * tex2d t1, i1, s2		; Read texel from chroma Cr texture
+	 * mov t0.z, t1.x		; Move Cr sample into .z component
+	 */
 	for (i = 0; i < 3; ++i)
 	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_OUTPUT, 0, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
+		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
+		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+		
+		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
 		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+		
 	}
+	
+	/* mul o0, t0, c0		; Rescale texel to correct range */
+	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
+	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 
 	/* end */
 	inst = vl_end();
@@ -701,10 +723,7 @@ static int vlCreateFragmentShaderFramePMC(struct VL_CONTEXT *context)
 		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
 	}
 	
-	/*
-	 * decl c0			; Multiplier to shift 9th bit of differential into place
-	 * decl c1			; Bias to get differential back to a signed value
-	 */
+	/* decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
 	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
 	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
 
@@ -729,38 +748,29 @@ static int vlCreateFragmentShaderFramePMC(struct VL_CONTEXT *context)
 	}
 	
 	/*
-	 * tex2d t0.xw, i0, s0		; Read texel from luma texture into .x and .w channels
-	 * mov t1.x, t0.w		; Move 9th bit from .w channel to .x
-	 * tex2d t0.yw, i1, s1		; Read texel from chroma Cb texture into .y and .w channels
-	 * mov t1.y, t0.w		; Move 9th bit from .w channel to .y
-	 * tex2d t0.zw, i1, s2		; Read texel from chroma Cr texture into .z and .w channels
-	 * mov t1.z, t0.w		; Move 9th bit from .w channel to .z
+	 * tex2d t1, i0, s0		; Read texel from luma texture
+	 * mov t0.x, t1.x		; Move luma sample into .x component
+	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
+	 * mov t0.y, t1.x		; Move Cb sample into .y component
+	 * tex2d t1, i1, s2		; Read texel from chroma Cr texture
+	 * mov t0.z, t1.x		; Move Cr sample into .z component
 	 */
 	for (i = 0; i < 3; ++i)
 	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
-		inst.FullDstRegisters[0].DstRegister.WriteMask = (TGSI_WRITEMASK_X << i) | TGSI_WRITEMASK_W;
+		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 		
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 0);
+		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
 		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_W;
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+		
 	}
 	
-	/* mul t1, t1, c0		; Muliply 9th bit by multiplier to shift it into place */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	
-	/* add t0, t0, t1		; Add luma and chroma low and high parts to get normalized unsigned 9-bit values */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	
-	/* sub t0, t0, c1		; Subtract bias to get back signed values */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
+	/* mul t0, t0, c0		; Rescale texel to correct range */
+	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
 	/* tex2d t1, i2, s3		; Read texel from ref macroblock */
@@ -826,11 +836,10 @@ static int vlCreateFragmentShaderFieldPMC(struct VL_CONTEXT *context)
 	}
 	
 	/*
-	 * decl c0			; Multiplier to shift 9th bit of differential into place
-	 * decl c1			; Bias to get differential back to a signed value
-	 * decl c2			; Constants 1/2 & 2 in .x, .y channels for Y-mod-2 top/bottom field selection
+	 * decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm
+	 * decl c1			; Constants 1/2 & 2 in .x, .y channels for Y-mod-2 top/bottom field selection
 	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 2);
+	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
 	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
 
 	/* decl o0			; Fragment color */
@@ -854,42 +863,35 @@ static int vlCreateFragmentShaderFieldPMC(struct VL_CONTEXT *context)
 	}
 	
 	/*
-	 * tex2d t0.xw, i0, s0		; Read texel from luma texture into .x and .w channels
-	 * mov t1.x, t0.w		; Move 9th bit from .w channel to .x
-	 * tex2d t0.yw, i1, s1		; Read texel from chroma Cb texture into .y and .w channels
-	 * mov t1.y, t0.w		; Move 9th bit from .w channel to .y
-	 * tex2d t0.zw, i1, s2		; Read texel from chroma Cr texture into .z and .w channels
-	 * mov t1.z, t0.w		; Move 9th bit from .w channel to .z
+	 * tex2d t1, i0, s0		; Read texel from luma texture
+	 * mov t0.x, t1.x		; Move luma sample into .x component
+	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
+	 * mov t0.y, t1.x		; Move Cb sample into .y component
+	 * tex2d t1, i1, s2		; Read texel from chroma Cr texture
+	 * mov t0.z, t1.x		; Move Cr sample into .z component
 	 */
 	for (i = 0; i < 3; ++i)
 	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
-		inst.FullDstRegisters[0].DstRegister.WriteMask = (TGSI_WRITEMASK_X << i) | TGSI_WRITEMASK_W;
+		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 		
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 0);
+		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
 		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_W;
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+		
 	}
 	
-	/* mul t1, t1, c0		; Muliply 9th bit by multiplier to shift it into place */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	
-	/* add t0, t0, t1		; Add luma and chroma low and high parts to get normalized unsigned 9-bit values */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	
-	/* sub t0, t0, c1		; Subtract bias to get back signed values */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
+	/* mul t0, t0, c0		; Rescale texel to correct range */
+	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
-	/* tex2d t1, i2, s3		; Read texel from ref macroblock top field
-	   tex2d t2, i3, s3		; Read texel from ref macroblock bottom field */
+	/*
+	 * tex2d t1, i2, s3		; Read texel from ref macroblock top field
+	 * tex2d t2, i3, s3		; Read texel from ref macroblock bottom field
+	 */
 	for (i = 0; i < 2; ++i)
 	{
 		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 2, TGSI_FILE_SAMPLER, 3);
@@ -897,8 +899,8 @@ static int vlCreateFragmentShaderFieldPMC(struct VL_CONTEXT *context)
 	}
 	
 	/* XXX: Pos values off by 0.5? */
-	/* sub t4, i4.y, c2.x		; Sub 0.5 from denormalized pos */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 4, TGSI_FILE_INPUT, 2);
+	/* sub t4, i4.y, c1.x		; Sub 0.5 from denormalized pos */
+	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 4, TGSI_FILE_CONSTANT, 1);
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
@@ -909,8 +911,8 @@ static int vlCreateFragmentShaderFieldPMC(struct VL_CONTEXT *context)
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
-	/* mul t3, t4, c2.x		; Multiply pos Y-coord by 1/2 */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 2);
+	/* mul t3, t4, c1.x		; Multiply pos Y-coord by 1/2 */
+	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 1);
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
@@ -921,8 +923,8 @@ static int vlCreateFragmentShaderFieldPMC(struct VL_CONTEXT *context)
 	inst = vl_inst2(TGSI_OPCODE_FLOOR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
-	/* mul t3, t3, c2.y		; Multiply by 2 */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 2);
+	/* mul t3, t3, c1.y		; Multiply by 2 */
+	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 1);
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
@@ -1224,11 +1226,10 @@ static int vlCreateFragmentShaderFrameBMC(struct VL_CONTEXT *context)
 	}
 	
 	/*
-	 * decl c0			; Multiplier to shift 9th bit of differential into place
-	 * decl c1			; Bias to get differential back to a signed value
-	 * decl c2			; Constant 1/2 in .x channel to use as weight to blend past and future texels
+	 * decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm
+	 * decl c1			; Constant 1/2 in .x channel to use as weight to blend past and future texels
 	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 2);
+	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
 	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
 
 	/* decl o0			; Fragment color */
@@ -1253,38 +1254,29 @@ static int vlCreateFragmentShaderFrameBMC(struct VL_CONTEXT *context)
 	}
 	
 	/*
-	 * tex2d t0.xw, i0, s0		; Read texel from luma texture into .x and .w channels
-	 * mov t1.x, t0.w		; Move 9th bit from .w channel to .x
-	 * tex2d t0.yw, i1, s1		; Read texel from chroma Cb texture into .y and .w channels
-	 * mov t1.y, t0.w		; Move 9th bit from .w channel to .y
-	 * tex2d t0.zw, i1, s2		; Read texel from chroma Cr texture into .z and .w channels
-	 * mov t1.z, t0.w		; Move 9th bit from .w channel to .z
+	 * tex2d t1, i0, s0		; Read texel from luma texture
+	 * mov t0.x, t1.x		; Move luma sample into .x component
+	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
+	 * mov t0.y, t1.x		; Move Cb sample into .y component
+	 * tex2d t1, i1, s2		; Read texel from chroma Cr texture
+	 * mov t0.z, t1.x		; Move Cr sample into .z component
 	 */
 	for (i = 0; i < 3; ++i)
 	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
-		inst.FullDstRegisters[0].DstRegister.WriteMask = (TGSI_WRITEMASK_X << i) | TGSI_WRITEMASK_W;
+		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 		
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 0);
+		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
 		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_W;
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+		
 	}
 	
-	/* mul t1, t1, c0		; Muliply 9th bit by multiplier to shift it into place */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	
-	/* add t0, t0, t1		; Add luma and chroma low and high parts to get normalized unsigned 9-bit values */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	
-	/* sub t0, t0, c1		; Subtract bias to get back signed values */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
+	/* mul t0, t0, c0		; Rescale texel to correct range */
+	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
 	/*
@@ -1297,8 +1289,8 @@ static int vlCreateFragmentShaderFrameBMC(struct VL_CONTEXT *context)
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	}
 	
-	/* lerp t1, c2.x, t1, t2	; Blend past and future texels */
-	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 2, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
+	/* lerp t1, c1.x, t1, t2	; Blend past and future texels */
+	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
@@ -1366,12 +1358,11 @@ static int vlCreateFragmentShaderFieldBMC(struct VL_CONTEXT *context)
 	}
 	
 	/*
-	 * decl c0			; Multiplier to shift 9th bit of differential into place
-	 * decl c1			; Bias to get differential back to a signed value
-	 * decl c2			; Constants 1/2 & 2 in .x, .y channels to use as weight to blend past and future texels
+	 * decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm
+	 * decl c1			; Constants 1/2 & 2 in .x, .y channels to use as weight to blend past and future texels
 	 *				; and for Y-mod-2 top/bottom field selection
 	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 2);
+	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
 	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
 
 	/* decl o0			; Fragment color */
@@ -1396,43 +1387,34 @@ static int vlCreateFragmentShaderFieldBMC(struct VL_CONTEXT *context)
 	}
 	
 	/*
-	 * tex2d t0.xw, i0, s0		; Read texel from luma texture into .x and .w channels
-	 * mov t1.x, t0.w		; Move 9th bit from .w channel to .x
-	 * tex2d t0.yw, i1, s1		; Read texel from chroma Cb texture into .y and .w channels
-	 * mov t1.y, t0.w		; Move 9th bit from .w channel to .y
-	 * tex2d t0.zw, i1, s2		; Read texel from chroma Cr texture into .z and .w channels
-	 * mov t1.z, t0.w		; Move 9th bit from .w channel to .z
+	 * tex2d t1, i0, s0		; Read texel from luma texture
+	 * mov t0.x, t1.x		; Move luma sample into .x component
+	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
+	 * mov t0.y, t1.x		; Move Cb sample into .y component
+	 * tex2d t1, i1, s2		; Read texel from chroma Cr texture
+	 * mov t0.z, t1.x		; Move Cr sample into .z component
 	 */
 	for (i = 0; i < 3; ++i)
 	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
-		inst.FullDstRegisters[0].DstRegister.WriteMask = (TGSI_WRITEMASK_X << i) | TGSI_WRITEMASK_W;
+		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i > 0 ? 1 : 0, TGSI_FILE_SAMPLER, i);
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 		
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 0);
+		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
 		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_W;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_W;
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+		
 	}
 	
-	/* mul t1, t1, c0		; Muliply 9th bit by multiplier to shift it into place */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	
-	/* add t0, t0, t1		; Add luma and chroma low and high parts to get normalized unsigned 9-bit values */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	
-	/* sub t0, t0, c1		; Subtract bias to get back signed values */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
+	/* mul t0, t0, c0		; Rescale texel to correct range */
+	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
 	/* XXX: Pos values off by 0.5? */
-	/* sub t4, i6.y, c2.x		; Sub 0.5 from denormalized pos */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 6, TGSI_FILE_CONSTANT, 2);
+	/* sub t4, i6.y, c1.x		; Sub 0.5 from denormalized pos */
+	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 6, TGSI_FILE_CONSTANT, 1);
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
@@ -1443,8 +1425,8 @@ static int vlCreateFragmentShaderFieldBMC(struct VL_CONTEXT *context)
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
-	/* mul t3, t4, c2.x		; Multiply pos Y-coord by 1/2 */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 2);
+	/* mul t3, t4, c1.x		; Multiply pos Y-coord by 1/2 */
+	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 1);
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
@@ -1455,8 +1437,8 @@ static int vlCreateFragmentShaderFieldBMC(struct VL_CONTEXT *context)
 	inst = vl_inst2(TGSI_OPCODE_FLOOR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
-	/* mul t3, t3, c2.y		; Multiply by 2 */
-	inst = vl_inst3( TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 2);
+	/* mul t3, t3, c1.y		; Multiply by 2 */
+	inst = vl_inst3( TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 1);
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
 	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
@@ -1497,8 +1479,8 @@ static int vlCreateFragmentShaderFieldBMC(struct VL_CONTEXT *context)
 	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 2, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 5);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	
-	/* lerp t1, c2.x, t1, t2	; Blend past and future texels */
-	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 2, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
+	/* lerp t1, c1.x, t1, t2	; Blend past and future texels */
+	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
@@ -1678,7 +1660,7 @@ static int vlInitMC(struct VL_CONTEXT *context)
 	
 	memset(&template, 0, sizeof(struct pipe_texture));
 	template.target = PIPE_TEXTURE_2D;
-	template.format = PIPE_FORMAT_A8L8_UNORM;
+	template.format = PIPE_FORMAT_R16_SNORM;
 	template.last_level = 0;
 	template.width[0] = 8;
 	template.height[0] = 8 * 4;
diff --git a/src/gallium/state_trackers/g3dvl/vl_data.c b/src/gallium/state_trackers/g3dvl/vl_data.c
index 7e6ee8ac12..0e5c8c77f9 100644
--- a/src/gallium/state_trackers/g3dvl/vl_data.c
+++ b/src/gallium/state_trackers/g3dvl/vl_data.c
@@ -86,8 +86,7 @@ const struct VL_TEXCOORD2F *vl_surface_texcoords = (const struct VL_TEXCOORD2F*)
  */
 const struct VL_MC_FS_CONSTS vl_mc_fs_consts =
 {
-	{256.0f, 256.0f, 256.0f, 0.0f},
-	{256.0f / 255.0f, 256.0f / 255.0f, 256.0f / 255.0f, 0.0f},
+	{32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f},
 	{0.5f, 2.0f, 0.0f, 0.0f}
 };
 
diff --git a/src/gallium/state_trackers/g3dvl/vl_surface.c b/src/gallium/state_trackers/g3dvl/vl_surface.c
index 9b91ab4e22..6d4e14b95c 100644
--- a/src/gallium/state_trackers/g3dvl/vl_surface.c
+++ b/src/gallium/state_trackers/g3dvl/vl_surface.c
@@ -63,7 +63,7 @@ static int vlTransformBlock(short *src, short *dst, short bias)
 	return 0;
 }
 
-static int vlGrabFrameCodedFullBlock(short *src, short *dst, unsigned int dst_pitch)
+static int vlGrabFrameCodedBlock(short *src, short *dst, unsigned int dst_pitch)
 {
 	unsigned int y;
 	
@@ -78,18 +78,7 @@ static int vlGrabFrameCodedFullBlock(short *src, short *dst, unsigned int dst_pi
 	return 0;
 }
 
-static int vlGrabFrameCodedDiffBlock(short *src, short *dst, unsigned int dst_pitch)
-{
-	unsigned int x, y;
-	
-	for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
-		for (x = 0; x < VL_BLOCK_WIDTH; ++x)
-			dst[y * dst_pitch + x] = src[y * VL_BLOCK_WIDTH + x] + 0x100;
-	
-	return 0;
-}
-
-static int vlGrabFieldCodedFullBlock(short *src, short *dst, unsigned int dst_pitch)
+static int vlGrabFieldCodedBlock(short *src, short *dst, unsigned int dst_pitch)
 {
 	unsigned int y;
 	
@@ -114,30 +103,17 @@ static int vlGrabFieldCodedFullBlock(short *src, short *dst, unsigned int dst_pi
 	return 0;
 }
 
-static int vlGrabFieldCodedDiffBlock(short *src, short *dst, unsigned int dst_pitch)
-{
-	unsigned int x, y;
-	
-	for (y = 0; y < VL_BLOCK_HEIGHT / 2; ++y)
-		for (x = 0; x < VL_BLOCK_WIDTH; ++x)
-			dst[y * dst_pitch * 2 + x] = src[y * VL_BLOCK_WIDTH + x] + 0x100;
-	
-	dst += VL_BLOCK_HEIGHT * dst_pitch;
-	
-	for (; y < VL_BLOCK_HEIGHT; ++y)
-		for (x = 0; x < VL_BLOCK_WIDTH; ++x)
-			dst[y * dst_pitch * 2 + x] = src[y * VL_BLOCK_WIDTH + x] + 0x100;
-	
-	return 0;
-}
-
 static int vlGrabNoBlock(short *dst, unsigned int dst_pitch)
 {
-	unsigned int x, y;
+	unsigned int y;
 	
 	for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
-		for (x = 0; x < VL_BLOCK_WIDTH; ++x)
-			dst[y * dst_pitch + x] = 0x100;
+		memset
+		(
+			dst + y * dst_pitch,
+			0,
+			VL_BLOCK_WIDTH * 2
+		);
 	
 	return 0;
 }
@@ -156,7 +132,6 @@ static int vlGrabBlocks
 	unsigned int		tex_pitch;
 	unsigned int		tb, sb = 0;
 	
-	const int		do_idct = 1;
 	short			temp_block[64];
 	
 	assert(context);
@@ -176,80 +151,26 @@ static int vlGrabBlocks
 	{
 		if ((coded_block_pattern >> (5 - tb)) & 1)
 		{
+			if (sample_type == VL_FULL_SAMPLE)
+				vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 128);
+			else
+				vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 0);
+			
 			if (dct_type == VL_DCT_FRAME_CODED)
-				if (sample_type == VL_FULL_SAMPLE)
-					if (do_idct)
-					{
-						vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 128);
-						vlGrabFrameCodedFullBlock
-						(
-							temp_block,
-							texels + tb * tex_pitch * VL_BLOCK_HEIGHT,
-							tex_pitch
-						);
-					}
-					else
-					vlGrabFrameCodedFullBlock
-					(
-						blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT,
-						texels + tb * tex_pitch * VL_BLOCK_HEIGHT,
-						tex_pitch
-					);
-				else
-					if (do_idct)
-					{
-						vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 0);
-						vlGrabFrameCodedDiffBlock
-						(
-							temp_block,
-							texels + tb * tex_pitch * VL_BLOCK_HEIGHT,
-							tex_pitch
-						);
-					}
-					else
-					vlGrabFrameCodedDiffBlock
-					(
-						blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT,
-						texels + tb * tex_pitch * VL_BLOCK_HEIGHT,
-						tex_pitch
-					);
+				vlGrabFrameCodedBlock
+				(
+					temp_block,
+					texels + tb * tex_pitch * VL_BLOCK_HEIGHT,
+					tex_pitch
+				);
 			else
-				if (sample_type == VL_FULL_SAMPLE)
-					if (do_idct)
-					{
-						vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 128);
-						vlGrabFieldCodedFullBlock
-						(
-							temp_block,
-							texels + (tb % 2) * tex_pitch * VL_BLOCK_HEIGHT + (tb / 2) * tex_pitch,
-							tex_pitch
-						);
-					}
-					else
-					vlGrabFieldCodedFullBlock
-					(
-						blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT,
-						texels + (tb % 2) * tex_pitch * VL_BLOCK_HEIGHT + (tb / 2) * tex_pitch,
-						tex_pitch
-					);
-				else
-					if (do_idct)
-					{
-						vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 0);
-						vlGrabFieldCodedDiffBlock
-						(
-							temp_block,
-							texels + (tb % 2) * tex_pitch * VL_BLOCK_HEIGHT + (tb / 2) * tex_pitch,
-							tex_pitch
-						);
-					}
-					else
-					vlGrabFieldCodedDiffBlock
-					(
-						blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT,
-						texels + (tb % 2) * tex_pitch * VL_BLOCK_HEIGHT + (tb / 2) * tex_pitch,
-						tex_pitch
-					);
+				vlGrabFieldCodedBlock
+				(
+					blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT,
+					texels + (tb % 2) * tex_pitch * VL_BLOCK_HEIGHT + (tb / 2) * tex_pitch,
+					tex_pitch
+				);
+			
 			++sb;
 		}
 		else
@@ -272,43 +193,18 @@ static int vlGrabBlocks
 		tex_pitch = tex_surface->stride / tex_surface->block.size;
 		
 		if ((coded_block_pattern >> (1 - tb)) & 1)
-		{			
+		{
 			if (sample_type == VL_FULL_SAMPLE)
-				if (do_idct)
-				{
-					vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 128);
-					vlGrabFrameCodedFullBlock
-					(
-						temp_block,
-						texels,
-						tex_pitch
-					);
-				}
-				else
-				vlGrabFrameCodedFullBlock
-				(
-					blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT,
-					texels,
-					tex_pitch
-				);
+				vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 128);
 			else
-				if (do_idct)
-				{
-					vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 0);
-					vlGrabFrameCodedDiffBlock
-					(
-						temp_block,
-						texels,
-						tex_pitch
-					);
-				}
-				else
-				vlGrabFrameCodedDiffBlock
-				(
-					blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT,
-					texels,
-					tex_pitch
-				);
+				vlTransformBlock(blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT, temp_block, 0);
+			
+			vlGrabFrameCodedBlock
+			(
+				temp_block,
+				texels,
+				tex_pitch
+			);
 			
 			++sb;
 		}