Name EXT_render_snorm Name Strings GL_EXT_render_snorm Contributors Daniel Koch, NVIDIA Jan-Harald Fredriksen, ARM Mathias Heyer, NVIDIA Contact Mathias Heyer, NVIDIA (mheyer [at] nvidia.com) Status Complete Version Last Modified Date: 2014-10-24 Revision: 4 Number OpenGL ES Extension #206 Dependencies OpenGL ES 3.1 is required. This extension is written against the OpenGL ES 3.1 (June 4, 2014) specification. This extension interacts with EXT_texture_norm16. Overview OpenGL ES 3.1 supports a variety of signed normalized texture and renderbuffer formats which are not color-renderable. This extension enables signed normalized texture and renderbuffer formats to be color-renderable. New Procedures and Functions None New Tokens Accepted by the parameter of ReadPixels BYTE 0x1400 // core OpenGL ES 3.1 SHORT 0x1402 // core OpenGL ES 3.1 Accepted by the parameter of RenderbufferStorage and RenderbufferStorageMultisample: R8_SNORM 0x8F94 // core OpenGL ES 3.1 RG8_SNORM 0x8F95 // core OpenGL ES 3.1 RGBA8_SNORM 0x8F97 // core OpenGL ES 3.1 R16_SNORM_EXT 0x8F98 // EXT_texture_norm16 RG16_SNORM_EXT 0x8F99 // EXT_texture_norm16 RGBA16_SNORM_EXT 0x8F9B // EXT_texture_norm16 Additions to Chapter 8 of the OpenGL ES 3.1 Specification (Textures and Samplers) Changes to Section 8.6 "Alternate Texture ImageSpecification Commands": Modify INVALID_OPERATION errors section on CopyTexImage2D, first bullet point from: "if signed integer, unsigned integer, or fixed-point RGBA data is required and the format of the current color buffer does not match the required format." to "if FRAMEBUFFER_ATTACHMENT_COMPONENT_TYPE for the framebuffer attachment corresponding to the read buffer does not match the component type of the requested ." Change Table 8.13 "Correspondence of sized internal formats to base internal formats,...": Sized Base Bits/component CR TF Req. Req. Internal Internal S are shared bits rend. tex. Format Format R G B A S ------------ ----------- ---- ---- ---- --- ---- --- --- ---- --- R8_SNORM RED s8 X X X X RG8_SNORM RG s8 s8 X X X X RGBA8_SNORM RGBA s8 s8 s8 s8 X X X X R16_SNORM_EXT RED s16 X X X X RG16_SNORM_EXT RG s16 s16 X X X X RGBA16_SNORM_EXT RGBA s16 s16 s16 s16 X X X X Change Table 8.15: "ReadPixels format and type used during CopyTex*." Replace the first row with the following: Read Buffer Format format type ------------------------------------ ------ ------------- 8bit Unsigned Normalized Fixed-point RGBA UNSIGNED_BYTE 8bit Signed Normalized Fixed-point RGBA BYTE 16bit Signed Normalized Fixed-point RGBA SHORT Additions to Chapter 15 of the OpenGL ES 3.1 Specification (Writing Fragments and Samples to the Framebuffer) Changes to Section 15.1.7 "Blending": Replace "The components of the source and destination values and blend factors are clamped to [0, 1] prior to evaluating the blend equation." with "If the color buffer is fixed-point, the components of the source and destination values and blend factors are each clamped to [0, 1] or [-1, 1] respectively for an unsigned normalized or signed normalized color buffer prior to evaluating the blend equation." Changes to Section 15.1.7.1 "Blend Equation": Replace "Unsigned normalized fixed-point destination (framebuffer) components are represented as described in section 2.3.4." with "Normalized fixed-point destination (framebuffer) components are represented as described in section 2.3.4." Replace "Prior to blending, unsigned normalized fixed-point color components undergo an implied conversion to floating-point using equation 2.1." with "Prior to blending, unsigned and signed normalized fixed-point color components undergo an implied conversion to floating-point using equation 2.1 and 2.2, respectively." Changes to Section 15.1.7.3 "Blend Color" Replace "If destination framebuffer components use an unsigned normalized fixed-point representation, the constant color components are clamped to the range [0, 1] when computing blend factors." with "If destination framebuffer components use an unsigned or signed normalized fixed-point representation, the constant color components are clamped to the range [0, 1] or [-1, 1], respectively, when computing blend factors." Changes to Section 15.2.3 "Clearing the Buffers": Replace "Unsigned normalized fixed-point RGBA color buffers are cleared to color values derived by clamping each component of the clear color to the range [0, 1], then converting the (possibly sRGB converted and/or dithered) color to fixed-point using equations 2.3 or 2.4, respectively." with "Unsigned normalized fixed-point or signed normalized fixed-point RGBA color buffers are cleared to color values derived by clamping each component of the clear color to the range [0, 1] or [-1, 1] respectively, then converting the (possibly sRGB converted and/or dithered) color to fixed-point using equations 2.3 or 2.4, respectively." Add to the second paragraph of Section 16.1.2 "ReadPixels": "For 8bit signed normalized fixed-point rendering surfaces, the combination format RGBA and type BYTE is accepted. For a 16bit signed normalized fixed point buffer, the combination RGBA and SHORT is accepted." Add to Section 16.1.4 "Conversion of RGBA values": For signed normalized fixed-point color buffer, each element is converted to floating-point using equation 2.2. Errors No new errors. Interactions with EXT_texture_norm16 If EXT_texture_norm16 is not supported, remove references to R16_SNORM_EXT, RG16_SNORM_EXT, RGB16_SNORM_EXT, RGBA16_SNORM_EXT. Remove language and additions referring to 16bit (signed / unsigned) normalized fixed point buffers. Issues 1. How does this extension differ from the functionality offered by GL4.4? On GL, the affected formats are generally exposed as 'color renderable' but not as 'required renderbuffer format', which might be interpreted as 'optionally renderable'. EXT_render_snorm tries to avoid this ambiguity by introducing formats as 'neither color-renderable nor req. renderbuffer format' or as 'color-renderable plus required renderbuffer format', but not a mixture of both. DX 10.1 level hardware is expected to supported these formats as being renderable. 3-component snorm formats were left out as they are not expected to map well to native hardware formats. 2. Are format conversions between UNORM and SNORM formats allowed for CopyTexImage2D? The OpenGL ES specs describe the CopyTexImage2D operation in terms of ReadPixels followed by TexImage2D. The ReadPixels command will use the appropriate format/type combination from Table 8.15. For an internal format of RGBA8_SNORM, this will be GL_RGBA and GL_BYTE. (Special rules apply to framebuffer attachments that have been created using unsized internalformats. These rules don't apply to attachments of _SNORM formats as they could only have been created using sized internalformats.) The following 'virtual' TexImage2D call will use the same format/type combination together with the provided internalformat. The resulting combination of internalformat/format/type must be one of those listed in Table 8.12. For signed fixed point formats it is not possible to create them from unsigned fixed point data and vice versa. Effectively this means, CopyTexImage2D cannot perform any conversions that glTexImage could not do (except dropping color components); in particular it cannot convert between mismatching component types. Source Effective | Requested | CopyTexImage Supported Internalformat | Internalformat | -----------------+------------------------+------------------------------------ sized SNORM | sized SNORM | yes (provided component sizes match | | and the same or more components | | are present in the source buffer) sized SNORM | sized or unsized UNORM | no (conversions not supported) Revision History: Revision: 1 2014-07-18 (mheyer) initial revision Revision: 2 2014-08-08 (mheyer) rename extension to EXT_render_snorm Revision: 3 2014-08-20 (mheyer) clarify CopyTexImage2D behavior Revision: 4 2014-10-24 (dkoch) mark complete, publishing cleanup