From: Stefan Dösinger Subject: Re: [PATCH v2] wined3d: Avoid using float literals for MAX_FLOAT in GLSL when possible. Message-Id: <77128814-f51c-2dc7-622c-ea9a5792db8c@gmail.com> Date: Sat, 6 Apr 2019 12:06:27 +0300 In-Reply-To: References: <20190405180809.15555-1-gofmanp@gmail.com> Am 06.04.19 um 00:28 schrieb Paul Gofman: > After testing a bit more I figured out the exact conditions which lead > to shader compilation failure, and given those conditions are reliably > avoided by CSMT I don't think anymore that this  patch is much useful. > Please disregard it, sorry for the noise. I suspect the application changed the FPU precision to single precision or something like that. I remember that there is (was?) some bug where the nvidia driver didn't parse a float literal that was not MAX_FLOAT correctly and the resulting loss of precision broke the shader. Other drivers parsed it correctly even with the FPU in single precision mode. -----BEGIN PGP SIGNATURE----- iQIzBAEBCAAdFiEEQxb0tqoFWyeVMl1sPRO8yFRPGiIFAlyobBcACgkQPRO8yFRP GiK/9A/9EAhDD0KKLI/WB1LCSGZeZkgxkX66ESIyP2RbyvvdIfMAs0xoQzKZ8cIv Wd3bEe1yuqSMlLpKVViU6noKq9GWdhNbsxd5+HVVDf6RDItmB9Lm1I2EaZSn8l5M sbPlVLfmhDmE1RFsgMBV27jmeWRy488PTiBAKVOEFPCmA9WEolrGJiOFtlGtysn4 PHwacQkJHthsmvsZSoM+vkA43c9wKbnAF7dUid0erAnHqDrxHUTkKCIBtrbTnb2u GZR+NdZ6Yq/OZnloWIhr/AQVQq7XFVVP4yVWBS+LczMoe+yCklypAJr9Z0Agb2Eg Z/uJ5hv/PFe635ec7289TOsySmZ8w3J4iaJOyLvRtLVvc3TplVLg1B6fXSs/Jr6N RWWfDW4kU8TnYnu3MKRkeER4HmPhDPBohvJz1L/8hDJt2vxWVDr2qHX2aHODtc2k fYjxLExtDvbQIbYa4HVYJQPEWjvt8dsRKcPi2nRwXFQMN6+33dxdBwhFVdCpRC7f xkFBUTRjmMIYBZsC1k4cKKBkyVOHOTkCg4HKI8cbKcxkDRPBxrI6Y5hOHEzM8tpk HXKKipFEi2DHw+SbU9I1vSZ5e7p7qVaS8nHdUMtB3rfL+UHGZt63K0/bSx2KcZ1X zguYMtWGj27UneAxD29r/sjFByljQpQKluutc41gX86GfVap7xw= =Jrc5 -----END PGP SIGNATURE-----