Re: [Mesa-dev] GLSLstd450NMin/NMax/NClamp
On 11/17/2020 01:59 PM, Brian Paul wrote: On 11/17/2020 11:45 AM, Ian Romanick wrote: On 11/17/20 9:25 AM, Brian Paul wrote: It appears these SPIR-V extension functions don't behave as they should on Intel (don't know about other Vulkan drivers). They're supposed to be NaN-aware such that if one argument is NaN, the other argument is returned. From our testing, it looks like NMax works as expected, but not NMin or NClamp. Do you have some specific test cases that fail? Yeah, but they come from WHCK and I don't have a small test case. I'll try your patch series and see what happens. -Brian I'm using your patch and I'm also making use of VK_KHR_shader_float_control but neither has helped. Though, the later does help with some issues in Nvidia. -Brian ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] GLSLstd450NMin/NMax/NClamp
On 11/17/2020 11:45 AM, Ian Romanick wrote: On 11/17/20 9:25 AM, Brian Paul wrote: It appears these SPIR-V extension functions don't behave as they should on Intel (don't know about other Vulkan drivers). They're supposed to be NaN-aware such that if one argument is NaN, the other argument is returned. From our testing, it looks like NMax works as expected, but not NMin or NClamp. Do you have some specific test cases that fail? Yeah, but they come from WHCK and I don't have a small test case. I'll try your patch series and see what happens. -Brian Looking at the SPIR-V/nir/intel code it's hard to tell what's going on and whether these semantics are actually being followed. Any comments? -Brian ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.freedesktop.org%2Fmailman%2Flistinfo%2Fmesa-dev&data=04%7C01%7Cbrianp%40vmware.com%7Ca9f0c2e6cfa846f3b20408d88b28f40d%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C1%7C637412355305449546%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=dWV4%2Bsq4B7xBiCxRmyTj7zM4b2Xfl%2BLjtdll6qHhUAg%3D&reserved=0 ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] GLSLstd450NMin/NMax/NClamp
On 11/17/20 9:25 AM, Brian Paul wrote: > > It appears these SPIR-V extension functions don't behave as they should > on Intel (don't know about other Vulkan drivers). > > They're supposed to be NaN-aware such that if one argument is NaN, the > other argument is returned. From our testing, it looks like NMax works > as expected, but not NMin or NClamp. Do you have some specific test cases that fail? > Looking at the SPIR-V/nir/intel code it's hard to tell what's going on > and whether these semantics are actually being followed. > > Any comments? > > -Brian > ___ > mesa-dev mailing list > mesa-dev@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/mesa-dev ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
Re: [Mesa-dev] GLSLstd450NMin/NMax/NClamp
Brian, There are some NaN related fixes by Ian soon to be merged in https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6358. Could you check if the MR helps this and the Select issue? Thanks, Caio From: mesa-dev on behalf of Brian Paul Sent: Tuesday, November 17, 2020 9:25:01 AM To: mesa-dev@lists.freedesktop.org Subject: [Mesa-dev] GLSLstd450NMin/NMax/NClamp It appears these SPIR-V extension functions don't behave as they should on Intel (don't know about other Vulkan drivers). They're supposed to be NaN-aware such that if one argument is NaN, the other argument is returned. From our testing, it looks like NMax works as expected, but not NMin or NClamp. Looking at the SPIR-V/nir/intel code it's hard to tell what's going on and whether these semantics are actually being followed. Any comments? -Brian ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
[Mesa-dev] GLSLstd450NMin/NMax/NClamp
It appears these SPIR-V extension functions don't behave as they should on Intel (don't know about other Vulkan drivers). They're supposed to be NaN-aware such that if one argument is NaN, the other argument is returned. From our testing, it looks like NMax works as expected, but not NMin or NClamp. Looking at the SPIR-V/nir/intel code it's hard to tell what's going on and whether these semantics are actually being followed. Any comments? -Brian ___ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev