diff options
author | Dave Airlie <[email protected]> | 2017-03-17 12:11:30 +1000 |
---|---|---|
committer | Dave Airlie <[email protected]> | 2017-05-03 12:55:34 +1000 |
commit | 83e58b036e1c34f26c99d04615df2b530f3045d9 (patch) | |
tree | 1c89241329944a5b75c2f4b3a1ea925f649309ad /src/amd/common/sid.h | |
parent | eeff7e11544f333d211c8f8ad3679db814050cfa (diff) |
radv: flush f32->f16 conversion denormals to zero. (v2)
SPIR-V defines the f32->f16 operation as flushing denormals to 0,
this compares the class using amd class opcode.
Thanks to Matt Arsenault for figuring it out.
This fix is VI+ only, add a TODO for SI/CIK.
This fixes:
dEQP-VK.spirv_assembly.instruction.compute.opquantize.flush_to_zero
Acked-by: Bas Nieuwenhuizen <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
Diffstat (limited to 'src/amd/common/sid.h')
-rw-r--r-- | src/amd/common/sid.h | 13 |
1 files changed, 13 insertions, 0 deletions
diff --git a/src/amd/common/sid.h b/src/amd/common/sid.h index 08cdfd77f07..d329ad9493f 100644 --- a/src/amd/common/sid.h +++ b/src/amd/common/sid.h @@ -9095,5 +9095,18 @@ #define CIK_SDMA_PACKET_SRBM_WRITE 0xe #define CIK_SDMA_COPY_MAX_SIZE 0x3fffe0 +enum amd_cmp_class_flags { + S_NAN = 1 << 0, // Signaling NaN + Q_NAN = 1 << 1, // Quiet NaN + N_INFINITY = 1 << 2, // Negative infinity + N_NORMAL = 1 << 3, // Negative normal + N_SUBNORMAL = 1 << 4, // Negative subnormal + N_ZERO = 1 << 5, // Negative zero + P_ZERO = 1 << 6, // Positive zero + P_SUBNORMAL = 1 << 7, // Positive subnormal + P_NORMAL = 1 << 8, // Positive normal + P_INFINITY = 1 << 9 // Positive infinity +}; + #endif /* _SID_H */ |