AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Revert "spirv: Use a simpler and more correct implementaiton of tanh()"

Browse Source 
This reverts commit da1c49171d.

The reduced formula has precision problems on fp16 around 0.  Bring
back the old formula, but make sure to keep the clamping.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4054>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4054>
This commit is contained in:
Kristian H. Kristensen
2020-02-27 11:38:53 -08:00
committed by Marge Bot
parent 986e92f0ea
commit 9f9432d56c
1 changed files with 13 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/compiler/spirv/vtn_glsl450.c
+13 -14
View File
@@ -458,25 +458,24 @@ handle_glsl450_alu(struct vtn_builder *b, enum GLSLstd450 entrypoint,
return;
case GLSLstd450Tanh: {
/* tanh(x) := (0.5 * (e^x - e^(-x))) / (0.5 * (e^x + e^(-x)))
/* tanh(x) := (e^x - e^(-x)) / (e^x + e^(-x))
*
* With a little algebra this reduces to (e^2x - 1) / (e^2x + 1)
* We clamp x to [-10, +10] to avoid precision problems. When x > 10,
* e^x dominates the sum, e^(-x) is lost and tanh(x) is 1.0 for 32 bit
* floating point.
*
* We clamp x to (-inf, +10] to avoid precision problems. When x > 10,
* e^2x is so much larger than 1.0 that 1.0 gets flushed to zero in the
* computation e^2x +/- 1 so it can be ignored.
*
* For 16-bit precision we clamp x to (-inf, +4.2] since the maximum
* representable number is only 65,504 and e^(2*6) exceeds that. Also,
* if x > 4.2, tanh(x) will return 1.0 in fp16.
* For 16-bit precision this we clamp x to [-4.2, +4.2].
*/
const uint32_t bit_size = src[0]->bit_size;
const double clamped_x = bit_size > 16 ? 10.0 : 4.2;
nir_ssa_def *x = nir_fmin(nb, src[0],
nir_imm_floatN_t(nb, clamped_x, bit_size));
nir_ssa_def *exp2x = build_exp(nb, nir_fmul_imm(nb, x, 2.0));
val->ssa->def = nir_fdiv(nb, nir_fadd_imm(nb, exp2x, -1.0),
nir_fadd_imm(nb, exp2x, 1.0));
nir_ssa_def *x = nir_fclamp(nb, src[0],
nir_imm_floatN_t(nb, -clamped_x, bit_size),
nir_imm_floatN_t(nb, clamped_x, bit_size));
val->ssa->def =
nir_fdiv(nb, nir_fsub(nb, build_exp(nb, x),
build_exp(nb, nir_fneg(nb, x))),
nir_fadd(nb, build_exp(nb, x),
build_exp(nb, nir_fneg(nb, x))));
return;
}