OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nir: make fdph lowering match fdot

Browse Source 
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20812>
This commit is contained in:
Rhys Perry
2023-01-20 11:38:43 +00:00
committed by Marge Bot
parent 3668da7c83
commit 50f7e21481
1 changed files with 22 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
src/compiler/nir/nir_lower_alu_width.c
View File

@@ -294,15 +294,29 @@ lower_alu_instr_width(nir_builder *b, nir_instr *instr, void *_data)
nir_ssa_def *src0_vec = nir_ssa_for_alu_src(b, alu, 0); nir_ssa_def *src0_vec = nir_ssa_for_alu_src(b, alu, 0);
nir_ssa_def *src1_vec = nir_ssa_for_alu_src(b, alu, 1); nir_ssa_def *src1_vec = nir_ssa_for_alu_src(b, alu, 1);
nir_ssa_def *sum[4]; /* Only use reverse order for imprecise fdph, see explanation in lower_fdot. */
for (unsigned i = 0; i < 3; i++) { bool reverse_order = !b->exact;
sum[i] = nir_fmul(b, nir_channel(b, src0_vec, i), if (will_lower_ffma(b->shader, alu->dest.dest.ssa.bit_size)) {
nir_channel(b, src1_vec, i)); nir_ssa_def *sum[4];
} for (unsigned i = 0; i < 3; i++) {
sum[3] = nir_channel(b, src1_vec, 3); int dest = reverse_order ? 3 - i : i;
sum[dest] = nir_fmul(b, nir_channel(b, src0_vec, i),
nir_channel(b, src1_vec, i));
}
sum[reverse_order ? 0 : 3] = nir_channel(b, src1_vec, 3);
return nir_fadd(b, nir_fadd(b, sum[0], sum[1]), return nir_fadd(b, nir_fadd(b, nir_fadd(b, sum[0], sum[1]), sum[2]), sum[3]);
nir_fadd(b, sum[2], sum[3])); } else if (reverse_order) {
nir_ssa_def *sum = nir_channel(b, src1_vec, 3);
for (int i = 2; i >= 0; i--)
sum = nir_ffma(b, nir_channel(b, src0_vec, i), nir_channel(b, src1_vec, i), sum);
return sum;
} else {
nir_ssa_def *sum = nir_fmul(b, nir_channel(b, src0_vec, 0), nir_channel(b, src1_vec, 0));
sum = nir_ffma(b, nir_channel(b, src0_vec, 1), nir_channel(b, src1_vec, 1), sum);
sum = nir_ffma(b, nir_channel(b, src0_vec, 2), nir_channel(b, src1_vec, 2), sum);
return nir_fadd(b, sum, nir_channel(b, src1_vec, 3));
}
} }
case nir_op_pack_64_2x32: { case nir_op_pack_64_2x32: {