ac/llvm: don't assert on 32bit ffma before gfx9

Yes it's slow, but it's better than nothing.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41649>
This commit is contained in:
Georg Lehmann 2026-05-18 14:39:53 +02:00 committed by Marge Bot
parent a92d0356eb
commit 015b1d9e6d

View file

@ -760,8 +760,6 @@ static bool visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr)
}
break;
case nir_op_ffma:
/* FMA is slow on gfx6-8, so it shouldn't be used. */
assert(instr->def.bit_size != 32 || ctx->ac.gfx_level >= GFX9);
result = emit_fp_intrinsic(&ctx->ac, "llvm.fma", def_type, src[0], src[1], src[2]);
break;
case nir_op_ffmaz: