Module: Mesa Branch: main Commit: 7b6103ccc85908fd4ffffd542d33afb255b5eecf URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=7b6103ccc85908fd4ffffd542d33afb255b5eecf
Author: Faith Ekstrand <faith.ekstr...@collabora.com> Date: Thu Nov 23 21:42:57 2023 -0600 nak: Add a builder helper for OpPrmt The builder also has some extra smarts in it for avoiding the PRMT when the operation happens to exactly select one of the two sources. Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26348> --- src/nouveau/compiler/nak_builder.rs | 26 ++++++++++++++++++++++++++ src/nouveau/compiler/nak_from_nir.rs | 7 +------ 2 files changed, 27 insertions(+), 6 deletions(-) diff --git a/src/nouveau/compiler/nak_builder.rs b/src/nouveau/compiler/nak_builder.rs index 31cb220b6c8..f8090243acd 100644 --- a/src/nouveau/compiler/nak_builder.rs +++ b/src/nouveau/compiler/nak_builder.rs @@ -49,6 +49,26 @@ pub trait Builder { } } + fn prmt_to(&mut self, dst: Dst, x: Src, y: Src, sel: [u8;4]) { + if sel == [0, 1, 2, 3] { + self.copy_to(dst, x); + } else if sel == [4, 5, 6, 7] { + self.copy_to(dst, y); + } else { + let mut sel_u32 = 0; + for i in 0..4 { + assert!(sel[i] < 16); + sel_u32 |= u32::from(sel[i]) << (i * 4); + } + + self.push_op(OpPrmt { + dst: dst, + srcs: [x, y], + selection: sel_u32.into(), + }); + } + } + fn copy_to(&mut self, dst: Dst, src: Src) { self.push_op(OpCopy { dst: dst, src: src }); } @@ -206,6 +226,12 @@ pub trait SSABuilder: Builder { dst } + fn prmt(&mut self, x: Src, y: Src, sel: [u8;4]) -> SSARef { + let dst = self.alloc_ssa(RegFile::GPR, 1); + self.prmt_to(dst.into(), x, y, sel); + dst + } + fn sel(&mut self, cond: Src, x: Src, y: Src) -> SSARef { assert!(cond.src_ref.is_predicate()); assert!(x.is_predicate() == y.is_predicate()); diff --git a/src/nouveau/compiler/nak_from_nir.rs b/src/nouveau/compiler/nak_from_nir.rs index 7e7653dad56..40e87c63dcf 100644 --- a/src/nouveau/compiler/nak_from_nir.rs +++ b/src/nouveau/compiler/nak_from_nir.rs @@ -811,12 +811,7 @@ impl<'a> ShaderFromNir<'a> { high: false, }); - b.push_op(OpPrmt { - dst: dst.into(), - srcs: [low.into(), high.into()], - selection: 0x5410.into(), - }); - dst + b.prmt(low.into(), high.into(), [0, 1, 4, 5]) } nir_op_u2f32 => { assert!(alu.def.bit_size() == 32);
