From de073a10e6ee90658d2ff760d8fb0467d29656f8 Mon Sep 17 00:00:00 2001
From: Faith Ekstrand <faith.ekstrand@collabora.com>
Date: Mon, 30 Jan 2023 20:53:17 -0600
Subject: [PATCH] nak: Implement 32-bit logic ops

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24998>
---
 src/nouveau/compiler/nak_encode_tu102.rs | 10 ++++
 src/nouveau/compiler/nak_from_nir.rs     | 27 ++++++++++
 src/nouveau/compiler/nak_ir.rs           | 64 +++++++++++++++++++++++-
 3 files changed, 100 insertions(+), 1 deletion(-)

diff --git a/src/nouveau/compiler/nak_encode_tu102.rs b/src/nouveau/compiler/nak_encode_tu102.rs
index f0a6b4d8c47..6f45b6621a6 100644
--- a/src/nouveau/compiler/nak_encode_tu102.rs
+++ b/src/nouveau/compiler/nak_encode_tu102.rs
@@ -204,6 +204,15 @@ fn encode_iadd3(bs: &mut impl BitSetMut, instr: &Instr) {
     bs.set_field(84..87, 7_u32); /* pred */
 }
 
+fn encode_lop3(bs: &mut impl BitSetMut, instr: &Instr, op: &LogicOp) {
+    encode_alu(bs, instr, 0x012);
+    bs.set_field(72..80, op.lut);
+    bs.set_bit(80, false); /* .PAND */
+    bs.set_field(81..84, 7_u32); /* pred */
+    bs.set_field(84..87, 7_u32); /* pred */
+    bs.set_bit(90, true);
+}
+
 fn encode_shl(bs: &mut impl BitSetMut, instr: &Instr) {
     encode_alu(bs, instr, 0x019);
 
@@ -316,6 +325,7 @@ pub fn encode_instr(instr: &Instr) -> [u32; 4] {
         Opcode::S2R(i) => encode_s2r(&mut bs, instr, *i),
         Opcode::MOV => encode_mov(&mut bs, instr),
         Opcode::IADD3 => encode_iadd3(&mut bs, instr),
+        Opcode::LOP3(op) => encode_lop3(&mut bs, instr, &op),
         Opcode::SHL => encode_shl(&mut bs, instr),
         Opcode::ALD(a) => encode_ald(&mut bs, instr, &a),
         Opcode::AST(a) => encode_ast(&mut bs, instr, &a),
diff --git a/src/nouveau/compiler/nak_from_nir.rs b/src/nouveau/compiler/nak_from_nir.rs
index e5cf2d82854..9fb2efa6953 100644
--- a/src/nouveau/compiler/nak_from_nir.rs
+++ b/src/nouveau/compiler/nak_from_nir.rs
@@ -88,6 +88,33 @@ impl<'a> ShaderFromNir<'a> {
             nir_op_iadd => {
                 self.instrs.push(Instr::new_iadd(dst, srcs[0], srcs[1]));
             }
+            nir_op_iand => {
+                self.instrs.push(Instr::new_lop3(
+                    dst,
+                    LogicOp::new_lut(&|x, y, _| x & y),
+                    srcs[0],
+                    srcs[1],
+                    Src::Zero,
+                ));
+            }
+            nir_op_inot => {
+                self.instrs.push(Instr::new_lop3(
+                    dst,
+                    LogicOp::new_lut(&|x, _, _| !x),
+                    srcs[0],
+                    Src::Zero,
+                    Src::Zero,
+                ));
+            }
+            nir_op_ior => {
+                self.instrs.push(Instr::new_lop3(
+                    dst,
+                    LogicOp::new_lut(&|x, y, _| x | y),
+                    srcs[0],
+                    srcs[1],
+                    Src::Zero,
+                ));
+            }
             nir_op_ishl => {
                 self.instrs.push(Instr::new_shl(dst, srcs[0], srcs[1]));
             }
diff --git a/src/nouveau/compiler/nak_ir.rs b/src/nouveau/compiler/nak_ir.rs
index 9ab969cd0af..7753e44a108 100644
--- a/src/nouveau/compiler/nak_ir.rs
+++ b/src/nouveau/compiler/nak_ir.rs
@@ -4,7 +4,7 @@
  */
 
 use std::fmt;
-use std::ops::Range;
+use std::ops::{BitAnd, BitOr, Not, Range};
 use std::slice;
 
 #[derive(Clone, Copy)]
@@ -386,6 +386,61 @@ impl InstrRefs {
     }
 }
 
+pub struct LogicOp {
+    pub lut: u8,
+}
+
+impl LogicOp {
+    #[inline]
+    pub fn new_lut<F: Fn(u8, u8, u8) -> u8>(f: &F) -> LogicOp {
+        LogicOp {
+            lut: f(0xf0, 0xcc, 0xaa),
+        }
+    }
+
+    pub fn eval<
+        T: BitAnd<Output = T> + BitOr<Output = T> + Copy + Not<Output = T>,
+    >(
+        &self,
+        x: T,
+        y: T,
+        z: T,
+    ) -> T {
+        let mut res = x & !x; /* zero */
+        if (self.lut & (1 << 0)) != 0 {
+            res = res | (!x & !y & !z);
+        }
+        if (self.lut & (1 << 1)) != 0 {
+            res = res | (!x & !y & z);
+        }
+        if (self.lut & (1 << 2)) != 0 {
+            res = res | (!x & y & !z);
+        }
+        if (self.lut & (1 << 3)) != 0 {
+            res = res | (!x & y & z);
+        }
+        if (self.lut & (1 << 4)) != 0 {
+            res = res | (x & !y & !z);
+        }
+        if (self.lut & (1 << 5)) != 0 {
+            res = res | (x & !y & z);
+        }
+        if (self.lut & (1 << 6)) != 0 {
+            res = res | (x & y & !z);
+        }
+        if (self.lut & (1 << 7)) != 0 {
+            res = res | (x & y & z);
+        }
+        res
+    }
+}
+
+impl fmt::Display for LogicOp {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "LUT[{:#x}]", self.lut)
+    }
+}
+
 pub struct AttrAccess {
     pub addr: u16,
     pub comps: u8,
@@ -632,6 +687,10 @@ impl Instr {
         Instr::new(Opcode::IADD3, slice::from_ref(&dst), &[Src::Zero, x, y])
     }
 
+    pub fn new_lop3(dst: Dst, op: LogicOp, x: Src, y: Src, z: Src) -> Instr {
+        Instr::new(Opcode::LOP3(op), slice::from_ref(&dst), &[x, y, z])
+    }
+
     pub fn new_shl(dst: Dst, x: Src, shift: Src) -> Instr {
         Instr::new(Opcode::SHL, slice::from_ref(&dst), &[x, shift])
     }
@@ -754,6 +813,7 @@ impl Instr {
             | Opcode::FMNMX
             | Opcode::FMUL
             | Opcode::IADD3
+            | Opcode::LOP3(_)
             | Opcode::SHL => Some(6),
             Opcode::MOV => Some(15),
             Opcode::S2R(_) => None,
@@ -800,6 +860,7 @@ pub enum Opcode {
     FMUL,
 
     IADD3,
+    LOP3(LogicOp),
     SHL,
 
     S2R(u8),
@@ -827,6 +888,7 @@ impl fmt::Display for Opcode {
             Opcode::FMNMX => write!(f, "FMNMX"),
             Opcode::FMUL => write!(f, "FMUL"),
             Opcode::IADD3 => write!(f, "IADD3"),
+            Opcode::LOP3(op) => write!(f, "LOP3.{}", op),
             Opcode::SHL => write!(f, "SHL"),
             Opcode::S2R(i) => write!(f, "S2R({})", i),
             Opcode::MOV => write!(f, "MOV"),