From cba68090a60c3de8eadbf8eb53e37620a1d66683 Mon Sep 17 00:00:00 2001
From: Veikka Tuominen <git@vexu.eu>
Date: Fri, 11 Mar 2022 13:12:32 +0200
Subject: stage2: implement `@shuffle` at runtime

---
 src/codegen/c.zig    | 16 ++++++++++++++++
 src/codegen/llvm.zig | 38 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 54 insertions(+)

(limited to 'src/codegen')

diff --git a/src/codegen/c.zig b/src/codegen/c.zig
index 4a30bf023b..c179b65046 100644
--- a/src/codegen/c.zig
+++ b/src/codegen/c.zig
@@ -1716,6 +1716,7 @@ fn genBody(f: *Function, body: []const Air.Inst.Index) error{ AnalysisFail, OutO
             .tag_name         => try airTagName(f, inst),
             .error_name       => try airErrorName(f, inst),
             .splat            => try airSplat(f, inst),
+            .shuffle          => try airShuffle(f, inst),
             .aggregate_init   => try airAggregateInit(f, inst),
             .union_init       => try airUnionInit(f, inst),
             .prefetch         => try airPrefetch(f, inst),
@@ -3557,6 +3558,21 @@ fn airSplat(f: *Function, inst: Air.Inst.Index) !CValue {
     return f.fail("TODO: C backend: implement airSplat", .{});
 }
 
+fn airShuffle(f: *Function, inst: Air.Inst.Index) !CValue {
+    if (f.liveness.isUnused(inst)) return CValue.none;
+
+    const inst_ty = f.air.typeOfIndex(inst);
+    const ty_op = f.air.instructions.items(.data)[inst].ty_op;
+    const operand = try f.resolveInst(ty_op.operand);
+    const writer = f.object.writer();
+    const local = try f.allocLocal(inst_ty, .Const);
+    try writer.writeAll(" = ");
+
+    _ = operand;
+    _ = local;
+    return f.fail("TODO: C backend: implement airShuffle", .{});
+}
+
 fn airAggregateInit(f: *Function, inst: Air.Inst.Index) !CValue {
     if (f.liveness.isUnused(inst)) return CValue.none;
 
diff --git a/src/codegen/llvm.zig b/src/codegen/llvm.zig
index d7df161b00..190fcd0ee6 100644
--- a/src/codegen/llvm.zig
+++ b/src/codegen/llvm.zig
@@ -3204,6 +3204,7 @@ pub const FuncGen = struct {
                 .tag_name       => try self.airTagName(inst),
                 .error_name     => try self.airErrorName(inst),
                 .splat          => try self.airSplat(inst),
+                .shuffle        => try self.airShuffle(inst),
                 .aggregate_init => try self.airAggregateInit(inst),
                 .union_init     => try self.airUnionInit(inst),
                 .prefetch       => try self.airPrefetch(inst),
@@ -5850,6 +5851,43 @@ pub const FuncGen = struct {
         return self.builder.buildShuffleVector(op_vector, undef_vector, mask_llvm_ty.constNull(), "");
     }
 
+    fn airShuffle(self: *FuncGen, inst: Air.Inst.Index) !?*const llvm.Value {
+        if (self.liveness.isUnused(inst)) return null;
+
+        const ty_pl = self.air.instructions.items(.data)[inst].ty_pl;
+        const extra = self.air.extraData(Air.Shuffle, ty_pl.payload).data;
+        const a = try self.resolveInst(extra.a);
+        const b = try self.resolveInst(extra.b);
+        const mask = self.air.values[extra.mask];
+        const mask_len = extra.mask_len;
+        const a_len = self.air.typeOf(extra.a).vectorLen();
+
+        // LLVM uses integers larger than the length of the first array to
+        // index into the second array. This was deemed unnecessarily fragile
+        // when changing code, so Zig uses negative numbers to index the
+        // second vector. These start at -1 and go down, and are easiest to use
+        // with the ~ operator. Here we convert between the two formats.
+        const values = try self.gpa.alloc(*const llvm.Value, mask_len);
+        defer self.gpa.free(values);
+
+        const llvm_i32 = self.context.intType(32);
+
+        for (values) |*val, i| {
+            var buf: Value.ElemValueBuffer = undefined;
+            const elem = mask.elemValueBuffer(i, &buf);
+            if (elem.isUndef()) {
+                val.* = llvm_i32.getUndef();
+            } else {
+                const int = elem.toSignedInt();
+                const unsigned = if (int >= 0) @intCast(u32, int) else @intCast(u32, ~int + a_len);
+                val.* = llvm_i32.constInt(unsigned, .False);
+            }
+        }
+
+        const llvm_mask_value = llvm.constVector(values.ptr, mask_len);
+        return self.builder.buildShuffleVector(a, b, llvm_mask_value, "");
+    }
+
     fn airAggregateInit(self: *FuncGen, inst: Air.Inst.Index) !?*const llvm.Value {
         if (self.liveness.isUnused(inst)) return null;
 
-- 
cgit v1.2.3