From d94e16eecdef5f6112b7cf0f1e42233a8ba92342 Mon Sep 17 00:00:00 2001
From: Roland Conybeare <rconybeare@gmail.com>
Date: Mon, 24 Jun 2024 15:09:27 -0400
Subject: [PATCH] xo-expression: + several arithmetic intrinsics

---
 include/xo/expression/llvmintrinsic.hpp | 37 +++++++++++++++++++++++--
 1 file changed, 34 insertions(+), 3 deletions(-)

diff --git a/include/xo/expression/llvmintrinsic.hpp b/include/xo/expression/llvmintrinsic.hpp
index 37f231fe..052baa9e 100644
--- a/include/xo/expression/llvmintrinsic.hpp
+++ b/include/xo/expression/llvmintrinsic.hpp
@@ -19,9 +19,18 @@ namespace xo {
          *  can use the associated llvm instrinsic instead of generating a function call
          *  @c Primitive::value
          *
-         *  @note llvm will still need to use @c Primitive::value,  for example
-         *  when handling an @c xo::ast::Apply instance where the function position
-         *  is a computed function.
+         *  @note llvm will still sometimes need to use
+         *  @c Primitive::value (and generate a function call sequence),
+         *  for example when handling an @c xo::ast::Apply instance
+         *  where the function position is a computed function.
+         *  @endnote
+         *
+         *  @note
+         *  LLVM requires separate intrinsics for {ints, floats}.
+         *  It does not need separate intrinsics for different sizes.
+         *  For example IRBuilder::CreateAdd works for
+         *  {8-bit, 16-bit, 32-bit, 64-bit, 128-bit} x {signed, unsigned} integers
+         *  Integer division is an exception;  need to choose between i_sdiv and i_udiv
          *  @endnote
          *
          *  @note
@@ -30,21 +39,38 @@ namespace xo {
          *  @endnote
          **/
         enum class llvmintrinsic {
+            // see /nix/store/x5yz...llvm-18.1.5-dev/include/llvm/IR/IRBuilder.h
+
             /** sentinel value **/
             invalid = -1,
 
+            /** -> IRBuilder::CreateNeg (negate 1 integer) **/
+            i_neg,
+
             /** -> IRBuilder::CreateAdd (add 2 integers, overflow silently) **/
             i_add,
 
+            /** -> IRBuilder::CreateSub (subtract 2 integers, overflow silently) **/
+            i_sub,
+
             /** -> IRBuilder::CreateMul (multiply 2 integers, overflow silently) **/
             i_mul,
 
+            /** -> IRBuilder::CreateSdiv (divide 2 signed integers) **/
+            i_sdiv,
+
+            /** -> IRBuilder::CreateUdiv (divide 2 unsigned integers) **/
+            i_udiv,
+
             /** -> IRBuilder::CreateFadd (add 2 floating-point numbers) **/
             fp_add,
 
             /** -> IRBuilder::CreateFmul (multiply 2 floating-point numbers) **/
             fp_mul,
 
+            /** -> IRBuilder::CreateFdiv (divide 2 floating-point numbers) **/
+            fp_div,
+
             /**
              *  want to do whatever llvm IR @c llvm.sqrt.f64 and friends do.
              *  Not sure if that's an always-available function of something else
@@ -72,10 +98,15 @@ namespace xo {
         {
             switch(x) {
             case llvmintrinsic::invalid: return "?llvminstrinsic";
+            case llvmintrinsic::i_neg: return "i_neg";
             case llvmintrinsic::i_add: return "i_add";
+            case llvmintrinsic::i_sub: return "i_sub";
             case llvmintrinsic::i_mul: return "i_mul";
+            case llvmintrinsic::i_sdiv: return "i_sdiv";
+            case llvmintrinsic::i_udiv: return "i_udiv";
             case llvmintrinsic::fp_add: return "fp_add";
             case llvmintrinsic::fp_mul: return "fp_mul";
+            case llvmintrinsic::fp_div: return "fp_div";
             case llvmintrinsic::fp_sqrt: return "fp_sqrt";
             case llvmintrinsic::fp_pow: return "fp_pow";
             case llvmintrinsic::fp_sin: return "fp_sin";