[Mlir-commits] [mlir] [mlir] Rewrites for I2 to I8 signed and unsigned extension (PR #121298)

[Andrzej Warzyński]

================
@@ -1179,70 +1183,166 @@ Value BitCastRewriter::genericRewriteStep(
   return runningResult;
 }
 
-/// Rewrite the i4 -> i8 signed extension into a sequence of shuffles and
-/// bitwise ops that take advantage of high-level information to avoid leaving
-/// LLVM to scramble with peephole optimizations.
-static Value rewriteI4ToI8SignedExt(PatternRewriter &rewriter, Location loc,
-                                    Value srcValue) {
-  VectorType srcVecType = cast<VectorType>(srcValue.getType());
-  assert(srcVecType.getElementType().isSignlessInteger(4) &&
-         "Expected i4 type");
+/// Bitcasts the aligned `subByteVec` vector to a vector of i8.
+/// Where aligned means it satisfies the alignedConversionPreconditions.
+///
+/// Example:
+/// vector<16x16xi2> -> vector<16x2xi8>
+/// vector<16x16xi4> -> vector<16x4xi8>
+static Value bitcastSubByteVectorToI8(PatternRewriter &rewriter, Location loc,
+                                      Value subByteVec) {
+  auto srcVecType = cast<VectorType>(subByteVec.getType());
+  int64_t srcBitwidth = srcVecType.getElementType().getIntOrFloatBitWidth();
+  assert(8 % srcBitwidth == 0 &&
+         "Unsupported sub-byte type (not a divisor of i8)");
+  int64_t bitwidthFactor = 8 / srcBitwidth;
----------------
banach-space wrote:

To keep the naming consistent within the file, could this be `numSrcElemsPerByte`?

Similar variable elsewhere: https://github.com/llvm/llvm-project/blob/cfee344dda7394631f2177a15e56cfeee1d61fc4/mlir/lib/Dialect/Vector/Transforms/VectorEmulateNarrowType.cpp#L1098

https://github.com/llvm/llvm-project/pull/121298