llvm/test/Transforms/LowerMatrixIntrinsics/dot-product-volatile-align.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s

; The dot-product / flatten lowering rewrites a matrix.column.major.load feeding
; a 1xK * Kx1 matmul into a single vector load. The volatile flag and alignment
; of the original intrinsic load must be carried over to the rewritten load; the
; non-volatile operand must stay non-volatile.

define <1 x float> @dot_product_flatten_volatile_align(ptr %a, ptr %b) {
; CHECK-LABEL: @dot_product_flatten_volatile_align(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <4 x float>, ptr [[B:%.*]], align 4
; CHECK-NEXT:    [[TMP0:%.*]] = load volatile <4 x float>, ptr [[A:%.*]], align 8
; CHECK-NEXT:    [[TMP1:%.*]] = fmul <4 x float> [[TMP0]], [[COL_LOAD]]
; CHECK-NEXT:    [[TMP2:%.*]] = call reassoc float @llvm.vector.reduce.fadd.v4f32(float 0.000000e+00, <4 x float> [[TMP1]])
; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <1 x float> poison, float [[TMP2]], i64 0
; CHECK-NEXT:    ret <1 x float> [[TMP3]]
;
entry:
  %lhs = call <4 x float> @llvm.matrix.column.major.load.v4f32.i64(ptr align 8 %a, i64 1, i1 true, i32 1, i32 4)
  %rhs = call <4 x float> @llvm.matrix.column.major.load.v4f32.i64(ptr %b, i64 4, i1 false, i32 4, i32 1)
  %res = call reassoc <1 x float> @llvm.matrix.multiply.v1f32.v4f32.v4f32(<4 x float> %lhs, <4 x float> %rhs, i32 1, i32 4, i32 1)
  ret <1 x float> %res
}