# RUN: %stinkytofu-opt --arch gfx1250 %s --StinkyBuildImplicitDependencyPass --ds-read-order=AscendingCache --StinkyDAGSchedulerPass --print-output
#
# AscendingCache zigzag: AABB → ABBA across WMMA groups.
# Original: A0A0 A1A1 B0B0 B1B1. A0+B0 feed WMMA0, A1+B1 feed WMMA1.
# Zigzag alternates direction: group 0 ascending (A0 B0), group 1 descending (B1 A1).
# Result: A0A0 B0B0 B1B1 A1A1 — B0 and B1 adjacent for cache reuse.
#
# CHECK: v[8:11]{{.*}}ds_load_b128
# CHECK: v[12:15]{{.*}}ds_load_b128
# CHECK: v[60:63]{{.*}}ds_load_b128
# CHECK: v[64:67]{{.*}}ds_load_b128
# CHECK: v[68:71]{{.*}}ds_load_b128
# CHECK: v[72:75]{{.*}}ds_load_b128
# CHECK: v[16:19]{{.*}}ds_load_b128
# CHECK: v[20:23]{{.*}}ds_load_b128

st.func @test() {
^entry:
  v[8:11] = "st.ds_load_b128"(v100) { issueCycles = 1, latencyCycles = 56, mod.ds = { na = 1, offset = 0, gds = false }, mod.memtoken = { tokens = [0] } }
  v[12:15] = "st.ds_load_b128"(v100) { issueCycles = 1, latencyCycles = 56, mod.ds = { na = 1, offset = 32, gds = false }, mod.memtoken = { tokens = [0] } }
  v[16:19] = "st.ds_load_b128"(v100) { issueCycles = 1, latencyCycles = 56, mod.ds = { na = 1, offset = 8192, gds = false }, mod.memtoken = { tokens = [0] } }
  v[20:23] = "st.ds_load_b128"(v100) { issueCycles = 1, latencyCycles = 56, mod.ds = { na = 1, offset = 8224, gds = false }, mod.memtoken = { tokens = [0] } }
  v[60:63] = "st.ds_load_b128"(v101) { issueCycles = 1, latencyCycles = 56, mod.ds = { na = 1, offset = 0, gds = false }, mod.memtoken = { tokens = [0] } }
  v[64:67] = "st.ds_load_b128"(v101) { issueCycles = 1, latencyCycles = 56, mod.ds = { na = 1, offset = 32, gds = false }, mod.memtoken = { tokens = [0] } }
  v[68:71] = "st.ds_load_b128"(v101) { issueCycles = 1, latencyCycles = 56, mod.ds = { na = 1, offset = 256, gds = false }, mod.memtoken = { tokens = [0] } }
  v[72:75] = "st.ds_load_b128"(v101) { issueCycles = 1, latencyCycles = 56, mod.ds = { na = 1, offset = 288, gds = false }, mod.memtoken = { tokens = [0] } }
  v[0:7] = "st.v_wmma_f32_16x16x16_bf16"(v[8:15], v[60:67], v[0:7]) { mod.mfma = { reuseA = false, reuseB = false } }
  v[0:7] = "st.v_wmma_f32_16x16x16_bf16"(v[16:23], v[68:75], v[0:7]) { mod.mfma = { reuseA = false, reuseB = false } }
}