[llvm] [AMDGPU][NFC] Pre-commit memcpy test with complex constant length (PR #182170)

[via llvm-commits]

================
@@ -0,0 +1,91 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a < %s | FileCheck %s
+
+; Test mempcy lowering where length is given by a complex but constant expression.
+; Loop guard should not be necessary since length is positive.
+
+ at src_array = global [128 x i8] zeroinitializer, align 1
+ at dst_array = global [128 x i8] zeroinitializer, align 1
+
+define amdgpu_kernel void @_start() {
+; CHECK-LABEL: _start:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_add_u32 flat_scratch_lo, s12, s17
+; CHECK-NEXT:    s_addc_u32 flat_scratch_hi, s13, 0
+; CHECK-NEXT:    s_mov_b64 s[0:1], 0
+; CHECK-NEXT:  .LBB0_1: ; %dynamic-memcpy-expansion-main-body
+; CHECK-NEXT:    ; =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:    s_getpc_b64 s[2:3]
+; CHECK-NEXT:    s_add_u32 s2, s2, src_array at gotpcrel32@lo+4
+; CHECK-NEXT:    s_addc_u32 s3, s3, src_array at gotpcrel32@hi+12
+; CHECK-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
+; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
+; CHECK-NEXT:    s_add_u32 s2, s2, s0
+; CHECK-NEXT:    s_addc_u32 s3, s3, s1
+; CHECK-NEXT:    v_pk_mov_b32 v[0:1], s[2:3], s[2:3] op_sel:[0,1]
+; CHECK-NEXT:    flat_load_dwordx4 v[2:5], v[0:1]
+; CHECK-NEXT:    s_getpc_b64 s[2:3]
+; CHECK-NEXT:    s_add_u32 s2, s2, dst_array at gotpcrel32@lo+4
+; CHECK-NEXT:    s_addc_u32 s3, s3, dst_array at gotpcrel32@hi+12
+; CHECK-NEXT:    s_load_dwordx2 s[2:3], s[2:3], 0x0
+; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
+; CHECK-NEXT:    s_add_u32 s2, s2, s0
+; CHECK-NEXT:    s_addc_u32 s3, s3, s1
+; CHECK-NEXT:    s_add_u32 s0, s0, 16
+; CHECK-NEXT:    s_addc_u32 s1, s1, 0
+; CHECK-NEXT:    v_pk_mov_b32 v[0:1], s[2:3], s[2:3] op_sel:[0,1]
+; CHECK-NEXT:    v_cmp_lt_u64_e64 s[2:3], s[0:1], 16
+; CHECK-NEXT:    s_and_b64 vcc, exec, s[2:3]
+; CHECK-NEXT:    s_waitcnt vmcnt(0)
+; CHECK-NEXT:    flat_store_dwordx4 v[0:1], v[2:5]
+; CHECK-NEXT:    s_cbranch_vccnz .LBB0_1
+; CHECK-NEXT:  ; %bb.2: ; %dynamic-memcpy-expansion-residual-cond
+; FIXME: Compare should be evaluated at compile time
+; CHECK-NEXT:    s_cmp_eq_u64 13, 0
+; CHECK-NEXT:    s_cbranch_scc1 .LBB0_5
+; CHECK-NEXT:  ; %bb.3: ; %dynamic-memcpy-expansion-residual-body.preheader
+; CHECK-NEXT:    s_sub_u32 s4, 29, 13
+; CHECK-NEXT:    s_subb_u32 s5, 0, 0
+; CHECK-NEXT:    s_getpc_b64 s[0:1]
+; CHECK-NEXT:    s_add_u32 s0, s0, src_array at gotpcrel32@lo+4
+; CHECK-NEXT:    s_addc_u32 s1, s1, src_array at gotpcrel32@hi+12
+; CHECK-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
+; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
+; CHECK-NEXT:    s_add_u32 s2, s0, s4
+; CHECK-NEXT:    s_addc_u32 s3, s1, s5
+; CHECK-NEXT:    s_getpc_b64 s[0:1]
+; CHECK-NEXT:    s_add_u32 s0, s0, dst_array at gotpcrel32@lo+4
+; CHECK-NEXT:    s_addc_u32 s1, s1, dst_array at gotpcrel32@hi+12
+; CHECK-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
+; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
+; CHECK-NEXT:    s_add_u32 s4, s0, s4
+; CHECK-NEXT:    s_addc_u32 s5, s1, s5
+; CHECK-NEXT:    s_mov_b64 s[0:1], 0
+; CHECK-NEXT:  .LBB0_4: ; %dynamic-memcpy-expansion-residual-body
+; CHECK-NEXT:    ; =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:    s_add_u32 s6, s2, s0
+; CHECK-NEXT:    s_addc_u32 s7, s3, s1
+; CHECK-NEXT:    v_pk_mov_b32 v[0:1], s[6:7], s[6:7] op_sel:[0,1]
+; CHECK-NEXT:    flat_load_ubyte v2, v[0:1]
+; CHECK-NEXT:    s_add_u32 s6, s4, s0
+; CHECK-NEXT:    s_addc_u32 s7, s5, s1
+; CHECK-NEXT:    s_add_u32 s0, s0, 1
+; CHECK-NEXT:    s_addc_u32 s1, s1, 0
+; CHECK-NEXT:    v_pk_mov_b32 v[0:1], s[6:7], s[6:7] op_sel:[0,1]
+; CHECK-NEXT:    v_cmp_lt_u64_e64 s[6:7], s[0:1], 13
+; CHECK-NEXT:    s_and_b64 vcc, exec, s[6:7]
+; CHECK-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; CHECK-NEXT:    flat_store_byte v[0:1], v2
+; CHECK-NEXT:    s_cbranch_vccnz .LBB0_4
+; CHECK-NEXT:  .LBB0_5: ; %dynamic-memcpy-post-expansion
+; CHECK-NEXT:    s_endpgm
+  %src_ptr = getelementptr inbounds [4 x i8], [4 x i8]* @src_array, i64 0, i64 0
----------------
LU-JOHN wrote:

Done.

https://github.com/llvm/llvm-project/pull/182170