[llvm] [NFC][AMDGPU] Move the rem tests in `div_i128.ll` into `rem_i128.ll` (PR #83307)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Feb 28 09:56:18 PST 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Shilei Tian (shiltian)
<details>
<summary>Changes</summary>
---
Patch is 323.91 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/83307.diff
2 Files Affected:
- (modified) llvm/test/CodeGen/AMDGPU/div_i128.ll (-3003)
- (added) llvm/test/CodeGen/AMDGPU/rem_i128.ll (+3014)
``````````diff
diff --git a/llvm/test/CodeGen/AMDGPU/div_i128.ll b/llvm/test/CodeGen/AMDGPU/div_i128.ll
index 5296ad3ab51d31..2f3d5d9d140c2c 100644
--- a/llvm/test/CodeGen/AMDGPU/div_i128.ll
+++ b/llvm/test/CodeGen/AMDGPU/div_i128.ll
@@ -2310,2860 +2310,6 @@ define i128 @v_udiv_i128_vv(i128 %lhs, i128 %rhs) {
ret i128 %div
}
-define i128 @v_srem_i128_vv(i128 %lhs, i128 %rhs) {
-; GFX9-LABEL: v_srem_i128_vv:
-; GFX9: ; %bb.0: ; %_udiv-special-cases
-; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_ashrrev_i32_e32 v20, 31, v3
-; GFX9-NEXT: v_xor_b32_e32 v0, v0, v20
-; GFX9-NEXT: v_xor_b32_e32 v10, v2, v20
-; GFX9-NEXT: v_xor_b32_e32 v1, v1, v20
-; GFX9-NEXT: v_sub_co_u32_e32 v2, vcc, v0, v20
-; GFX9-NEXT: v_xor_b32_e32 v9, v3, v20
-; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v1, v20, vcc
-; GFX9-NEXT: v_ashrrev_i32_e32 v8, 31, v7
-; GFX9-NEXT: v_subb_co_u32_e32 v0, vcc, v10, v20, vcc
-; GFX9-NEXT: v_xor_b32_e32 v4, v4, v8
-; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v9, v20, vcc
-; GFX9-NEXT: v_xor_b32_e32 v5, v5, v8
-; GFX9-NEXT: v_sub_co_u32_e32 v23, vcc, v4, v8
-; GFX9-NEXT: v_xor_b32_e32 v6, v6, v8
-; GFX9-NEXT: v_subb_co_u32_e32 v21, vcc, v5, v8, vcc
-; GFX9-NEXT: v_xor_b32_e32 v7, v7, v8
-; GFX9-NEXT: v_subb_co_u32_e32 v4, vcc, v6, v8, vcc
-; GFX9-NEXT: v_subb_co_u32_e32 v5, vcc, v7, v8, vcc
-; GFX9-NEXT: v_or_b32_e32 v7, v21, v5
-; GFX9-NEXT: v_or_b32_e32 v6, v23, v4
-; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7]
-; GFX9-NEXT: v_or_b32_e32 v7, v3, v1
-; GFX9-NEXT: v_or_b32_e32 v6, v2, v0
-; GFX9-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[6:7]
-; GFX9-NEXT: v_ffbh_u32_e32 v6, v4
-; GFX9-NEXT: v_add_u32_e32 v6, 32, v6
-; GFX9-NEXT: v_ffbh_u32_e32 v7, v5
-; GFX9-NEXT: v_min_u32_e32 v6, v6, v7
-; GFX9-NEXT: v_ffbh_u32_e32 v7, v23
-; GFX9-NEXT: v_add_u32_e32 v7, 32, v7
-; GFX9-NEXT: v_ffbh_u32_e32 v8, v21
-; GFX9-NEXT: v_min_u32_e32 v7, v7, v8
-; GFX9-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
-; GFX9-NEXT: v_add_co_u32_e32 v7, vcc, 64, v7
-; GFX9-NEXT: v_addc_co_u32_e64 v8, s[6:7], 0, 0, vcc
-; GFX9-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[4:5]
-; GFX9-NEXT: v_ffbh_u32_e32 v9, v1
-; GFX9-NEXT: v_cndmask_b32_e32 v6, v7, v6, vcc
-; GFX9-NEXT: v_ffbh_u32_e32 v7, v0
-; GFX9-NEXT: v_add_u32_e32 v7, 32, v7
-; GFX9-NEXT: v_min_u32_e32 v7, v7, v9
-; GFX9-NEXT: v_ffbh_u32_e32 v9, v2
-; GFX9-NEXT: v_add_u32_e32 v9, 32, v9
-; GFX9-NEXT: v_ffbh_u32_e32 v10, v3
-; GFX9-NEXT: v_min_u32_e32 v9, v9, v10
-; GFX9-NEXT: v_cndmask_b32_e64 v8, v8, 0, vcc
-; GFX9-NEXT: v_add_co_u32_e32 v9, vcc, 64, v9
-; GFX9-NEXT: v_addc_co_u32_e64 v10, s[6:7], 0, 0, vcc
-; GFX9-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
-; GFX9-NEXT: s_mov_b64 s[6:7], 0x7f
-; GFX9-NEXT: v_cndmask_b32_e32 v7, v9, v7, vcc
-; GFX9-NEXT: v_cndmask_b32_e64 v10, v10, 0, vcc
-; GFX9-NEXT: v_sub_co_u32_e32 v6, vcc, v6, v7
-; GFX9-NEXT: v_subb_co_u32_e32 v7, vcc, v8, v10, vcc
-; GFX9-NEXT: v_mov_b32_e32 v9, 0
-; GFX9-NEXT: v_subbrev_co_u32_e32 v8, vcc, 0, v9, vcc
-; GFX9-NEXT: v_subbrev_co_u32_e32 v9, vcc, 0, v9, vcc
-; GFX9-NEXT: v_cmp_lt_u64_e32 vcc, s[6:7], v[6:7]
-; GFX9-NEXT: v_or_b32_e32 v13, v7, v9
-; GFX9-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GFX9-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[8:9]
-; GFX9-NEXT: v_mov_b32_e32 v22, v20
-; GFX9-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[8:9]
-; GFX9-NEXT: v_cndmask_b32_e32 v10, v11, v10, vcc
-; GFX9-NEXT: v_and_b32_e32 v10, 1, v10
-; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 1, v10
-; GFX9-NEXT: v_xor_b32_e32 v10, 0x7f, v6
-; GFX9-NEXT: v_or_b32_e32 v12, v10, v8
-; GFX9-NEXT: s_or_b64 s[4:5], s[4:5], vcc
-; GFX9-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[12:13]
-; GFX9-NEXT: s_xor_b64 s[6:7], s[4:5], -1
-; GFX9-NEXT: v_cndmask_b32_e64 v11, v1, 0, s[4:5]
-; GFX9-NEXT: v_cndmask_b32_e64 v12, v0, 0, s[4:5]
-; GFX9-NEXT: v_cndmask_b32_e64 v10, v3, 0, s[4:5]
-; GFX9-NEXT: v_cndmask_b32_e64 v13, v2, 0, s[4:5]
-; GFX9-NEXT: s_and_b64 s[4:5], s[6:7], vcc
-; GFX9-NEXT: s_and_saveexec_b64 s[8:9], s[4:5]
-; GFX9-NEXT: s_cbranch_execz .LBB2_6
-; GFX9-NEXT: ; %bb.1: ; %udiv-bb1
-; GFX9-NEXT: v_add_co_u32_e32 v24, vcc, 1, v6
-; GFX9-NEXT: v_addc_co_u32_e32 v25, vcc, 0, v7, vcc
-; GFX9-NEXT: v_addc_co_u32_e32 v26, vcc, 0, v8, vcc
-; GFX9-NEXT: v_sub_u32_e32 v13, 0x7f, v6
-; GFX9-NEXT: v_addc_co_u32_e32 v27, vcc, 0, v9, vcc
-; GFX9-NEXT: v_sub_u32_e32 v11, 64, v13
-; GFX9-NEXT: v_or_b32_e32 v8, v25, v27
-; GFX9-NEXT: v_or_b32_e32 v7, v24, v26
-; GFX9-NEXT: v_lshlrev_b64 v[9:10], v13, v[0:1]
-; GFX9-NEXT: v_lshrrev_b64 v[11:12], v11, v[2:3]
-; GFX9-NEXT: v_sub_u32_e32 v6, 63, v6
-; GFX9-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[7:8]
-; GFX9-NEXT: v_lshlrev_b64 v[6:7], v6, v[2:3]
-; GFX9-NEXT: v_or_b32_e32 v8, v10, v12
-; GFX9-NEXT: v_or_b32_e32 v9, v9, v11
-; GFX9-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v13
-; GFX9-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v13
-; GFX9-NEXT: v_lshlrev_b64 v[12:13], v13, v[2:3]
-; GFX9-NEXT: v_cndmask_b32_e64 v7, v7, v8, s[4:5]
-; GFX9-NEXT: v_cndmask_b32_e64 v6, v6, v9, s[4:5]
-; GFX9-NEXT: v_mov_b32_e32 v8, 0
-; GFX9-NEXT: v_mov_b32_e32 v10, 0
-; GFX9-NEXT: v_cndmask_b32_e64 v7, v7, v1, s[6:7]
-; GFX9-NEXT: v_cndmask_b32_e64 v6, v6, v0, s[6:7]
-; GFX9-NEXT: v_cndmask_b32_e64 v13, 0, v13, s[4:5]
-; GFX9-NEXT: v_mov_b32_e32 v9, 0
-; GFX9-NEXT: v_mov_b32_e32 v11, 0
-; GFX9-NEXT: v_cndmask_b32_e64 v12, 0, v12, s[4:5]
-; GFX9-NEXT: s_and_saveexec_b64 s[4:5], vcc
-; GFX9-NEXT: s_xor_b64 s[6:7], exec, s[4:5]
-; GFX9-NEXT: s_cbranch_execz .LBB2_5
-; GFX9-NEXT: ; %bb.2: ; %udiv-preheader
-; GFX9-NEXT: v_sub_u32_e32 v10, 64, v24
-; GFX9-NEXT: v_lshrrev_b64 v[8:9], v24, v[2:3]
-; GFX9-NEXT: v_lshlrev_b64 v[10:11], v10, v[0:1]
-; GFX9-NEXT: v_cmp_gt_u32_e32 vcc, 64, v24
-; GFX9-NEXT: v_or_b32_e32 v10, v8, v10
-; GFX9-NEXT: v_subrev_u32_e32 v8, 64, v24
-; GFX9-NEXT: v_or_b32_e32 v11, v9, v11
-; GFX9-NEXT: v_lshrrev_b64 v[8:9], v8, v[0:1]
-; GFX9-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v24
-; GFX9-NEXT: v_cndmask_b32_e32 v9, v9, v11, vcc
-; GFX9-NEXT: v_cndmask_b32_e64 v15, v9, v3, s[4:5]
-; GFX9-NEXT: v_cndmask_b32_e32 v10, v8, v10, vcc
-; GFX9-NEXT: v_lshrrev_b64 v[8:9], v24, v[0:1]
-; GFX9-NEXT: v_cndmask_b32_e64 v14, v10, v2, s[4:5]
-; GFX9-NEXT: v_cndmask_b32_e32 v17, 0, v9, vcc
-; GFX9-NEXT: v_cndmask_b32_e32 v16, 0, v8, vcc
-; GFX9-NEXT: v_add_co_u32_e32 v28, vcc, -1, v23
-; GFX9-NEXT: v_addc_co_u32_e32 v29, vcc, -1, v21, vcc
-; GFX9-NEXT: v_addc_co_u32_e32 v30, vcc, -1, v4, vcc
-; GFX9-NEXT: v_mov_b32_e32 v18, 0
-; GFX9-NEXT: v_mov_b32_e32 v10, 0
-; GFX9-NEXT: v_addc_co_u32_e32 v31, vcc, -1, v5, vcc
-; GFX9-NEXT: s_mov_b64 s[4:5], 0
-; GFX9-NEXT: v_mov_b32_e32 v19, 0
-; GFX9-NEXT: v_mov_b32_e32 v11, 0
-; GFX9-NEXT: v_mov_b32_e32 v9, 0
-; GFX9-NEXT: .LBB2_3: ; %udiv-do-while
-; GFX9-NEXT: ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT: v_lshrrev_b32_e32 v32, 31, v15
-; GFX9-NEXT: v_lshlrev_b64 v[14:15], 1, v[14:15]
-; GFX9-NEXT: v_lshrrev_b32_e32 v33, 31, v7
-; GFX9-NEXT: v_lshlrev_b64 v[6:7], 1, v[6:7]
-; GFX9-NEXT: v_lshrrev_b32_e32 v8, 31, v13
-; GFX9-NEXT: v_lshlrev_b64 v[16:17], 1, v[16:17]
-; GFX9-NEXT: v_or_b32_e32 v14, v14, v33
-; GFX9-NEXT: v_or3_b32 v6, v6, v8, v10
-; GFX9-NEXT: v_sub_co_u32_e32 v8, vcc, v28, v14
-; GFX9-NEXT: v_or_b32_e32 v16, v16, v32
-; GFX9-NEXT: v_subb_co_u32_e32 v8, vcc, v29, v15, vcc
-; GFX9-NEXT: v_subb_co_u32_e32 v8, vcc, v30, v16, vcc
-; GFX9-NEXT: v_lshlrev_b64 v[12:13], 1, v[12:13]
-; GFX9-NEXT: v_subb_co_u32_e32 v8, vcc, v31, v17, vcc
-; GFX9-NEXT: v_ashrrev_i32_e32 v8, 31, v8
-; GFX9-NEXT: v_or_b32_e32 v12, v18, v12
-; GFX9-NEXT: v_and_b32_e32 v18, v8, v23
-; GFX9-NEXT: v_or_b32_e32 v13, v19, v13
-; GFX9-NEXT: v_and_b32_e32 v19, v8, v21
-; GFX9-NEXT: v_sub_co_u32_e32 v14, vcc, v14, v18
-; GFX9-NEXT: v_and_b32_e32 v32, v8, v4
-; GFX9-NEXT: v_subb_co_u32_e32 v15, vcc, v15, v19, vcc
-; GFX9-NEXT: v_and_b32_e32 v33, v8, v5
-; GFX9-NEXT: v_subb_co_u32_e32 v16, vcc, v16, v32, vcc
-; GFX9-NEXT: v_subb_co_u32_e32 v17, vcc, v17, v33, vcc
-; GFX9-NEXT: v_add_co_u32_e32 v24, vcc, -1, v24
-; GFX9-NEXT: v_addc_co_u32_e32 v25, vcc, -1, v25, vcc
-; GFX9-NEXT: v_addc_co_u32_e32 v26, vcc, -1, v26, vcc
-; GFX9-NEXT: v_addc_co_u32_e32 v27, vcc, -1, v27, vcc
-; GFX9-NEXT: v_or_b32_e32 v18, v24, v26
-; GFX9-NEXT: v_or_b32_e32 v19, v25, v27
-; GFX9-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[18:19]
-; GFX9-NEXT: v_and_b32_e32 v8, 1, v8
-; GFX9-NEXT: v_mov_b32_e32 v19, v9
-; GFX9-NEXT: v_or3_b32 v7, v7, 0, v11
-; GFX9-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
-; GFX9-NEXT: v_mov_b32_e32 v18, v8
-; GFX9-NEXT: s_andn2_b64 exec, exec, s[4:5]
-; GFX9-NEXT: s_cbranch_execnz .LBB2_3
-; GFX9-NEXT: ; %bb.4: ; %Flow
-; GFX9-NEXT: s_or_b64 exec, exec, s[4:5]
-; GFX9-NEXT: .LBB2_5: ; %Flow2
-; GFX9-NEXT: s_or_b64 exec, exec, s[6:7]
-; GFX9-NEXT: v_lshlrev_b64 v[14:15], 1, v[12:13]
-; GFX9-NEXT: v_lshlrev_b64 v[6:7], 1, v[6:7]
-; GFX9-NEXT: v_lshrrev_b32_e32 v12, 31, v13
-; GFX9-NEXT: v_or3_b32 v11, v7, 0, v11
-; GFX9-NEXT: v_or3_b32 v12, v6, v12, v10
-; GFX9-NEXT: v_or_b32_e32 v10, v9, v15
-; GFX9-NEXT: v_or_b32_e32 v13, v8, v14
-; GFX9-NEXT: .LBB2_6: ; %Flow3
-; GFX9-NEXT: s_or_b64 exec, exec, s[8:9]
-; GFX9-NEXT: v_mul_lo_u32 v16, v13, v5
-; GFX9-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v23, v13, 0
-; GFX9-NEXT: v_mov_b32_e32 v15, 0
-; GFX9-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v13, v4, 0
-; GFX9-NEXT: v_mov_b32_e32 v14, v6
-; GFX9-NEXT: v_mad_u64_u32 v[13:14], s[4:5], v21, v13, v[14:15]
-; GFX9-NEXT: v_mul_lo_u32 v9, v10, v4
-; GFX9-NEXT: v_mul_lo_u32 v11, v11, v23
-; GFX9-NEXT: v_mov_b32_e32 v4, v14
-; GFX9-NEXT: v_mov_b32_e32 v14, v15
-; GFX9-NEXT: v_mad_u64_u32 v[13:14], s[4:5], v23, v10, v[13:14]
-; GFX9-NEXT: v_add3_u32 v8, v8, v16, v9
-; GFX9-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v12, v23, v[7:8]
-; GFX9-NEXT: v_mov_b32_e32 v8, v14
-; GFX9-NEXT: v_add_co_u32_e32 v8, vcc, v4, v8
-; GFX9-NEXT: v_addc_co_u32_e64 v9, s[4:5], 0, 0, vcc
-; GFX9-NEXT: v_mul_lo_u32 v12, v12, v21
-; GFX9-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v21, v10, v[8:9]
-; GFX9-NEXT: v_add3_u32 v4, v11, v7, v12
-; GFX9-NEXT: v_add_co_u32_e32 v6, vcc, v8, v6
-; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, v9, v4, vcc
-; GFX9-NEXT: v_mov_b32_e32 v7, v13
-; GFX9-NEXT: v_sub_co_u32_e32 v2, vcc, v2, v5
-; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v3, v7, vcc
-; GFX9-NEXT: v_subb_co_u32_e32 v0, vcc, v0, v6, vcc
-; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v4, vcc
-; GFX9-NEXT: v_xor_b32_e32 v5, v0, v20
-; GFX9-NEXT: v_xor_b32_e32 v0, v2, v20
-; GFX9-NEXT: v_xor_b32_e32 v4, v1, v22
-; GFX9-NEXT: v_xor_b32_e32 v1, v3, v22
-; GFX9-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v20
-; GFX9-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v22, vcc
-; GFX9-NEXT: v_subb_co_u32_e32 v2, vcc, v5, v20, vcc
-; GFX9-NEXT: v_subb_co_u32_e32 v3, vcc, v4, v22, vcc
-; GFX9-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX9-O0-LABEL: v_srem_i128_vv:
-; GFX9-O0: ; %bb.0: ; %_udiv-special-cases
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-O0-NEXT: s_xor_saveexec_b64 s[4:5], -1
-; GFX9-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:348 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:352 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:356 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 offset:360 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_mov_b64 exec, s[4:5]
-; GFX9-O0-NEXT: ; implicit-def: $vgpr8 : SGPR spill to VGPR lane
-; GFX9-O0-NEXT: v_mov_b32_e32 v8, v6
-; GFX9-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:120 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:116 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_mov_b32_e32 v14, v2
-; GFX9-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:120 ; 4-byte Folded Reload
-; GFX9-O0-NEXT: v_mov_b32_e32 v6, v1
-; GFX9-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:116 ; 4-byte Folded Reload
-; GFX9-O0-NEXT: v_mov_b32_e32 v4, v0
-; GFX9-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
-; GFX9-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
-; GFX9-O0-NEXT: s_mov_b64 exec, s[18:19]
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v3, v5
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 def $vgpr4_vgpr5 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v5, v6
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; kill: def $vgpr8 killed $vgpr8 def $vgpr8_vgpr9 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v9, v7
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; kill: def $vgpr14 killed $vgpr14 def $vgpr14_vgpr15 killed $exec
-; GFX9-O0-NEXT: s_waitcnt vmcnt(1)
-; GFX9-O0-NEXT: v_mov_b32_e32 v15, v1
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4_sgpr5
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4_sgpr5
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4_sgpr5
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4_sgpr5
-; GFX9-O0-NEXT: s_mov_b32 s4, 63
-; GFX9-O0-NEXT: v_mov_b32_e32 v6, v14
-; GFX9-O0-NEXT: v_mov_b32_e32 v7, v15
-; GFX9-O0-NEXT: v_ashrrev_i64 v[12:13], s4, v[6:7]
-; GFX9-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:108 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:112 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_mov_b32_e32 v6, v12
-; GFX9-O0-NEXT: v_mov_b32_e32 v7, v13
-; GFX9-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:104 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_mov_b32_e32 v6, v8
-; GFX9-O0-NEXT: v_mov_b32_e32 v7, v9
-; GFX9-O0-NEXT: v_ashrrev_i64 v[6:7], s4, v[6:7]
-; GFX9-O0-NEXT: v_mov_b32_e32 v1, v15
-; GFX9-O0-NEXT: v_mov_b32_e32 v10, v13
-; GFX9-O0-NEXT: v_xor_b32_e64 v1, v1, v10
-; GFX9-O0-NEXT: v_mov_b32_e32 v11, v14
-; GFX9-O0-NEXT: v_xor_b32_e64 v13, v11, v12
-; GFX9-O0-NEXT: ; kill: def $vgpr13 killed $vgpr13 def $vgpr13_vgpr14 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v14, v1
-; GFX9-O0-NEXT: v_mov_b32_e32 v1, v5
-; GFX9-O0-NEXT: v_xor_b32_e64 v1, v1, v10
-; GFX9-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 killed $vgpr4_vgpr5 killed $exec
-; GFX9-O0-NEXT: v_xor_b32_e64 v15, v4, v12
-; GFX9-O0-NEXT: ; kill: def $vgpr15 killed $vgpr15 def $vgpr15_vgpr16 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v16, v1
-; GFX9-O0-NEXT: v_mov_b32_e32 v1, v9
-; GFX9-O0-NEXT: v_mov_b32_e32 v4, v7
-; GFX9-O0-NEXT: v_xor_b32_e64 v1, v1, v4
-; GFX9-O0-NEXT: v_mov_b32_e32 v5, v8
-; GFX9-O0-NEXT: ; kill: def $vgpr6 killed $vgpr6 killed $vgpr6_vgpr7 killed $exec
-; GFX9-O0-NEXT: v_xor_b32_e64 v7, v5, v6
-; GFX9-O0-NEXT: ; kill: def $vgpr7 killed $vgpr7 def $vgpr7_vgpr8 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v8, v1
-; GFX9-O0-NEXT: v_mov_b32_e32 v1, v3
-; GFX9-O0-NEXT: v_xor_b32_e64 v1, v1, v4
-; GFX9-O0-NEXT: ; kill: def $vgpr2 killed $vgpr2 killed $vgpr2_vgpr3 killed $exec
-; GFX9-O0-NEXT: v_xor_b32_e64 v2, v2, v6
-; GFX9-O0-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v3, v1
-; GFX9-O0-NEXT: v_mov_b32_e32 v9, v15
-; GFX9-O0-NEXT: v_mov_b32_e32 v1, v16
-; GFX9-O0-NEXT: v_mov_b32_e32 v11, v13
-; GFX9-O0-NEXT: v_mov_b32_e32 v5, v14
-; GFX9-O0-NEXT: v_sub_co_u32_e32 v9, vcc, v9, v12
-; GFX9-O0-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v10, vcc
-; GFX9-O0-NEXT: v_subb_co_u32_e32 v13, vcc, v11, v12, vcc
-; GFX9-O0-NEXT: v_subb_co_u32_e32 v5, vcc, v5, v10, vcc
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; kill: def $vgpr13 killed $vgpr13 def $vgpr13_vgpr14 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v14, v5
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v10, v1
-; GFX9-O0-NEXT: v_mov_b32_e32 v1, v2
-; GFX9-O0-NEXT: ; kill: def $vgpr3 killed $vgpr3 killed $vgpr2_vgpr3 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v5, v7
-; GFX9-O0-NEXT: v_mov_b32_e32 v2, v8
-; GFX9-O0-NEXT: v_sub_co_u32_e32 v1, vcc, v1, v6
-; GFX9-O0-NEXT: v_subb_co_u32_e32 v3, vcc, v3, v4, vcc
-; GFX9-O0-NEXT: v_subb_co_u32_e32 v11, vcc, v5, v6, vcc
-; GFX9-O0-NEXT: v_subb_co_u32_e32 v2, vcc, v2, v4, vcc
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; kill: def $vgpr11 killed $vgpr11 def $vgpr11_vgpr12 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v12, v2
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; implicit-def: $sgpr4
-; GFX9-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v2, v3
-; GFX9-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v10, off, s[0:3], s32 offset:96 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_mov_b32_e32 v3, v11
-; GFX9-O0-NEXT: v_mov_b32_e32 v4, v12
-; GFX9-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_mov_b32_e32 v4, v2
-; GFX9-O0-NEXT: v_mov_b32_e32 v3, v1
-; GFX9-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_mov_b32_e32 v3, v13
-; GFX9-O0-NEXT: v_mov_b32_e32 v4, v14
-; GFX9-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_mov_b32_e32 v3, v9
-; GFX9-O0-NEXT: v_mov_b32_e32 v4, v10
-; GFX9-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
-; GFX9-O0-NEXT: v_mov_b32_e32 v7, v12
-; GFX9-O0-NEXT: v_mov_b32_e32 v8, v2
-; GFX9-O0-NEXT: v_or_b32_e64 v3, v8, v7
-; GFX9-O0-NEXT: v_mov_b32_e32 v6, v11
-; GFX9-O0-NEXT: v_mov_b32_e32 v5, v1
-; GFX9-O0-NEXT: v_or_b32_e64 v1, v5, v6
-; GFX9-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2 killed $exec
-; GFX9-O0-NEXT: v_mov_b32_e32 v2, v3
-; GFX9-O0-NEXT: s_mov_b64 s[6:7], 0
-; GFX9-O0-NEXT: v_writelane_b32 v0, s6, 0
-; GFX9-O0-NEXT: v_writelane_b32 v0, s7, 1
-; GFX9-O0-NEXT: v_cmp...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/83307
More information about the llvm-commits
mailing list