[llvm] [AMDGPU] Ensure divergence for v_alignbit (PR #129159)
Jeffrey Byrnes via llvm-commits
llvm-commits at lists.llvm.org
Thu Feb 27 16:10:33 PST 2025
https://github.com/jrbyrnes created https://github.com/llvm/llvm-project/pull/129159
Selecting vgpr for the uniform version of this pattern may lead to unnecessary vgpr and waterfall loops.
>From 7767fc6a848947cc02be44c12f4047d93a881364 Mon Sep 17 00:00:00 2001
From: Jeffrey Byrnes <Jeffrey.Byrnes at amd.com>
Date: Thu, 27 Feb 2025 15:20:57 -0800
Subject: [PATCH] [AMDGPU] Ensure divergence for v_alignbit
Change-Id: I2c9f75da895fb83755e57fd8cc36f850222d7c8c
---
llvm/lib/Target/AMDGPU/SIInstructions.td | 4 +-
llvm/test/CodeGen/AMDGPU/alignbit-pat.ll | 6 +-
.../AMDGPU/amdgpu-codegenprepare-idiv.ll | 750 ++++++++--------
.../branch-folding-implicit-def-subreg.ll | 846 +++++++++---------
.../CodeGen/AMDGPU/call-argument-types.ll | 8 +-
.../CodeGen/AMDGPU/cgp-bitfield-extract.ll | 5 +-
llvm/test/CodeGen/AMDGPU/fmed3.ll | 11 +-
llvm/test/CodeGen/AMDGPU/function-returns.ll | 10 +-
.../CodeGen/AMDGPU/insert-delay-alu-bug.ll | 12 +-
llvm/test/CodeGen/AMDGPU/minimummaximum.ll | 34 +-
llvm/test/CodeGen/AMDGPU/minmax.ll | 31 +-
llvm/test/CodeGen/AMDGPU/mul_int24.ll | 17 +-
llvm/test/CodeGen/AMDGPU/sdiv64.ll | 70 +-
.../test/CodeGen/AMDGPU/shift-and-i64-ubfe.ll | 4 +-
.../CodeGen/AMDGPU/shrink-add-sub-constant.ll | 17 +-
llvm/test/CodeGen/AMDGPU/srem64.ll | 82 +-
llvm/test/CodeGen/AMDGPU/udiv64.ll | 32 +-
llvm/test/CodeGen/AMDGPU/uniform-alignbit.ll | 38 +
llvm/test/CodeGen/AMDGPU/urem64.ll | 24 +-
llvm/test/CodeGen/AMDGPU/v_pack.ll | 36 +-
20 files changed, 1040 insertions(+), 997 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/uniform-alignbit.ll
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 6f80dbcfe5e71..e618cbebadb7c 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -2500,11 +2500,11 @@ def : AMDGPUPat <
def : ROTRPattern <V_ALIGNBIT_B32_e64>;
-def : GCNPat<(i32 (trunc (srl i64:$src0, (and i32:$src1, (i32 31))))),
+def : GCNPat<(i32 (DivergentUnaryFrag<trunc> (srl i64:$src0, (and i32:$src1, (i32 31))))),
(V_ALIGNBIT_B32_e64 (i32 (EXTRACT_SUBREG (i64 $src0), sub1)),
(i32 (EXTRACT_SUBREG (i64 $src0), sub0)), $src1)>;
-def : GCNPat<(i32 (trunc (srl i64:$src0, (i32 ShiftAmt32Imm:$src1)))),
+def : GCNPat<(i32 (DivergentUnaryFrag<trunc> (srl i64:$src0, (i32 ShiftAmt32Imm:$src1)))),
(V_ALIGNBIT_B32_e64 (i32 (EXTRACT_SUBREG (i64 $src0), sub1)),
(i32 (EXTRACT_SUBREG (i64 $src0), sub0)), $src1)>;
diff --git a/llvm/test/CodeGen/AMDGPU/alignbit-pat.ll b/llvm/test/CodeGen/AMDGPU/alignbit-pat.ll
index fc1326268fd1e..6451677db313c 100644
--- a/llvm/test/CodeGen/AMDGPU/alignbit-pat.ll
+++ b/llvm/test/CodeGen/AMDGPU/alignbit-pat.ll
@@ -3,7 +3,8 @@
; GCN-LABEL: {{^}}alignbit_shr_pat:
; GCN-DAG: s_load_dword s[[SHR:[0-9]+]]
; GCN-DAG: load_dwordx2 v[[[LO:[0-9]+]]:[[HI:[0-9]+]]]
-; GCN: v_alignbit_b32 v{{[0-9]+}}, v[[HI]], v[[LO]], s[[SHR]]
+; GCN-NEXT: s_mov_b32
+; GCN-NEXT: s_mov_b32
define amdgpu_kernel void @alignbit_shr_pat(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1, i32 %arg2) {
bb:
@@ -70,7 +71,8 @@ bb:
; GCN-LABEL: {{^}}alignbit_shr_pat_const30:
; GCN: load_dwordx2 v[[[LO:[0-9]+]]:[[HI:[0-9]+]]]
-; GCN: v_alignbit_b32 v{{[0-9]+}}, v[[HI]], v[[LO]], 30
+; GCN-NEXT: s_mov_b32
+; GCN-NEXT: s_mov_b32
define amdgpu_kernel void @alignbit_shr_pat_const30(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1) {
bb:
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll
index 8e16889c72e65..8e8aa409eafec 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll
@@ -4610,50 +4610,48 @@ define amdgpu_kernel void @udiv_v3i15(ptr addrspace(1) %out, <3 x i15> %x, <3 x
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: s_lshr_b64 s[6:7], s[10:11], 30
+; GFX6-NEXT: s_mov_b32 s0, s8
+; GFX6-NEXT: s_and_b32 s8, s6, 0x7fff
+; GFX6-NEXT: s_and_b32 s6, s4, 0x7fff
+; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s6
; GFX6-NEXT: s_and_b32 s6, s10, 0x7fff
-; GFX6-NEXT: s_and_b32 s7, s4, 0x7fff
-; GFX6-NEXT: v_cvt_f32_u32_e32 v1, s7
-; GFX6-NEXT: v_mov_b32_e32 v2, s4
+; GFX6-NEXT: v_cvt_f32_u32_e32 v1, s6
+; GFX6-NEXT: s_lshr_b64 s[6:7], s[4:5], 30
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v2, v0
; GFX6-NEXT: s_bfe_u32 s4, s4, 0xf000f
-; GFX6-NEXT: v_cvt_f32_u32_e32 v3, s6
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v4, v1
-; GFX6-NEXT: v_cvt_f32_u32_e32 v5, s4
-; GFX6-NEXT: s_bfe_u32 s7, s10, 0xf000f
-; GFX6-NEXT: v_alignbit_b32 v2, s5, v2, 30
-; GFX6-NEXT: v_mul_f32_e32 v4, v3, v4
-; GFX6-NEXT: v_cvt_f32_u32_e32 v6, s7
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v7, v5
-; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v2
-; GFX6-NEXT: v_trunc_f32_e32 v4, v4
-; GFX6-NEXT: v_mad_f32 v3, -v4, v1, v3
-; GFX6-NEXT: v_cvt_u32_f32_e32 v4, v4
-; GFX6-NEXT: v_cvt_f32_u32_e32 v2, v2
-; GFX6-NEXT: v_mov_b32_e32 v0, s10
-; GFX6-NEXT: v_alignbit_b32 v0, s11, v0, 30
-; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v1
-; GFX6-NEXT: v_mul_f32_e32 v1, v6, v7
-; GFX6-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX6-NEXT: v_trunc_f32_e32 v1, v1
-; GFX6-NEXT: v_addc_u32_e32 v3, vcc, 0, v4, vcc
-; GFX6-NEXT: v_mad_f32 v4, -v1, v5, v6
-; GFX6-NEXT: v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT: v_cvt_f32_u32_e32 v0, v0
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v6, v2
-; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, v5
-; GFX6-NEXT: v_addc_u32_e32 v4, vcc, 0, v1, vcc
-; GFX6-NEXT: v_mul_f32_e32 v1, v0, v6
-; GFX6-NEXT: v_trunc_f32_e32 v1, v1
-; GFX6-NEXT: v_cvt_u32_f32_e32 v5, v1
-; GFX6-NEXT: v_mad_f32 v0, -v1, v2, v0
-; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v2
-; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v3
-; GFX6-NEXT: v_addc_u32_e32 v0, vcc, 0, v5, vcc
-; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v4
+; GFX6-NEXT: v_cvt_f32_u32_e32 v3, s4
+; GFX6-NEXT: s_mov_b32 s1, s9
+; GFX6-NEXT: s_bfe_u32 s9, s10, 0xf000f
+; GFX6-NEXT: v_mul_f32_e32 v2, v1, v2
+; GFX6-NEXT: v_trunc_f32_e32 v2, v2
+; GFX6-NEXT: v_cvt_f32_u32_e32 v4, s9
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v5, v3
+; GFX6-NEXT: s_and_b32 s5, s6, 0x7fff
+; GFX6-NEXT: v_mad_f32 v1, -v2, v0, v1
+; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, v0
+; GFX6-NEXT: v_cvt_f32_u32_e32 v1, s5
+; GFX6-NEXT: v_cvt_u32_f32_e32 v2, v2
+; GFX6-NEXT: v_mul_f32_e32 v0, v4, v5
+; GFX6-NEXT: v_trunc_f32_e32 v0, v0
+; GFX6-NEXT: v_mad_f32 v4, -v0, v3, v4
+; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0
+; GFX6-NEXT: v_cvt_f32_u32_e32 v5, s8
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v6, v1
+; GFX6-NEXT: v_addc_u32_e32 v2, vcc, 0, v2, vcc
+; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, v3
+; GFX6-NEXT: v_addc_u32_e32 v3, vcc, 0, v0, vcc
+; GFX6-NEXT: v_mul_f32_e32 v0, v5, v6
+; GFX6-NEXT: v_trunc_f32_e32 v0, v0
+; GFX6-NEXT: v_cvt_u32_f32_e32 v4, v0
+; GFX6-NEXT: v_mad_f32 v0, -v0, v1, v5
+; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1
+; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v3
+; GFX6-NEXT: v_addc_u32_e32 v0, vcc, 0, v4, vcc
; GFX6-NEXT: v_lshl_b64 v[0:1], v[0:1], 30
+; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v2
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 15, v3
; GFX6-NEXT: v_or_b32_e32 v2, v3, v2
-; GFX6-NEXT: s_mov_b32 s0, s8
-; GFX6-NEXT: s_mov_b32 s1, s9
; GFX6-NEXT: v_or_b32_e32 v0, v2, v0
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_waitcnt expcnt(0)
@@ -4667,46 +4665,44 @@ define amdgpu_kernel void @udiv_v3i15(ptr addrspace(1) %out, <3 x i15> %x, <3 x
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-NEXT: v_mov_b32_e32 v0, s2
-; GFX9-NEXT: s_and_b32 s5, s6, 0x7fff
+; GFX9-NEXT: s_lshr_b64 s[4:5], s[2:3], 30
+; GFX9-NEXT: s_and_b32 s3, s6, 0x7fff
+; GFX9-NEXT: v_cvt_f32_u32_e32 v0, s3
+; GFX9-NEXT: s_and_b32 s5, s2, 0x7fff
+; GFX9-NEXT: s_bfe_u32 s8, s2, 0xf000f
+; GFX9-NEXT: s_lshr_b64 s[2:3], s[6:7], 30
; GFX9-NEXT: v_cvt_f32_u32_e32 v1, s5
-; GFX9-NEXT: s_and_b32 s4, s2, 0x7fff
-; GFX9-NEXT: v_alignbit_b32 v0, s3, v0, 30
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v3, v0
; GFX9-NEXT: s_bfe_u32 s3, s6, 0xf000f
-; GFX9-NEXT: v_cvt_f32_u32_e32 v4, s4
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v5, v1
-; GFX9-NEXT: v_cvt_f32_u32_e32 v6, s3
-; GFX9-NEXT: s_bfe_u32 s2, s2, 0xf000f
-; GFX9-NEXT: v_mov_b32_e32 v3, s6
-; GFX9-NEXT: v_alignbit_b32 v3, s7, v3, 30
-; GFX9-NEXT: v_mul_f32_e32 v5, v4, v5
-; GFX9-NEXT: v_cvt_f32_u32_e32 v7, s2
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v8, v6
-; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v3
-; GFX9-NEXT: v_trunc_f32_e32 v5, v5
-; GFX9-NEXT: v_mad_f32 v4, -v5, v1, v4
-; GFX9-NEXT: v_cvt_u32_f32_e32 v5, v5
-; GFX9-NEXT: v_cvt_f32_u32_e32 v3, v3
-; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, v1
-; GFX9-NEXT: v_mul_f32_e32 v1, v7, v8
-; GFX9-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX9-NEXT: v_trunc_f32_e32 v1, v1
-; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v5, vcc
-; GFX9-NEXT: v_mad_f32 v5, -v1, v6, v7
-; GFX9-NEXT: v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT: v_cvt_f32_u32_e32 v0, v0
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v7, v3
-; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v5|, v6
-; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v1, vcc
-; GFX9-NEXT: v_mul_f32_e32 v1, v0, v7
-; GFX9-NEXT: v_trunc_f32_e32 v1, v1
-; GFX9-NEXT: v_cvt_u32_f32_e32 v6, v1
-; GFX9-NEXT: v_mad_f32 v0, -v1, v3, v0
-; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v3
-; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v4
-; GFX9-NEXT: v_addc_co_u32_e32 v0, vcc, 0, v6, vcc
-; GFX9-NEXT: v_and_b32_e32 v4, 0x7fff, v5
+; GFX9-NEXT: v_cvt_f32_u32_e32 v4, s3
+; GFX9-NEXT: v_cvt_f32_u32_e32 v5, s8
+; GFX9-NEXT: v_mul_f32_e32 v3, v1, v3
+; GFX9-NEXT: v_trunc_f32_e32 v3, v3
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v6, v4
+; GFX9-NEXT: s_and_b32 s2, s2, 0x7fff
+; GFX9-NEXT: v_mad_f32 v1, -v3, v0, v1
+; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, v0
+; GFX9-NEXT: v_cvt_f32_u32_e32 v1, s2
+; GFX9-NEXT: v_cvt_u32_f32_e32 v3, v3
+; GFX9-NEXT: v_mul_f32_e32 v0, v5, v6
+; GFX9-NEXT: s_and_b32 s4, s4, 0x7fff
+; GFX9-NEXT: v_trunc_f32_e32 v0, v0
+; GFX9-NEXT: v_mad_f32 v5, -v0, v4, v5
+; GFX9-NEXT: v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT: v_cvt_f32_u32_e32 v6, s4
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v7, v1
+; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
+; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v5|, v4
+; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v0, vcc
+; GFX9-NEXT: v_mul_f32_e32 v0, v6, v7
+; GFX9-NEXT: v_trunc_f32_e32 v0, v0
+; GFX9-NEXT: v_cvt_u32_f32_e32 v5, v0
+; GFX9-NEXT: v_mad_f32 v0, -v0, v1, v6
+; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1
+; GFX9-NEXT: v_and_b32_e32 v4, 0x7fff, v4
+; GFX9-NEXT: v_addc_co_u32_e32 v0, vcc, 0, v5, vcc
; GFX9-NEXT: v_lshlrev_b64 v[0:1], 30, v[0:1]
+; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v3
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 15, v4
; GFX9-NEXT: v_or_b32_e32 v3, v3, v4
; GFX9-NEXT: v_or_b32_e32 v0, v3, v0
@@ -4797,58 +4793,56 @@ define amdgpu_kernel void @urem_v3i15(ptr addrspace(1) %out, <3 x i15> %x, <3 x
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: s_lshr_b64 s[6:7], s[10:11], 30
; GFX6-NEXT: s_mov_b32 s0, s8
-; GFX6-NEXT: s_and_b32 s8, s4, 0x7fff
-; GFX6-NEXT: v_cvt_f32_u32_e32 v1, s8
-; GFX6-NEXT: s_and_b32 s7, s10, 0x7fff
-; GFX6-NEXT: v_cvt_f32_u32_e32 v3, s7
-; GFX6-NEXT: v_mov_b32_e32 v2, s4
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v4, v1
-; GFX6-NEXT: v_alignbit_b32 v2, s5, v2, 30
-; GFX6-NEXT: s_bfe_u32 s5, s4, 0xf000f
-; GFX6-NEXT: v_cvt_f32_u32_e32 v5, s5
-; GFX6-NEXT: v_mul_f32_e32 v4, v3, v4
-; GFX6-NEXT: v_trunc_f32_e32 v4, v4
-; GFX6-NEXT: v_mad_f32 v3, -v4, v1, v3
-; GFX6-NEXT: v_cvt_u32_f32_e32 v4, v4
-; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v1
-; GFX6-NEXT: s_bfe_u32 s8, s10, 0xf000f
-; GFX6-NEXT: v_cvt_f32_u32_e32 v3, s8
-; GFX6-NEXT: v_addc_u32_e32 v1, vcc, 0, v4, vcc
-; GFX6-NEXT: v_mul_lo_u32 v1, v1, s4
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v4, v5
-; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v2
-; GFX6-NEXT: v_mov_b32_e32 v0, s10
-; GFX6-NEXT: v_sub_i32_e32 v6, vcc, s10, v1
-; GFX6-NEXT: v_mul_f32_e32 v1, v3, v4
-; GFX6-NEXT: v_cvt_f32_u32_e32 v4, v2
-; GFX6-NEXT: v_alignbit_b32 v0, s11, v0, 30
-; GFX6-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX6-NEXT: v_cvt_f32_u32_e32 v7, v0
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v8, v4
-; GFX6-NEXT: v_trunc_f32_e32 v1, v1
-; GFX6-NEXT: v_mad_f32 v3, -v1, v5, v3
-; GFX6-NEXT: v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v5
-; GFX6-NEXT: v_mul_f32_e32 v3, v7, v8
-; GFX6-NEXT: v_trunc_f32_e32 v3, v3
-; GFX6-NEXT: v_cvt_u32_f32_e32 v5, v3
-; GFX6-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; GFX6-NEXT: v_mad_f32 v3, -v3, v4, v7
+; GFX6-NEXT: s_and_b32 s8, s6, 0x7fff
+; GFX6-NEXT: s_and_b32 s6, s4, 0x7fff
+; GFX6-NEXT: v_cvt_f32_u32_e32 v0, s6
+; GFX6-NEXT: s_and_b32 s6, s10, 0x7fff
+; GFX6-NEXT: v_cvt_f32_u32_e32 v1, s6
+; GFX6-NEXT: s_lshr_b64 s[6:7], s[4:5], 30
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v2, v0
+; GFX6-NEXT: s_and_b32 s5, s6, 0x7fff
+; GFX6-NEXT: s_bfe_u32 s6, s4, 0xf000f
+; GFX6-NEXT: v_cvt_f32_u32_e32 v3, s6
+; GFX6-NEXT: v_mul_f32_e32 v2, v1, v2
+; GFX6-NEXT: v_trunc_f32_e32 v2, v2
+; GFX6-NEXT: v_mad_f32 v1, -v2, v0, v1
+; GFX6-NEXT: v_cvt_u32_f32_e32 v2, v2
+; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, v0
+; GFX6-NEXT: s_bfe_u32 s11, s10, 0xf000f
+; GFX6-NEXT: v_cvt_f32_u32_e32 v1, s11
+; GFX6-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc
+; GFX6-NEXT: v_mul_lo_u32 v0, v0, s4
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v2, v3
+; GFX6-NEXT: v_cvt_f32_u32_e32 v5, s8
; GFX6-NEXT: s_lshr_b32 s4, s4, 15
-; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v4
-; GFX6-NEXT: v_mul_lo_u32 v1, v1, s4
-; GFX6-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc
-; GFX6-NEXT: v_mul_lo_u32 v2, v3, v2
-; GFX6-NEXT: s_lshr_b32 s6, s10, 15
-; GFX6-NEXT: v_sub_i32_e32 v3, vcc, s6, v1
-; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v2
-; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v3
-; GFX6-NEXT: v_lshl_b64 v[0:1], v[0:1], 30
-; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v6
-; GFX6-NEXT: v_lshlrev_b32_e32 v3, 15, v3
-; GFX6-NEXT: v_or_b32_e32 v2, v3, v2
+; GFX6-NEXT: v_sub_i32_e32 v4, vcc, s10, v0
+; GFX6-NEXT: v_mul_f32_e32 v0, v1, v2
+; GFX6-NEXT: v_cvt_f32_u32_e32 v2, s5
+; GFX6-NEXT: v_trunc_f32_e32 v0, v0
+; GFX6-NEXT: v_mad_f32 v1, -v0, v3, v1
+; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v6, v2
+; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, v3
; GFX6-NEXT: s_mov_b32 s1, s9
+; GFX6-NEXT: v_addc_u32_e32 v0, vcc, 0, v0, vcc
+; GFX6-NEXT: v_mul_f32_e32 v1, v5, v6
+; GFX6-NEXT: v_trunc_f32_e32 v1, v1
+; GFX6-NEXT: v_cvt_u32_f32_e32 v3, v1
+; GFX6-NEXT: v_mad_f32 v1, -v1, v2, v5
+; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, v2
+; GFX6-NEXT: v_mul_lo_u32 v0, v0, s4
+; GFX6-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc
+; GFX6-NEXT: v_mul_lo_u32 v1, v1, s5
+; GFX6-NEXT: s_lshr_b32 s9, s10, 15
+; GFX6-NEXT: v_sub_i32_e32 v2, vcc, s9, v0
+; GFX6-NEXT: v_sub_i32_e32 v0, vcc, s8, v1
+; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v2
+; GFX6-NEXT: v_lshl_b64 v[0:1], v[0:1], 30
+; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v4
+; GFX6-NEXT: v_lshlrev_b32_e32 v2, 15, v2
+; GFX6-NEXT: v_or_b32_e32 v2, v2, v3
; GFX6-NEXT: v_or_b32_e32 v0, v2, v0
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_waitcnt expcnt(0)
@@ -4862,54 +4856,52 @@ define amdgpu_kernel void @urem_v3i15(ptr addrspace(1) %out, <3 x i15> %x, <3 x
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-NEXT: v_mov_b32_e32 v0, s2
-; GFX9-NEXT: v_alignbit_b32 v0, s3, v0, 30
-; GFX9-NEXT: s_and_b32 s3, s6, 0x7fff
+; GFX9-NEXT: s_lshr_b64 s[4:5], s[2:3], 30
+; GFX9-NEXT: s_and_b32 s5, s6, 0x7fff
+; GFX9-NEXT: v_cvt_f32_u32_e32 v0, s5
+; GFX9-NEXT: s_and_b32 s3, s2, 0x7fff
; GFX9-NEXT: v_cvt_f32_u32_e32 v1, s3
-; GFX9-NEXT: s_and_b32 s4, s2, 0x7fff
-; GFX9-NEXT: v_cvt_f32_u32_e32 v4, s4
-; GFX9-NEXT: s_bfe_u32 s4, s6, 0xf000f
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v5, v1
-; GFX9-NEXT: v_cvt_f32_u32_e32 v6, s4
-; GFX9-NEXT: v_mov_b32_e32 v3, s6
-; GFX9-NEXT: v_alignbit_b32 v3, s7, v3, 30
-; GFX9-NEXT: v_mul_f32_e32 v5, v4, v5
-; GFX9-NEXT: v_trunc_f32_e32 v5, v5
-; GFX9-NEXT: v_mad_f32 v4, -v5, v1, v4
-; GFX9-NEXT: v_cvt_u32_f32_e32 v5, v5
-; GFX9-NEXT: s_bfe_u32 s5, s2, 0xf000f
-; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v3
-; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, v1
-; GFX9-NEXT: v_cvt_f32_u32_e32 v7, s5
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v8, v6
-; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v5, vcc
-; GFX9-NEXT: v_cvt_f32_u32_e32 v5, v3
-; GFX9-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX9-NEXT: v_mul_f32_e32 v4, v7, v8
-; GFX9-NEXT: v_cvt_f32_u32_e32 v8, v0
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v9, v5
+; GFX9-NEXT: s_and_b32 s8, s4, 0x7fff
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v3, v0
+; GFX9-NEXT: s_lshr_b64 s[4:5], s[6:7], 30
+; GFX9-NEXT: s_bfe_u32 s5, s6, 0xf000f
+; GFX9-NEXT: v_cvt_f32_u32_e32 v4, s5
+; GFX9-NEXT: v_mul_f32_e32 v3, v1, v3
+; GFX9-NEXT: v_trunc_f32_e32 v3, v3
+; GFX9-NEXT: v_mad_f32 v1, -v3, v0, v1
+; GFX9-NEXT: v_cvt_u32_f32_e32 v3, v3
+; GFX9-NEXT: s_bfe_u32 s9, s2, 0xf000f
+; GFX9-NEXT: s_and_b32 s3, s4, 0x7fff
+; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, v0
+; GFX9-NEXT: v_cvt_f32_u32_e32 v5, s9
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v6, v4
+; GFX9-NEXT: v_addc_co_u32_e32 v0, vcc, 0, v3, vcc
+; GFX9-NEXT: v_cvt_f32_u32_e32 v3, s3
+; GFX9-NEXT: v_mul_f32_e32 v1, v5, v6
+; GFX9-NEXT: v_cvt_f32_u32_e32 v6, s8
+; GFX9-NEXT: v_trunc_f32_e32 v1, v1
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v7, v3
+; GFX9-NEXT: v_mad_f32 v5, -v1, v4, v5
+; GFX9-NEXT: v_cvt_u32_f32_e32 v1, v1
+; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v5|, v4
+; GFX9-NEXT: v_mul_f32_e32 v4, v6, v7
; GFX9-NEXT: v_trunc_f32_e32 v4, v4
-; GFX9-NEXT: v_mad_f32 v7, -v4, v6, v7
-; GFX9-NEXT: v_cvt_u32_f32_e32 v4, v4
-; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v7|, v6
-; GFX9-NEXT: v_mul_f32_e32 v6, v8, v9
-; GFX9-NEXT: v_trunc_f32_e32 v6, v6
-; GFX9-NEXT: v_cvt_u32_f32_e32 v7, v6
-; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v4, vcc
-; GFX9-NEXT: v_mad_f32 v6, -v6, v5, v8
-; GFX9-NEXT: s_lshr_b32 s3, s6, 15
-; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v6|, v5
-; GFX9-NEXT: v_mul_lo_u32 v4, v4, s3
-; GFX9-NEXT: v_addc_co_u32_e32 v5, vcc, 0, v7, vcc
-; GFX9-NEXT: v_mul_lo_u32 v1, v1, s6
-; GFX9-NEXT: v_mul_lo_u32 v3, v5, v3
+; GFX9-NEXT: v_cvt_u32_f32_e32 v5, v4
+; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
+; GFX9-NEXT: v_mad_f32 v4, -v4, v3, v6
+; GFX9-NEXT: s_lshr_b32 s4, s6, 15
+; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, v3
+; GFX9-NEXT: v_mul_lo_u32 v0, v0, s6
+; GFX9-NEXT: v_mul_lo_u32 v1, v1, s4
+; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v5, vcc
+; GFX9-NEXT: v_mul_lo_u32 v3, v3, s3
; GFX9-NEXT: s_lshr_b32 s3, s2, 15
-; GFX9-NEXT: v_sub_u32_e32 v4, s3, v4
-; GFX9-NEXT: v_sub_u32_e32 v5, s2, v1
-; GFX9-NEXT: v_sub_u32_e32 v0, v0, v3
-; GFX9-NEXT: v_and_b32_e32 v4, 0x7fff, v4
+; GFX9-NEXT: v_sub_u32_e32 v4, s2, v0
+; GFX9-NEXT: v_sub_u32_e32 v5, s3, v1
+; GFX9-NEXT: v_sub_u32_e32 v0, s8, v3
+; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v4
+; GFX9-NEXT: v_and_b32_e32 v4, 0x7fff, v5
; GFX9-NEXT: v_lshlrev_b64 v[0:1], 30, v[0:1]
-; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v5
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 15, v4
; GFX9-NEXT: v_or_b32_e32 v3, v3, v4
; GFX9-NEXT: v_or_b32_e32 v0, v3, v0
@@ -5006,64 +4998,63 @@ define amdgpu_kernel void @sdiv_v3i15(ptr addrspace(1) %out, <3 x i15> %x, <3 x
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
-; GFX6-NEXT: v_mov_b32_e32 v0, s10
-; GFX6-NEXT: s_bfe_i32 s6, s4, 0xf0000
-; GFX6-NEXT: v_cvt_f32_i32_e32 v2, s6
-; GFX6-NEXT: v_mov_b32_e32 v1, s4
-; GFX6-NEXT: v_alignbit_b32 v1, s5, v1, 30
+; GFX6-NEXT: s_lshr_b64 s[6:7], s[10:11], 30
+; GFX6-NEXT: s_bfe_i32 s7, s4, 0xf0000
+; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s7
+; GFX6-NEXT: s_mov_b32 s0, s8
+; GFX6-NEXT: s_mov_b32 s1, s9
+; GFX6-NEXT: s_lshr_b64 s[8:9], s[4:5], 30
; GFX6-NEXT: s_bfe_i32 s5, s10, 0xf0000
-; GFX6-NEXT: v_cvt_f32_i32_e32 v3, s5
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v4, v2
-; GFX6-NEXT: s_xor_b32 s5, s5, s6
+; GFX6-NEXT: v_cvt_f32_i32_e32 v1, s5
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v2, v0
+; GFX6-NEXT: s_xor_b32 s5, s5, s7
; GFX6-NEXT: s_ashr_i32 s5, s5, 30
; GFX6-NEXT: s_or_b32 s5, s5, 1
-; GFX6-NEXT: v_mul_f32_e32 v4, v3, v4
-; GFX6-NEXT: v_trunc_f32_e32 v4, v4
-; GFX6-NEXT: v_mad_f32 v3, -v4, v2, v3
-; GFX6-NEXT: v_cmp_ge_f32_e64 s[6:7], |v3|, |v2|
-; GFX6-NEXT: s_and_b64 s[6:7], s[6:7], exec
-; GFX6-NEXT: v_cvt_i32_f32_e32 v4, v4
+; GFX6-NEXT: v_mul_f32_e32 v2, v1, v2
+; GFX6-NEXT: v_trunc_f32_e32 v2, v2
+; GFX6-NEXT: v_mad_f32 v1, -v2, v0, v1
+; GFX6-NEXT: v_cmp_ge_f32_e64 s[12:13], |v1|, |v0|
+; GFX6-NEXT: s_and_b64 s[12:13], s[12:13], exec
+; GFX6-NEXT: v_cvt_i32_f32_e32 v2, v2
; GFX6-NEXT: s_cselect_b32 s5, s5, 0
; GFX6-NEXT: s_bfe_i32 s4, s4, 0xf000f
-; GFX6-NEXT: v_cvt_f32_i32_e32 v2, s4
-; GFX6-NEXT: v_add_i32_e32 v3, vcc, s5, v4
+; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s4
+; GFX6-NEXT: v_add_i32_e32 v2, vcc, s5, v2
; GFX6-NEXT: s_bfe_i32 s5, s10, 0xf000f
-; GFX6-NEXT: v_cvt_f32_i32_e32 v4, s5
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v5, v2
+; GFX6-NEXT: v_cvt_f32_i32_e32 v1, s5
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v3, v0
; GFX6-NEXT: s_xor_b32 s4, s5, s4
; GFX6-NEXT: s_ashr_i32 s4, s4, 30
-; GFX6-NEXT: v_bfe_i32 v1, v1, 0, 15
-; GFX6-NEXT: v_mul_f32_e32 v5, v4, v5
-; GFX6-NEXT: v_trunc_f32_e32 v5, v5
-; GFX6-NEXT: v_mad_f32 v4, -v5, v2, v4
+; GFX6-NEXT: s_or_b32 s7, s4, 1
+; GFX6-NEXT: v_mul_f32_e32 v3, v1, v3
+; GFX6-NEXT: v_trunc_f32_e32 v3, v3
+; GFX6-NEXT: v_mad_f32 v1, -v3, v0, v1
+; GFX6-NEXT: v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GFX6-NEXT: s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT: v_cvt_i32_f32_e32 v3, v3
+; GFX6-NEXT: s_cselect_b32 s4, s7, 0
+; GFX6-NEXT: s_bfe_i32 s5, s8, 0xf0000
+; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s5
+; GFX6-NEXT: v_add_i32_e32 v3, vcc, s4, v3
+; GFX6-NEXT: s_bfe_i32 s4, s6, 0xf0000
+; GFX6-NEXT: v_cvt_f32_i32_e32 v1, s4
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v4, v0
+; GFX6-NEXT: s_xor_b32 s4, s4, s5
+; GFX6-NEXT: s_ashr_i32 s4, s4, 30
; GFX6-NEXT: s_or_b32 s6, s4, 1
-; GFX6-NEXT: v_cvt_i32_f32_e32 v5, v5
-; GFX6-NEXT: v_cmp_ge_f32_e64 s[4:5], |v4|, |v2|
-; GFX6-NEXT: v_cvt_f32_i32_e32 v2, v1
-; GFX6-NEXT: v_alignbit_b32 v0, s11, v0, 30
+; GFX6-NEXT: v_mul_f32_e32 v4, v1, v4
+; GFX6-NEXT: v_trunc_f32_e32 v4, v4
+; GFX6-NEXT: v_mad_f32 v1, -v4, v0, v1
+; GFX6-NEXT: v_cvt_i32_f32_e32 v4, v4
+; GFX6-NEXT: v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
; GFX6-NEXT: s_and_b64 s[4:5], s[4:5], exec
; GFX6-NEXT: s_cselect_b32 s4, s6, 0
-; GFX6-NEXT: v_bfe_i32 v0, v0, 0, 15
-; GFX6-NEXT: v_add_i32_e32 v4, vcc, s4, v5
-; GFX6-NEXT: v_cvt_f32_i32_e32 v5, v0
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v6, v2
-; GFX6-NEXT: v_xor_b32_e32 v0, v0, v1
-; GFX6-NEXT: v_ashrrev_i32_e32 v0, 30, v0
-; GFX6-NEXT: v_or_b32_e32 v0, 1, v0
-; GFX6-NEXT: v_mul_f32_e32 v1, v5, v6
-; GFX6-NEXT: v_trunc_f32_e32 v1, v1
-; GFX6-NEXT: v_mad_f32 v5, -v1, v2, v5
-; GFX6-NEXT: v_cvt_i32_f32_e32 v1, v1
-; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v5|, |v2|
-; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v3
-; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v1
-; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v4
+; GFX6-NEXT: v_add_i32_e32 v0, vcc, s4, v4
+; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v3
; GFX6-NEXT: v_lshl_b64 v[0:1], v[0:1], 30
+; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v2
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 15, v3
; GFX6-NEXT: v_or_b32_e32 v2, v3, v2
-; GFX6-NEXT: s_mov_b32 s0, s8
-; GFX6-NEXT: s_mov_b32 s1, s9
; GFX6-NEXT: v_or_b32_e32 v0, v2, v0
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_waitcnt expcnt(0)
@@ -5077,60 +5068,59 @@ define amdgpu_kernel void @sdiv_v3i15(ptr addrspace(1) %out, <3 x i15> %x, <3 x
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-NEXT: v_mov_b32_e32 v0, s2
-; GFX9-NEXT: s_bfe_i32 s4, s6, 0xf0000
-; GFX9-NEXT: v_cvt_f32_i32_e32 v3, s4
-; GFX9-NEXT: v_alignbit_b32 v0, s3, v0, 30
-; GFX9-NEXT: s_bfe_i32 s3, s2, 0xf0000
-; GFX9-NEXT: v_cvt_f32_i32_e32 v4, s3
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v5, v3
-; GFX9-NEXT: s_xor_b32 s3, s3, s4
+; GFX9-NEXT: s_lshr_b64 s[4:5], s[2:3], 30
+; GFX9-NEXT: s_bfe_i32 s3, s6, 0xf0000
+; GFX9-NEXT: v_cvt_f32_i32_e32 v0, s3
+; GFX9-NEXT: s_bfe_i32 s5, s2, 0xf0000
+; GFX9-NEXT: v_cvt_f32_i32_e32 v1, s5
+; GFX9-NEXT: s_xor_b32 s3, s5, s3
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v3, v0
; GFX9-NEXT: s_ashr_i32 s3, s3, 30
+; GFX9-NEXT: s_lshr_b64 s[8:9], s[6:7], 30
; GFX9-NEXT: s_or_b32 s3, s3, 1
-; GFX9-NEXT: v_mul_f32_e32 v5, v4, v5
-; GFX9-NEXT: v_trunc_f32_e32 v5, v5
-; GFX9-NEXT: v_mad_f32 v4, -v5, v3, v4
-; GFX9-NEXT: v_cmp_ge_f32_e64 s[4:5], |v4|, |v3|
-; GFX9-NEXT: s_and_b64 s[4:5], s[4:5], exec
+; GFX9-NEXT: v_mul_f32_e32 v3, v1, v3
+; GFX9-NEXT: v_trunc_f32_e32 v3, v3
+; GFX9-NEXT: v_mad_f32 v1, -v3, v0, v1
+; GFX9-NEXT: v_cmp_ge_f32_e64 s[10:11], |v1|, |v0|
+; GFX9-NEXT: s_and_b64 s[10:11], s[10:11], exec
; GFX9-NEXT: s_cselect_b32 s3, s3, 0
-; GFX9-NEXT: s_bfe_i32 s4, s6, 0xf000f
-; GFX9-NEXT: v_cvt_i32_f32_e32 v5, v5
-; GFX9-NEXT: v_cvt_f32_i32_e32 v3, s4
+; GFX9-NEXT: s_bfe_i32 s5, s6, 0xf000f
+; GFX9-NEXT: v_cvt_f32_i32_e32 v0, s5
; GFX9-NEXT: s_bfe_i32 s2, s2, 0xf000f
-; GFX9-NEXT: v_mov_b32_e32 v1, s6
-; GFX9-NEXT: v_add_u32_e32 v4, s3, v5
-; GFX9-NEXT: v_cvt_f32_i32_e32 v5, s2
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v6, v3
-; GFX9-NEXT: v_alignbit_b32 v1, s7, v1, 30
-; GFX9-NEXT: s_xor_b32 s2, s2, s4
+; GFX9-NEXT: v_cvt_f32_i32_e32 v1, s2
+; GFX9-NEXT: v_cvt_i32_f32_e32 v3, v3
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v4, v0
+; GFX9-NEXT: s_xor_b32 s2, s2, s5
+; GFX9-NEXT: s_ashr_i32 s2, s2, 30
+; GFX9-NEXT: v_add_u32_e32 v3, s3, v3
+; GFX9-NEXT: v_mul_f32_e32 v4, v1, v4
+; GFX9-NEXT: v_trunc_f32_e32 v4, v4
+; GFX9-NEXT: v_mad_f32 v1, -v4, v0, v1
+; GFX9-NEXT: s_or_b32 s5, s2, 1
+; GFX9-NEXT: v_cmp_ge_f32_e64 s[2:3], |v1|, |v0|
+; GFX9-NEXT: s_and_b64 s[2:3], s[2:3], exec
+; GFX9-NEXT: v_cvt_i32_f32_e32 v4, v4
+; GFX9-NEXT: s_cselect_b32 s2, s5, 0
+; GFX9-NEXT: s_bfe_i32 s3, s8, 0xf0000
+; GFX9-NEXT: v_cvt_f32_i32_e32 v0, s3
+; GFX9-NEXT: v_add_u32_e32 v4, s2, v4
+; GFX9-NEXT: s_bfe_i32 s2, s4, 0xf0000
+; GFX9-NEXT: v_cvt_f32_i32_e32 v1, s2
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v5, v0
+; GFX9-NEXT: s_xor_b32 s2, s2, s3
; GFX9-NEXT: s_ashr_i32 s2, s2, 30
-; GFX9-NEXT: v_mul_f32_e32 v6, v5, v6
-; GFX9-NEXT: v_trunc_f32_e32 v6, v6
-; GFX9-NEXT: v_mad_f32 v5, -v6, v3, v5
-; GFX9-NEXT: v_bfe_i32 v1, v1, 0, 15
; GFX9-NEXT: s_or_b32 s4, s2, 1
-; GFX9-NEXT: v_cvt_i32_f32_e32 v6, v6
-; GFX9-NEXT: v_cmp_ge_f32_e64 s[2:3], |v5|, |v3|
-; GFX9-NEXT: v_cvt_f32_i32_e32 v3, v1
+; GFX9-NEXT: v_mul_f32_e32 v5, v1, v5
+; GFX9-NEXT: v_trunc_f32_e32 v5, v5
+; GFX9-NEXT: v_mad_f32 v1, -v5, v0, v1
+; GFX9-NEXT: v_cvt_i32_f32_e32 v5, v5
+; GFX9-NEXT: v_cmp_ge_f32_e64 s[2:3], |v1|, |v0|
; GFX9-NEXT: s_and_b64 s[2:3], s[2:3], exec
; GFX9-NEXT: s_cselect_b32 s2, s4, 0
-; GFX9-NEXT: v_bfe_i32 v0, v0, 0, 15
-; GFX9-NEXT: v_add_u32_e32 v5, s2, v6
-; GFX9-NEXT: v_cvt_f32_i32_e32 v6, v0
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v7, v3
-; GFX9-NEXT: v_xor_b32_e32 v0, v0, v1
-; GFX9-NEXT: v_ashrrev_i32_e32 v0, 30, v0
-; GFX9-NEXT: v_or_b32_e32 v0, 1, v0
-; GFX9-NEXT: v_mul_f32_e32 v1, v6, v7
-; GFX9-NEXT: v_trunc_f32_e32 v1, v1
-; GFX9-NEXT: v_cvt_i32_f32_e32 v7, v1
-; GFX9-NEXT: v_mad_f32 v1, -v1, v3, v6
-; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, |v3|
-; GFX9-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX9-NEXT: v_add_u32_e32 v0, v7, v0
-; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v4
-; GFX9-NEXT: v_and_b32_e32 v4, 0x7fff, v5
+; GFX9-NEXT: v_add_u32_e32 v0, s2, v5
+; GFX9-NEXT: v_and_b32_e32 v4, 0x7fff, v4
; GFX9-NEXT: v_lshlrev_b64 v[0:1], 30, v[0:1]
+; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v3
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 15, v4
; GFX9-NEXT: v_or_b32_e32 v3, v3, v4
; GFX9-NEXT: v_or_b32_e32 v0, v3, v0
@@ -5233,74 +5223,73 @@ define amdgpu_kernel void @srem_v3i15(ptr addrspace(1) %out, <3 x i15> %x, <3 x
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
-; GFX6-NEXT: s_bfe_i32 s6, s10, 0xf0000
-; GFX6-NEXT: v_mov_b32_e32 v2, s4
-; GFX6-NEXT: v_alignbit_b32 v2, s5, v2, 30
-; GFX6-NEXT: s_bfe_i32 s5, s4, 0xf0000
-; GFX6-NEXT: v_cvt_f32_i32_e32 v4, s5
-; GFX6-NEXT: v_cvt_f32_i32_e32 v5, s6
-; GFX6-NEXT: s_xor_b32 s5, s6, s5
-; GFX6-NEXT: s_ashr_i32 s5, s5, 30
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v6, v4
; GFX6-NEXT: s_mov_b32 s0, s8
; GFX6-NEXT: s_mov_b32 s1, s9
-; GFX6-NEXT: s_lshr_b32 s8, s10, 15
-; GFX6-NEXT: v_mul_f32_e32 v6, v5, v6
-; GFX6-NEXT: v_trunc_f32_e32 v6, v6
-; GFX6-NEXT: v_mad_f32 v5, -v6, v4, v5
-; GFX6-NEXT: v_cvt_i32_f32_e32 v6, v6
-; GFX6-NEXT: s_lshr_b32 s9, s4, 15
+; GFX6-NEXT: s_lshr_b64 s[8:9], s[4:5], 30
+; GFX6-NEXT: s_bfe_i32 s5, s4, 0xf0000
+; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s5
+; GFX6-NEXT: s_bfe_i32 s12, s10, 0xf0000
+; GFX6-NEXT: v_cvt_f32_i32_e32 v1, s12
+; GFX6-NEXT: s_xor_b32 s5, s12, s5
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v2, v0
+; GFX6-NEXT: s_lshr_b64 s[6:7], s[10:11], 30
+; GFX6-NEXT: s_ashr_i32 s5, s5, 30
+; GFX6-NEXT: s_and_b32 s7, s6, 0x7fff
+; GFX6-NEXT: v_mul_f32_e32 v2, v1, v2
+; GFX6-NEXT: v_trunc_f32_e32 v2, v2
+; GFX6-NEXT: v_mad_f32 v1, -v2, v0, v1
+; GFX6-NEXT: v_cvt_i32_f32_e32 v2, v2
+; GFX6-NEXT: s_lshr_b32 s11, s10, 15
+; GFX6-NEXT: s_and_b32 s9, s8, 0x7fff
+; GFX6-NEXT: s_lshr_b32 s14, s4, 15
; GFX6-NEXT: s_or_b32 s5, s5, 1
-; GFX6-NEXT: v_cmp_ge_f32_e64 s[6:7], |v5|, |v4|
-; GFX6-NEXT: s_and_b64 s[6:7], s[6:7], exec
+; GFX6-NEXT: v_cmp_ge_f32_e64 s[12:13], |v1|, |v0|
+; GFX6-NEXT: s_and_b64 s[12:13], s[12:13], exec
; GFX6-NEXT: s_cselect_b32 s5, s5, 0
-; GFX6-NEXT: v_add_i32_e32 v4, vcc, s5, v6
-; GFX6-NEXT: v_mul_lo_u32 v4, v4, s4
+; GFX6-NEXT: v_add_i32_e32 v0, vcc, s5, v2
+; GFX6-NEXT: v_mul_lo_u32 v0, v0, s4
; GFX6-NEXT: s_bfe_i32 s4, s4, 0xf000f
-; GFX6-NEXT: v_cvt_f32_i32_e32 v5, s4
+; GFX6-NEXT: v_cvt_f32_i32_e32 v1, s4
; GFX6-NEXT: s_bfe_i32 s5, s10, 0xf000f
-; GFX6-NEXT: v_cvt_f32_i32_e32 v6, s5
+; GFX6-NEXT: v_sub_i32_e32 v2, vcc, s10, v0
+; GFX6-NEXT: v_cvt_f32_i32_e32 v0, s5
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v3, v1
; GFX6-NEXT: s_xor_b32 s4, s5, s4
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v7, v5
-; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v2
; GFX6-NEXT: s_ashr_i32 s4, s4, 30
-; GFX6-NEXT: v_bfe_i32 v2, v2, 0, 15
-; GFX6-NEXT: v_mul_f32_e32 v7, v6, v7
-; GFX6-NEXT: v_trunc_f32_e32 v7, v7
-; GFX6-NEXT: v_mad_f32 v6, -v7, v5, v6
+; GFX6-NEXT: s_or_b32 s10, s4, 1
+; GFX6-NEXT: v_mul_f32_e32 v3, v0, v3
+; GFX6-NEXT: v_trunc_f32_e32 v3, v3
+; GFX6-NEXT: v_mad_f32 v0, -v3, v1, v0
+; GFX6-NEXT: v_cvt_i32_f32_e32 v3, v3
+; GFX6-NEXT: v_cmp_ge_f32_e64 s[4:5], |v0|, |v1|
+; GFX6-NEXT: s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT: s_cselect_b32 s4, s10, 0
+; GFX6-NEXT: v_add_i32_e32 v0, vcc, s4, v3
+; GFX6-NEXT: s_bfe_i32 s4, s8, 0xf0000
+; GFX6-NEXT: v_cvt_f32_i32_e32 v1, s4
+; GFX6-NEXT: s_bfe_i32 s5, s6, 0xf0000
+; GFX6-NEXT: v_cvt_f32_i32_e32 v3, s5
+; GFX6-NEXT: s_xor_b32 s4, s5, s4
+; GFX6-NEXT: v_rcp_iflag_f32_e32 v4, v1
+; GFX6-NEXT: s_ashr_i32 s4, s4, 30
; GFX6-NEXT: s_or_b32 s6, s4, 1
-; GFX6-NEXT: v_cvt_i32_f32_e32 v7, v7
-; GFX6-NEXT: v_cmp_ge_f32_e64 s[4:5], |v6|, |v5|
-; GFX6-NEXT: v_cvt_f32_i32_e32 v6, v2
-; GFX6-NEXT: v_mov_b32_e32 v0, s10
-; GFX6-NEXT: v_alignbit_b32 v0, s11, v0, 30
+; GFX6-NEXT: v_mul_lo_u32 v0, v0, s14
+; GFX6-NEXT: v_mul_f32_e32 v4, v3, v4
+; GFX6-NEXT: v_trunc_f32_e32 v4, v4
+; GFX6-NEXT: v_mad_f32 v3, -v4, v1, v3
+; GFX6-NEXT: v_cvt_i32_f32_e32 v4, v4
+; GFX6-NEXT: v_cmp_ge_f32_e64 s[4:5], |v3|, |v1|
; GFX6-NEXT: s_and_b64 s[4:5], s[4:5], exec
-; GFX6-NEXT: v_and_b32_e32 v1, 0x7fff, v0
; GFX6-NEXT: s_cselect_b32 s4, s6, 0
-; GFX6-NEXT: v_bfe_i32 v0, v0, 0, 15
-; GFX6-NEXT: v_add_i32_e32 v5, vcc, s4, v7
-; GFX6-NEXT: v_cvt_f32_i32_e32 v7, v0
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v8, v6
-; GFX6-NEXT: v_xor_b32_e32 v0, v0, v2
-; GFX6-NEXT: v_sub_i32_e32 v4, vcc, s10, v4
-; GFX6-NEXT: v_mul_f32_e32 v2, v7, v8
-; GFX6-NEXT: v_trunc_f32_e32 v2, v2
-; GFX6-NEXT: v_mad_f32 v7, -v2, v6, v7
-; GFX6-NEXT: v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT: v_ashrrev_i32_e32 v0, 30, v0
-; GFX6-NEXT: v_or_b32_e32 v0, 1, v0
-; GFX6-NEXT: v_cmp_ge_f32_e64 vcc, |v7|, |v6|
-; GFX6-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX6-NEXT: v_mul_lo_u32 v5, v5, s9
-; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v2
-; GFX6-NEXT: v_mul_lo_u32 v0, v0, v3
-; GFX6-NEXT: v_sub_i32_e32 v2, vcc, s8, v5
-; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v2
-; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT: v_add_i32_e32 v1, vcc, s4, v4
+; GFX6-NEXT: v_mul_lo_u32 v1, v1, s9
+; GFX6-NEXT: v_sub_i32_e32 v3, vcc, s11, v0
+; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v3
+; GFX6-NEXT: v_sub_i32_e32 v0, vcc, s7, v1
; GFX6-NEXT: v_lshl_b64 v[0:1], v[0:1], 30
-; GFX6-NEXT: v_and_b32_e32 v3, 0x7fff, v4
-; GFX6-NEXT: v_lshlrev_b32_e32 v2, 15, v2
-; GFX6-NEXT: v_or_b32_e32 v2, v2, v3
+; GFX6-NEXT: v_and_b32_e32 v2, 0x7fff, v2
+; GFX6-NEXT: v_lshlrev_b32_e32 v3, 15, v3
+; GFX6-NEXT: v_or_b32_e32 v2, v3, v2
; GFX6-NEXT: v_or_b32_e32 v0, v2, v0
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_waitcnt expcnt(0)
@@ -5312,78 +5301,77 @@ define amdgpu_kernel void @srem_v3i15(ptr addrspace(1) %out, <3 x i15> %x, <3 x
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX9-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x34
-; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-NEXT: v_mov_b32_e32 v0, s2
-; GFX9-NEXT: v_alignbit_b32 v0, s3, v0, 30
-; GFX9-NEXT: s_bfe_i32 s3, s6, 0xf0000
-; GFX9-NEXT: v_cvt_f32_i32_e32 v4, s3
-; GFX9-NEXT: s_bfe_i32 s4, s2, 0xf0000
-; GFX9-NEXT: v_cvt_f32_i32_e32 v5, s4
-; GFX9-NEXT: s_xor_b32 s3, s4, s3
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v6, v4
-; GFX9-NEXT: v_mov_b32_e32 v1, s6
-; GFX9-NEXT: s_ashr_i32 s3, s3, 30
-; GFX9-NEXT: s_lshr_b32 s8, s2, 15
-; GFX9-NEXT: v_mul_f32_e32 v6, v5, v6
-; GFX9-NEXT: v_trunc_f32_e32 v6, v6
-; GFX9-NEXT: v_mad_f32 v5, -v6, v4, v5
-; GFX9-NEXT: v_cvt_i32_f32_e32 v6, v6
-; GFX9-NEXT: v_alignbit_b32 v1, s7, v1, 30
-; GFX9-NEXT: s_lshr_b32 s7, s6, 15
-; GFX9-NEXT: s_or_b32 s3, s3, 1
-; GFX9-NEXT: v_cmp_ge_f32_e64 s[4:5], |v5|, |v4|
-; GFX9-NEXT: s_and_b64 s[4:5], s[4:5], exec
-; GFX9-NEXT: s_cselect_b32 s3, s3, 0
-; GFX9-NEXT: v_add_u32_e32 v4, s3, v6
-; GFX9-NEXT: s_bfe_i32 s3, s6, 0xf000f
-; GFX9-NEXT: v_cvt_f32_i32_e32 v5, s3
-; GFX9-NEXT: s_bfe_i32 s4, s2, 0xf000f
-; GFX9-NEXT: v_cvt_f32_i32_e32 v6, s4
-; GFX9-NEXT: s_xor_b32 s3, s4, s3
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v7, v5
-; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v1
-; GFX9-NEXT: s_ashr_i32 s3, s3, 30
-; GFX9-NEXT: v_bfe_i32 v1, v1, 0, 15
-; GFX9-NEXT: v_mul_f32_e32 v7, v6, v7
-; GFX9-NEXT: v_trunc_f32_e32 v7, v7
-; GFX9-NEXT: v_mad_f32 v6, -v7, v5, v6
-; GFX9-NEXT: v_cvt_i32_f32_e32 v7, v7
-; GFX9-NEXT: s_or_b32 s3, s3, 1
-; GFX9-NEXT: v_cmp_ge_f32_e64 s[4:5], |v6|, |v5|
-; GFX9-NEXT: v_cvt_f32_i32_e32 v6, v1
+; GFX9-NEXT: s_lshr_b64 s[4:5], s[2:3], 30
+; GFX9-NEXT: s_bfe_i32 s5, s6, 0xf0000
+; GFX9-NEXT: v_cvt_f32_i32_e32 v0, s5
+; GFX9-NEXT: s_lshr_b64 s[8:9], s[6:7], 30
+; GFX9-NEXT: s_bfe_i32 s7, s2, 0xf0000
+; GFX9-NEXT: v_cvt_f32_i32_e32 v1, s7
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v2, v0
+; GFX9-NEXT: s_xor_b32 s5, s7, s5
+; GFX9-NEXT: s_ashr_i32 s5, s5, 30
+; GFX9-NEXT: s_lshr_b32 s3, s2, 15
+; GFX9-NEXT: v_mul_f32_e32 v2, v1, v2
+; GFX9-NEXT: v_trunc_f32_e32 v2, v2
+; GFX9-NEXT: v_mad_f32 v1, -v2, v0, v1
+; GFX9-NEXT: v_cvt_i32_f32_e32 v2, v2
+; GFX9-NEXT: s_and_b32 s9, s4, 0x7fff
+; GFX9-NEXT: s_and_b32 s12, s8, 0x7fff
+; GFX9-NEXT: s_lshr_b32 s13, s6, 15
+; GFX9-NEXT: s_or_b32 s5, s5, 1
+; GFX9-NEXT: v_cmp_ge_f32_e64 s[10:11], |v1|, |v0|
+; GFX9-NEXT: s_and_b64 s[10:11], s[10:11], exec
+; GFX9-NEXT: s_cselect_b32 s5, s5, 0
+; GFX9-NEXT: v_add_u32_e32 v0, s5, v2
+; GFX9-NEXT: s_bfe_i32 s5, s6, 0xf000f
+; GFX9-NEXT: v_cvt_f32_i32_e32 v1, s5
+; GFX9-NEXT: v_mul_lo_u32 v0, v0, s6
+; GFX9-NEXT: s_bfe_i32 s6, s2, 0xf000f
+; GFX9-NEXT: v_cvt_f32_i32_e32 v2, s6
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v3, v1
+; GFX9-NEXT: s_xor_b32 s5, s6, s5
+; GFX9-NEXT: s_ashr_i32 s5, s5, 30
+; GFX9-NEXT: s_or_b32 s5, s5, 1
+; GFX9-NEXT: v_mul_f32_e32 v3, v2, v3
+; GFX9-NEXT: v_trunc_f32_e32 v3, v3
+; GFX9-NEXT: v_mad_f32 v2, -v3, v1, v2
+; GFX9-NEXT: v_cvt_i32_f32_e32 v3, v3
+; GFX9-NEXT: v_cmp_ge_f32_e64 s[6:7], |v2|, |v1|
+; GFX9-NEXT: s_and_b64 s[6:7], s[6:7], exec
+; GFX9-NEXT: s_cselect_b32 s5, s5, 0
+; GFX9-NEXT: v_add_u32_e32 v1, s5, v3
+; GFX9-NEXT: s_bfe_i32 s5, s8, 0xf0000
+; GFX9-NEXT: v_cvt_f32_i32_e32 v2, s5
+; GFX9-NEXT: s_bfe_i32 s4, s4, 0xf0000
+; GFX9-NEXT: v_cvt_f32_i32_e32 v3, s4
+; GFX9-NEXT: s_xor_b32 s4, s4, s5
+; GFX9-NEXT: v_rcp_iflag_f32_e32 v4, v2
+; GFX9-NEXT: s_ashr_i32 s4, s4, 30
+; GFX9-NEXT: s_or_b32 s6, s4, 1
+; GFX9-NEXT: v_mul_lo_u32 v1, v1, s13
+; GFX9-NEXT: v_mul_f32_e32 v4, v3, v4
+; GFX9-NEXT: v_trunc_f32_e32 v4, v4
+; GFX9-NEXT: v_mad_f32 v3, -v4, v2, v3
+; GFX9-NEXT: v_cvt_i32_f32_e32 v4, v4
+; GFX9-NEXT: v_cmp_ge_f32_e64 s[4:5], |v3|, |v2|
; GFX9-NEXT: s_and_b64 s[4:5], s[4:5], exec
-; GFX9-NEXT: s_cselect_b32 s3, s3, 0
-; GFX9-NEXT: v_add_u32_e32 v5, s3, v7
-; GFX9-NEXT: v_bfe_i32 v7, v0, 0, 15
-; GFX9-NEXT: v_cvt_f32_i32_e32 v8, v7
-; GFX9-NEXT: v_rcp_iflag_f32_e32 v9, v6
-; GFX9-NEXT: v_xor_b32_e32 v1, v7, v1
-; GFX9-NEXT: v_ashrrev_i32_e32 v1, 30, v1
-; GFX9-NEXT: v_or_b32_e32 v1, 1, v1
-; GFX9-NEXT: v_mul_f32_e32 v7, v8, v9
-; GFX9-NEXT: v_trunc_f32_e32 v7, v7
-; GFX9-NEXT: v_cvt_i32_f32_e32 v9, v7
-; GFX9-NEXT: v_mad_f32 v7, -v7, v6, v8
-; GFX9-NEXT: v_cmp_ge_f32_e64 vcc, |v7|, |v6|
-; GFX9-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc
-; GFX9-NEXT: v_mul_lo_u32 v4, v4, s6
-; GFX9-NEXT: v_mul_lo_u32 v5, v5, s7
-; GFX9-NEXT: v_add_u32_e32 v1, v9, v1
-; GFX9-NEXT: v_mul_lo_u32 v1, v1, v3
-; GFX9-NEXT: v_and_b32_e32 v0, 0x7fff, v0
-; GFX9-NEXT: v_sub_u32_e32 v3, s2, v4
-; GFX9-NEXT: v_sub_u32_e32 v4, s8, v5
-; GFX9-NEXT: v_sub_u32_e32 v0, v0, v1
-; GFX9-NEXT: v_and_b32_e32 v4, 0x7fff, v4
+; GFX9-NEXT: s_cselect_b32 s4, s6, 0
+; GFX9-NEXT: v_add_u32_e32 v2, s4, v4
+; GFX9-NEXT: v_mul_lo_u32 v2, v2, s12
+; GFX9-NEXT: v_sub_u32_e32 v4, s2, v0
+; GFX9-NEXT: v_sub_u32_e32 v5, s3, v1
+; GFX9-NEXT: v_mov_b32_e32 v3, 0
+; GFX9-NEXT: v_sub_u32_e32 v0, s9, v2
+; GFX9-NEXT: v_and_b32_e32 v2, 0x7fff, v4
+; GFX9-NEXT: v_and_b32_e32 v4, 0x7fff, v5
; GFX9-NEXT: v_lshlrev_b64 v[0:1], 30, v[0:1]
-; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff, v3
; GFX9-NEXT: v_lshlrev_b32_e32 v4, 15, v4
-; GFX9-NEXT: v_or_b32_e32 v3, v3, v4
-; GFX9-NEXT: v_or_b32_e32 v0, v3, v0
-; GFX9-NEXT: global_store_dword v2, v0, s[0:1]
+; GFX9-NEXT: v_or_b32_e32 v2, v2, v4
+; GFX9-NEXT: v_or_b32_e32 v0, v2, v0
+; GFX9-NEXT: global_store_dword v3, v0, s[0:1]
; GFX9-NEXT: v_and_b32_e32 v0, 0x1fff, v1
-; GFX9-NEXT: global_store_short v2, v0, s[0:1] offset:4
+; GFX9-NEXT: global_store_short v3, v0, s[0:1] offset:4
; GFX9-NEXT: s_endpgm
%r = srem <3 x i15> %x, %y
store <3 x i15> %r, ptr addrspace(1) %out
diff --git a/llvm/test/CodeGen/AMDGPU/branch-folding-implicit-def-subreg.ll b/llvm/test/CodeGen/AMDGPU/branch-folding-implicit-def-subreg.ll
index 98136347ab702..ff678e66bdaa3 100644
--- a/llvm/test/CodeGen/AMDGPU/branch-folding-implicit-def-subreg.ll
+++ b/llvm/test/CodeGen/AMDGPU/branch-folding-implicit-def-subreg.ll
@@ -17,52 +17,52 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: early-clobber renamable $sgpr20_sgpr21_sgpr22_sgpr23 = S_LOAD_DWORDX4_IMM_ec renamable $sgpr8_sgpr9, 24, 0 :: (dereferenceable invariant load (s128) from %ir.arg6.kernarg.offset.align.down, align 8, addrspace 4)
; GFX90A-NEXT: renamable $sgpr17 = S_LOAD_DWORD_IMM renamable $sgpr8_sgpr9, 40, 0 :: (dereferenceable invariant load (s32) from %ir.arg6.kernarg.offset.align.down + 16, align 8, addrspace 4)
; GFX90A-NEXT: renamable $sgpr24_sgpr25_sgpr26_sgpr27 = S_LOAD_DWORDX4_IMM renamable $sgpr8_sgpr9, 0, 0 :: (dereferenceable invariant load (s128) from %ir.arg.kernarg.offset1, addrspace 4)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_LOAD_DWORDX2_IMM renamable $sgpr8_sgpr9, 16, 0 :: (dereferenceable invariant load (s64) from %ir.arg.kernarg.offset1 + 16, align 16, addrspace 4)
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_LOAD_DWORDX2_IMM renamable $sgpr8_sgpr9, 16, 0 :: (dereferenceable invariant load (s64) from %ir.arg.kernarg.offset1 + 16, align 16, addrspace 4)
; GFX90A-NEXT: S_BITCMP1_B32 renamable $sgpr33, 0, implicit-def $scc
; GFX90A-NEXT: renamable $sgpr12_sgpr13 = S_CSELECT_B64 -1, 0, implicit killed $scc
- ; GFX90A-NEXT: renamable $sgpr34_sgpr35 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_MOV_B64 -1
; GFX90A-NEXT: renamable $sgpr28_sgpr29 = S_XOR_B64 renamable $sgpr12_sgpr13, -1, implicit-def dead $scc
; GFX90A-NEXT: S_BITCMP1_B32 renamable $sgpr33, 8, implicit-def $scc
- ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_CSELECT_B64 -1, 0, implicit killed $scc
- ; GFX90A-NEXT: renamable $sgpr30_sgpr31 = S_XOR_B64 killed renamable $sgpr18_sgpr19, -1, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr30_sgpr31 = S_CSELECT_B64 -1, 0, implicit killed $scc
+ ; GFX90A-NEXT: renamable $sgpr30_sgpr31 = S_XOR_B64 killed renamable $sgpr30_sgpr31, -1, implicit-def dead $scc
; GFX90A-NEXT: renamable $vgpr3 = V_MOV_B32_e32 0, implicit $exec
; GFX90A-NEXT: renamable $vgpr2 = DS_READ_B32_gfx9 renamable $vgpr3, 0, 0, implicit $exec :: (load (s32) from `ptr addrspace(3) null`, align 8, addrspace 3)
- ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, renamable $sgpr28_sgpr29, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_VCCZ %bb.2, implicit $vcc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.1.bb103:
- ; GFX90A-NEXT: successors: %bb.58(0x40000000), %bb.2(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr33, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x00000000000000FF, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000FF, $vgpr2_vgpr3:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: successors: %bb.59(0x40000000), %bb.2(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr33, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr44_sgpr45, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x00000000000000FF, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000FF, $vgpr2_vgpr3:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr34_sgpr35 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, renamable $sgpr30_sgpr31, implicit-def dead $scc
- ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.58, implicit $vcc
+ ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.59, implicit $vcc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.2:
; GFX90A-NEXT: successors: %bb.3(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8, $sgpr9, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr56, $sgpr57, $sgpr20_sgpr21_sgpr22, $sgpr24_sgpr25_sgpr26, $sgpr26_sgpr27, $vgpr2, $vgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8, $sgpr9, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr44_sgpr45, $sgpr58, $sgpr59, $sgpr20_sgpr21_sgpr22, $sgpr24_sgpr25_sgpr26, $sgpr26_sgpr27, $vgpr2, $vgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr23 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
- ; GFX90A-NEXT: renamable $vgpr19 = IMPLICIT_DEF implicit-def $vgpr18
- ; GFX90A-NEXT: renamable $vgpr21 = IMPLICIT_DEF implicit-def $vgpr20
+ ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
; GFX90A-NEXT: renamable $vgpr23 = IMPLICIT_DEF implicit-def $vgpr22
; GFX90A-NEXT: renamable $vgpr25 = IMPLICIT_DEF implicit-def $vgpr24
+ ; GFX90A-NEXT: renamable $vgpr27 = IMPLICIT_DEF implicit-def $vgpr26
+ ; GFX90A-NEXT: renamable $vgpr29 = IMPLICIT_DEF implicit-def $vgpr28
; GFX90A-NEXT: renamable $sgpr36_sgpr37 = S_MOV_B64 0
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.3.Flow17:
- ; GFX90A-NEXT: successors: %bb.4(0x40000000), %bb.57(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr23, $sgpr33, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000FF, $vgpr2_vgpr3:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr18_vgpr19:0x000000000000000F, $vgpr20_vgpr21:0x000000000000000F, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: successors: %bb.4(0x40000000), %bb.58(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr23, $sgpr33, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr44_sgpr45, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000FF, $vgpr2_vgpr3:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $vgpr26_vgpr27:0x000000000000000F, $vgpr28_vgpr29:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr30 = V_AND_B32_e32 1023, $vgpr31, implicit $exec
- ; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, killed renamable $sgpr34_sgpr35, implicit-def dead $scc
- ; GFX90A-NEXT: S_CBRANCH_VCCZ %bb.57, implicit $vcc
+ ; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, killed renamable $sgpr18_sgpr19, implicit-def dead $scc
+ ; GFX90A-NEXT: S_CBRANCH_VCCZ %bb.58, implicit $vcc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.4.bb15:
; GFX90A-NEXT: successors: %bb.35(0x40000000), %bb.5(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000FF, $vgpr2_vgpr3:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr18_sgpr19
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000FF, $vgpr2_vgpr3:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr44_sgpr45
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr0_vgpr1 = V_LSHLREV_B64_e64 2, $vgpr2_vgpr3, implicit $exec
; GFX90A-NEXT: renamable $vgpr4 = COPY renamable $sgpr25, implicit $exec
@@ -76,16 +76,16 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.5:
; GFX90A-NEXT: successors: %bb.6(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr18_sgpr19
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr44_sgpr45
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $sgpr34_sgpr35 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_MOV_B64 0
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr40_sgpr41 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr38_sgpr39 = S_MOV_B64 0
@@ -99,100 +99,102 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: renamable $vgpr56_vgpr57 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr44_vgpr45 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr42_vgpr43 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.6.Flow20:
; GFX90A-NEXT: successors: %bb.7(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $vgpr19 = COPY renamable $sgpr17, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr18 = COPY $sgpr17, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr21 = COPY $sgpr17, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr20 = COPY $sgpr17, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr23 = COPY $sgpr17, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr23 = COPY renamable $sgpr17, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr13 = COPY killed renamable $sgpr18, implicit $exec
; GFX90A-NEXT: renamable $vgpr22 = COPY $sgpr17, implicit $exec
; GFX90A-NEXT: renamable $vgpr25 = COPY $sgpr17, implicit $exec
; GFX90A-NEXT: renamable $vgpr24 = COPY $sgpr17, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr27 = COPY $sgpr17, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr26 = COPY $sgpr17, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr29 = COPY $sgpr17, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr28 = COPY $sgpr17, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.7.Flow19:
- ; GFX90A-NEXT: successors: %bb.62(0x40000000), %bb.8(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr18_vgpr19:0x000000000000000F, $vgpr20_vgpr21:0x000000000000000F, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: successors: %bb.63(0x40000000), %bb.8(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000F, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $vgpr26_vgpr27:0x000000000000000F, $vgpr28_vgpr29:0x000000000000000F, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_MOV_B64 0
- ; GFX90A-NEXT: $sgpr24_sgpr25 = S_AND_SAVEEXEC_B64 $sgpr36_sgpr37, implicit-def $exec, implicit-def $scc, implicit $exec
- ; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.62, implicit $exec
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_MOV_B64 0
+ ; GFX90A-NEXT: $sgpr18_sgpr19 = S_AND_SAVEEXEC_B64 $sgpr36_sgpr37, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.63, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.8.Flow32:
; GFX90A-NEXT: successors: %bb.9(0x40000000), %bb.10(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr58_sgpr59, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr24_sgpr25, implicit-def $scc
- ; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $sgpr18_sgpr19, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr18_sgpr19, implicit-def $scc
+ ; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $sgpr44_sgpr45, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr12_sgpr13 = S_XOR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.10, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.9.bb89:
; GFX90A-NEXT: successors: %bb.10(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr58_sgpr59, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr9, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr8, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.10.Flow33:
; GFX90A-NEXT: successors: %bb.11(0x40000000), %bb.12(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr58_sgpr59, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def $scc
- ; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $sgpr58_sgpr59, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $sgpr60_sgpr61, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr12_sgpr13 = S_XOR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.12, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.11.bb84:
; GFX90A-NEXT: successors: %bb.12(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr7, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr6, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.12.Flow34:
; GFX90A-NEXT: successors: %bb.13(0x40000000), %bb.14(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def $scc
- ; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $sgpr54_sgpr55, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $sgpr56_sgpr57, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr12_sgpr13 = S_XOR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.14, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.13.bb79:
; GFX90A-NEXT: successors: %bb.14(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr56_sgpr57, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr5, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr4, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.14.Flow35:
; GFX90A-NEXT: successors: %bb.15(0x40000000), %bb.16(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr56_sgpr57, $vgpr0_vgpr1:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def $scc
- ; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $sgpr52_sgpr53, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $sgpr54_sgpr55, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr36_sgpr37 = S_XOR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.16, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.15.bb72:
; GFX90A-NEXT: successors: %bb.16(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr56_sgpr57, $vgpr0_vgpr1:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $sgpr8 = S_ADD_U32 renamable $sgpr8, 48, implicit-def $scc
; GFX90A-NEXT: renamable $sgpr9 = S_ADDC_U32 killed renamable $sgpr9, 0, implicit-def dead $scc, implicit killed $scc
@@ -202,79 +204,79 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: $sgpr13 = COPY killed renamable $sgpr15
; GFX90A-NEXT: $sgpr14 = COPY killed renamable $sgpr16
; GFX90A-NEXT: dead $sgpr30_sgpr31 = SI_CALL killed renamable $sgpr18_sgpr19, @f2, csr_amdgpu_gfx90ainsts, implicit $sgpr4_sgpr5, implicit $sgpr6_sgpr7, implicit $sgpr8_sgpr9, implicit $sgpr10_sgpr11, implicit $sgpr12, implicit $sgpr13, implicit $sgpr14, implicit undef $sgpr15, implicit $vgpr31, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.16.Flow36:
; GFX90A-NEXT: successors: %bb.17(0x40000000), %bb.18(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr36_sgpr37, implicit-def $scc
- ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr50_sgpr51, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr52_sgpr53, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr4_sgpr5 = S_XOR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.18, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.17.bb67:
; GFX90A-NEXT: successors: %bb.18(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr47, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr46, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.18.Flow37:
; GFX90A-NEXT: successors: %bb.19(0x40000000), %bb.20(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def $scc
- ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr48_sgpr49, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr50_sgpr51, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr4_sgpr5 = S_XOR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.20, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.19.bb62:
; GFX90A-NEXT: successors: %bb.20(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr63, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr62, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.20.Flow38:
; GFX90A-NEXT: successors: %bb.21(0x40000000), %bb.22(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def $scc
- ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr46_sgpr47, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr48_sgpr49, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr4_sgpr5 = S_XOR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.22, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.21.bb54:
; GFX90A-NEXT: successors: %bb.22(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr61, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr60, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.22.Flow39:
; GFX90A-NEXT: successors: %bb.23(0x40000000), %bb.24(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def $scc
- ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr44_sgpr45, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr46_sgpr47, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr4_sgpr5 = S_XOR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.24, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.23.bb47:
; GFX90A-NEXT: successors: %bb.24(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr59, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr58, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.24.Flow40:
; GFX90A-NEXT: successors: %bb.25(0x40000000), %bb.26(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def $scc
; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr42_sgpr43, implicit-def $exec, implicit-def $scc, implicit $exec
@@ -283,15 +285,15 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.25.bb40:
; GFX90A-NEXT: successors: %bb.26(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr57, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr56, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.26.Flow41:
; GFX90A-NEXT: successors: %bb.27(0x40000000), %bb.28(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def $scc
; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr40_sgpr41, implicit-def $exec, implicit-def $scc, implicit $exec
@@ -300,15 +302,15 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.27.bb33:
; GFX90A-NEXT: successors: %bb.28(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr45, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr44, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.28.Flow42:
; GFX90A-NEXT: successors: %bb.34(0x40000000), %bb.29(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def $scc
; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr38_sgpr39, implicit-def $exec, implicit-def $scc, implicit $exec
@@ -317,7 +319,7 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.29.Flow43:
; GFX90A-NEXT: successors: %bb.30(0x40000000), %bb.31(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr4_sgpr5, implicit-def $scc
; GFX90A-NEXT: $vcc = S_ANDN2_B64 $exec, killed renamable $sgpr34_sgpr35, implicit-def dead $scc
@@ -325,17 +327,17 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.30.bb19:
; GFX90A-NEXT: successors: %bb.31(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr41, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.31.Flow44:
; GFX90A-NEXT: successors: %bb.32(0x40000000), %bb.33(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr56_sgpr57, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr58_sgpr59, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr56_sgpr57, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr4_sgpr5 = S_AND_SAVEEXEC_B64 $sgpr58_sgpr59, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: S_CBRANCH_EXECZ %bb.33, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.32.UnifiedUnreachableBlock:
@@ -351,28 +353,28 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.34.bb26:
; GFX90A-NEXT: successors: %bb.29(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr56_sgpr57, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr4_sgpr5, $sgpr34_sgpr35, $sgpr58_sgpr59, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET renamable $vgpr43, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr42, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_OR_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_OR_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
; GFX90A-NEXT: S_BRANCH %bb.29
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.35.bb20:
; GFX90A-NEXT: successors: %bb.37(0x40000000), %bb.36(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr18_sgpr19
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr44_sgpr45
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr0 = GLOBAL_LOAD_SBYTE renamable $vgpr40_vgpr41, 1024, 0, implicit $exec :: (load (s8) from %ir.i21, addrspace 1)
; GFX90A-NEXT: renamable $vgpr42 = V_ADD_CO_U32_e32 1024, $vgpr40, implicit-def $vcc, implicit $exec
; GFX90A-NEXT: renamable $sgpr34_sgpr35 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr38_sgpr39 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_MOV_B64 0
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vgpr43, dead renamable $vcc = V_ADDC_U32_e64 0, $vgpr41, killed $vcc, 0, implicit $exec
; GFX90A-NEXT: renamable $vcc = V_CMP_LT_I16_e64 0, killed $vgpr0, implicit $exec
@@ -386,35 +388,37 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: renamable $vgpr58_vgpr59 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr56_vgpr57 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr44_vgpr45 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: $sgpr24_sgpr25 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.37, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.36.Flow21:
; GFX90A-NEXT: successors: %bb.6(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr24_sgpr25, implicit-def $scc
; GFX90A-NEXT: S_BRANCH %bb.6
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.37.bb27:
; GFX90A-NEXT: successors: %bb.39(0x40000000), %bb.38(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr18_sgpr19, $sgpr58_sgpr59, $sgpr54_sgpr55, $sgpr52_sgpr53, $sgpr50_sgpr51, $sgpr48_sgpr49, $sgpr46_sgpr47, $sgpr44_sgpr45, $sgpr42_sgpr43
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr44_sgpr45, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr54_sgpr55, $sgpr52_sgpr53, $sgpr50_sgpr51, $sgpr48_sgpr49, $sgpr46_sgpr47, $sgpr42_sgpr43
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr0 = GLOBAL_LOAD_UBYTE renamable $vgpr40_vgpr41, 2048, 0, implicit $exec :: (load (s8) from %ir.i28, addrspace 1)
; GFX90A-NEXT: renamable $vgpr44 = V_ADD_CO_U32_e32 2048, $vgpr40, implicit-def $vcc, implicit $exec
; GFX90A-NEXT: renamable $sgpr40_sgpr41 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = COPY renamable $sgpr36_sgpr37
+ ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = COPY renamable $sgpr36_sgpr37
; GFX90A-NEXT: renamable $vgpr45, dead renamable $vcc = V_ADDC_U32_e64 0, $vgpr41, killed $vcc, 0, implicit $exec
; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U16_e64 0, killed $vgpr0, implicit $exec
+ ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr6_vgpr7 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr4_vgpr5 = IMPLICIT_DEF
@@ -423,49 +427,51 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: renamable $vgpr60_vgpr61 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr58_vgpr59 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr56_vgpr57 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: $sgpr38_sgpr39 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.39, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.38.Flow22:
; GFX90A-NEXT: successors: %bb.36(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr62_sgpr63, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr38_sgpr39, implicit-def $scc
; GFX90A-NEXT: renamable $sgpr38_sgpr39 = S_XOR_B64 $exec, -1, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr40_sgpr41 = S_AND_B64 killed renamable $sgpr40_sgpr41, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_AND_B64 killed renamable $sgpr42_sgpr43, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_AND_B64 killed renamable $sgpr44_sgpr45, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_AND_B64 killed renamable $sgpr64_sgpr65, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_AND_B64 killed renamable $sgpr46_sgpr47, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_AND_B64 killed renamable $sgpr48_sgpr49, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_AND_B64 killed renamable $sgpr50_sgpr51, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_AND_B64 killed renamable $sgpr52_sgpr53, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_AND_B64 killed renamable $sgpr54_sgpr55, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_AND_B64 killed renamable $sgpr18_sgpr19, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_AND_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_AND_B64 killed renamable $sgpr60_sgpr61, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_AND_B64 killed renamable $sgpr44_sgpr45, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr36_sgpr37 = S_ANDN2_B64 killed renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_AND_B64 killed renamable $sgpr60_sgpr61, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr36_sgpr37 = S_OR_B64 killed renamable $sgpr36_sgpr37, killed renamable $sgpr56_sgpr57, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr62_sgpr63, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr36_sgpr37 = S_OR_B64 killed renamable $sgpr36_sgpr37, killed renamable $sgpr58_sgpr59, implicit-def dead $scc
; GFX90A-NEXT: S_BRANCH %bb.36
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.39.bb34:
; GFX90A-NEXT: successors: %bb.41(0x40000000), %bb.40(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr18_sgpr19, $sgpr58_sgpr59, $sgpr54_sgpr55, $sgpr52_sgpr53, $sgpr50_sgpr51, $sgpr48_sgpr49, $sgpr46_sgpr47, $sgpr44_sgpr45
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr42_sgpr43, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr54_sgpr55, $sgpr52_sgpr53, $sgpr50_sgpr51, $sgpr48_sgpr49, $sgpr46_sgpr47, $sgpr64_sgpr65
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr0 = GLOBAL_LOAD_UBYTE renamable $vgpr40_vgpr41, 3072, 0, implicit $exec :: (load (s8) from %ir.i35, addrspace 1)
; GFX90A-NEXT: renamable $vgpr56 = V_ADD_CO_U32_e32 3072, $vgpr40, implicit-def $vcc, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = COPY renamable $sgpr36_sgpr37
+ ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = COPY renamable $sgpr36_sgpr37
; GFX90A-NEXT: renamable $vgpr57, dead renamable $vcc = V_ADDC_U32_e64 0, $vgpr41, killed $vcc, 0, implicit $exec
; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U16_e64 0, killed $vgpr0, implicit $exec
+ ; GFX90A-NEXT: renamable $sgpr66_sgpr67 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr6_vgpr7 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr4_vgpr5 = IMPLICIT_DEF
@@ -473,328 +479,341 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: renamable $vgpr62_vgpr63 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr60_vgpr61 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr58_vgpr59 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: $sgpr40_sgpr41 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.41, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.40.Flow23:
; GFX90A-NEXT: successors: %bb.38(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr62_sgpr63, $sgpr66_sgpr67, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr40_sgpr41, implicit-def $scc
; GFX90A-NEXT: renamable $sgpr40_sgpr41 = S_XOR_B64 $exec, -1, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_AND_B64 killed renamable $sgpr42_sgpr43, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_AND_B64 killed renamable $sgpr44_sgpr45, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_AND_B64 killed renamable $sgpr46_sgpr47, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = S_AND_B64 killed renamable $sgpr44_sgpr45, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_AND_B64 killed renamable $sgpr66_sgpr67, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_AND_B64 killed renamable $sgpr48_sgpr49, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_AND_B64 killed renamable $sgpr50_sgpr51, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_AND_B64 killed renamable $sgpr52_sgpr53, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_AND_B64 killed renamable $sgpr54_sgpr55, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_AND_B64 killed renamable $sgpr18_sgpr19, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_AND_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_AND_B64 killed renamable $sgpr60_sgpr61, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_OR_B64 killed renamable $sgpr56_sgpr57, killed renamable $sgpr60_sgpr61, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_AND_B64 killed renamable $sgpr42_sgpr43, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr62_sgpr63, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = S_OR_B64 killed renamable $sgpr42_sgpr43, killed renamable $sgpr58_sgpr59, implicit-def dead $scc
; GFX90A-NEXT: S_BRANCH %bb.38
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.41.bb41:
; GFX90A-NEXT: successors: %bb.46(0x40000000), %bb.42(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr58_sgpr59, $sgpr54_sgpr55, $sgpr52_sgpr53, $sgpr50_sgpr51, $sgpr48_sgpr49, $sgpr46_sgpr47
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr46_sgpr47, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr54_sgpr55, $sgpr52_sgpr53, $sgpr50_sgpr51, $sgpr64_sgpr65
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr58 = V_ADD_CO_U32_e32 4096, $vgpr40, implicit-def $vcc, implicit $exec
; GFX90A-NEXT: renamable $sgpr18_sgpr19 = COPY $vcc
; GFX90A-NEXT: renamable $vgpr59, dead renamable $sgpr18_sgpr19 = V_ADDC_U32_e64 0, $vgpr41, killed $sgpr18_sgpr19, 0, implicit $exec
; GFX90A-NEXT: renamable $vgpr0 = GLOBAL_LOAD_UBYTE renamable $vgpr58_vgpr59, 0, 0, implicit $exec :: (load (s8) from %ir.i42, addrspace 1)
- ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_MOV_B64 0
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = COPY renamable $sgpr36_sgpr37
- ; GFX90A-NEXT: renamable $vgpr18, dead renamable $vcc = V_ADDC_U32_e64 0, $vgpr41, killed $vcc, 0, implicit $exec
+ ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = COPY renamable $sgpr36_sgpr37
+ ; GFX90A-NEXT: renamable $vgpr13, dead renamable $vcc = V_ADDC_U32_e64 0, $vgpr41, killed $vcc, 0, implicit $exec
; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U16_e64 0, killed $vgpr0, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr6_vgpr7 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr4_vgpr5 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr0_vgpr1 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr62_vgpr63 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr60_vgpr61 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: $sgpr42_sgpr43 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.46, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.42.Flow24:
; GFX90A-NEXT: successors: %bb.40(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr18, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr60_sgpr61, $sgpr62_sgpr63, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr13, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr62_sgpr63, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr42_sgpr43, implicit-def $scc
- ; GFX90A-NEXT: renamable $vgpr59 = COPY killed renamable $vgpr18, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_XOR_B64 $exec, -1, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_AND_B64 killed renamable $sgpr44_sgpr45, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_AND_B64 killed renamable $sgpr62_sgpr63, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_AND_B64 killed renamable $sgpr48_sgpr49, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $vgpr59 = COPY killed renamable $vgpr13, implicit $exec
+ ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_XOR_B64 $exec, -1, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr66_sgpr67 = S_AND_B64 killed renamable $sgpr48_sgpr49, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_AND_B64 killed renamable $sgpr64_sgpr65, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_AND_B64 killed renamable $sgpr50_sgpr51, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_AND_B64 killed renamable $sgpr52_sgpr53, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_AND_B64 killed renamable $sgpr54_sgpr55, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_AND_B64 killed renamable $sgpr18_sgpr19, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_AND_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_AND_B64 killed renamable $sgpr60_sgpr61, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_OR_B64 killed renamable $sgpr56_sgpr57, killed renamable $sgpr60_sgpr61, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_AND_B64 killed renamable $sgpr46_sgpr47, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr62_sgpr63, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = S_OR_B64 killed renamable $sgpr46_sgpr47, killed renamable $sgpr58_sgpr59, implicit-def dead $scc
; GFX90A-NEXT: S_BRANCH %bb.40
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.43.bb55:
; GFX90A-NEXT: successors: %bb.48(0x40000000), %bb.44(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr44_sgpr45, $sgpr52_sgpr53, $sgpr58_sgpr59, $sgpr54_sgpr55, $sgpr46_sgpr47
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr54_sgpr55, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr50_sgpr51, $sgpr48_sgpr49
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: S_BITCMP1_B32 killed renamable $sgpr33, 16, implicit-def $scc
- ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = S_CSELECT_B64 -1, 0, implicit killed $scc
- ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_XOR_B64 renamable $sgpr64_sgpr65, -1, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr66_sgpr67 = S_CSELECT_B64 -1, 0, implicit killed $scc
+ ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_XOR_B64 renamable $sgpr66_sgpr67, -1, implicit-def dead $scc
; GFX90A-NEXT: renamable $vgpr62 = V_ADD_CO_U32_e32 6144, $vgpr40, implicit-def $vcc, implicit $exec
; GFX90A-NEXT: renamable $vgpr63, dead renamable $vcc = V_ADDC_U32_e64 0, $vgpr41, killed $vcc, 0, implicit $exec
- ; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, renamable $sgpr48_sgpr49, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, renamable $sgpr18_sgpr19, implicit-def dead $scc
; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.48, implicit $vcc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.44:
; GFX90A-NEXT: successors: %bb.45(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr56, $vgpr47, $vgpr18, $vgpr30, $vgpr31, $vgpr58, $vgpr61, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8, $sgpr9, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $vgpr57, $vgpr63, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22, $sgpr22_sgpr23, $sgpr24_sgpr25_sgpr26, $sgpr26_sgpr27, $vgpr46, $vgpr45, $vgpr2, $vgpr3, $vgpr44, $vgpr43, $vgpr42, $vgpr41, $vgpr40, $vgpr60, $vgpr62
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr54_sgpr55, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr50_sgpr51
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = COPY renamable $sgpr36_sgpr37
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = COPY renamable $sgpr36_sgpr37
; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr6_vgpr7 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr4_vgpr5 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr0_vgpr1 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_MOV_B64 0
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.45.Flow26:
; GFX90A-NEXT: successors: %bb.47(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr18, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr13, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_XOR_B64 $exec, -1, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr70_sgpr71 = S_AND_B64 killed renamable $sgpr44_sgpr45, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr68_sgpr69 = S_AND_B64 killed renamable $sgpr46_sgpr47, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr66_sgpr67 = S_AND_B64 killed renamable $sgpr48_sgpr49, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_AND_B64 killed renamable $sgpr54_sgpr55, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_AND_B64 killed renamable $sgpr52_sgpr53, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_AND_B64 killed renamable $sgpr50_sgpr51, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = S_OR_B64 killed renamable $sgpr44_sgpr45, killed renamable $sgpr48_sgpr49, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = S_XOR_B64 $exec, -1, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr72_sgpr73 = S_AND_B64 killed renamable $sgpr46_sgpr47, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr70_sgpr71 = S_AND_B64 killed renamable $sgpr48_sgpr49, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr68_sgpr69 = S_AND_B64 killed renamable $sgpr50_sgpr51, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_AND_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_AND_B64 killed renamable $sgpr60_sgpr61, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_AND_B64 killed renamable $sgpr54_sgpr55, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_AND_B64 killed renamable $sgpr52_sgpr53, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr66_sgpr67 = S_OR_B64 killed renamable $sgpr48_sgpr49, killed renamable $sgpr50_sgpr51, implicit-def dead $scc
; GFX90A-NEXT: S_BRANCH %bb.47
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.46.bb48:
; GFX90A-NEXT: successors: %bb.43(0x40000000), %bb.47(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr46_sgpr47, $sgpr58_sgpr59, $sgpr54_sgpr55, $sgpr44_sgpr45, $sgpr52_sgpr53
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr33, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr46_sgpr47, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr54_sgpr55, $sgpr50_sgpr51, $sgpr48_sgpr49
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr60 = V_ADD_CO_U32_e32 5120, $vgpr40, implicit-def $vcc, implicit $exec
; GFX90A-NEXT: renamable $sgpr18_sgpr19 = COPY $vcc
; GFX90A-NEXT: renamable $vgpr0 = V_ADD_CO_U32_e32 4096, $vgpr40, implicit-def $vcc, implicit $exec
; GFX90A-NEXT: renamable $vgpr1, dead renamable $vcc = V_ADDC_U32_e64 0, $vgpr41, killed $vcc, 0, implicit $exec
; GFX90A-NEXT: renamable $vgpr0 = GLOBAL_LOAD_UBYTE killed renamable $vgpr0_vgpr1, 1024, 0, implicit $exec :: (load (s8) from %ir.i49, addrspace 1)
- ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = COPY renamable $sgpr36_sgpr37
- ; GFX90A-NEXT: renamable $sgpr66_sgpr67 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr66_sgpr67 = COPY renamable $sgpr36_sgpr37
; GFX90A-NEXT: renamable $sgpr68_sgpr69 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr70_sgpr71 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vgpr61, dead renamable $vcc = V_ADDC_U32_e64 0, $vgpr41, killed $sgpr18_sgpr19, 0, implicit $exec
; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U16_e64 0, killed $vgpr0, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr70_sgpr71 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr72_sgpr73 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr6_vgpr7 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr4_vgpr5 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr0_vgpr1 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr62_vgpr63 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
- ; GFX90A-NEXT: $sgpr18_sgpr19 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: $sgpr44_sgpr45 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.43, implicit $exec
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.47.Flow25:
; GFX90A-NEXT: successors: %bb.42(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr18, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr46_sgpr47, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr60_sgpr61, $sgpr64_sgpr65, $sgpr66_sgpr67, $sgpr68_sgpr69, $sgpr70_sgpr71, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
- ; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr18_sgpr19, implicit-def $scc
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_XOR_B64 $exec, -1, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = S_AND_B64 killed renamable $sgpr60_sgpr61, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_AND_B64 killed renamable $sgpr70_sgpr71, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_AND_B64 killed renamable $sgpr68_sgpr69, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_AND_B64 killed renamable $sgpr66_sgpr67, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_AND_B64 killed renamable $sgpr54_sgpr55, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr58_sgpr59, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_AND_B64 killed renamable $sgpr46_sgpr47, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_AND_B64 killed renamable $sgpr64_sgpr65, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_OR_B64 killed renamable $sgpr46_sgpr47, killed renamable $sgpr56_sgpr57, implicit-def dead $scc
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr13, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr62_sgpr63, $sgpr66_sgpr67, $sgpr68_sgpr69, $sgpr70_sgpr71, $sgpr72_sgpr73, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: {{ $}}
+ ; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr44_sgpr45, implicit-def $scc
+ ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_XOR_B64 $exec, -1, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = S_AND_B64 killed renamable $sgpr62_sgpr63, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_AND_B64 killed renamable $sgpr72_sgpr73, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_AND_B64 killed renamable $sgpr70_sgpr71, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_AND_B64 killed renamable $sgpr68_sgpr69, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_AND_B64 killed renamable $sgpr56_sgpr57, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_AND_B64 killed renamable $sgpr60_sgpr61, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_AND_B64 killed renamable $sgpr46_sgpr47, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr66_sgpr67, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = S_OR_B64 killed renamable $sgpr44_sgpr45, killed renamable $sgpr58_sgpr59, implicit-def dead $scc
; GFX90A-NEXT: S_BRANCH %bb.42
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.48.bb63:
- ; GFX90A-NEXT: successors: %bb.50(0x40000000), %bb.49(0x40000000)
- ; GFX90A-NEXT: liveins: $vcc, $sgpr14, $sgpr15, $sgpr16, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr48_sgpr49, $sgpr56_sgpr57:0x000000000000000F, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr52_sgpr53, $sgpr58_sgpr59, $sgpr54_sgpr55, $sgpr46_sgpr47
+ ; GFX90A-NEXT: successors: %bb.51(0x40000000), %bb.49(0x40000000)
+ ; GFX90A-NEXT: liveins: $vcc, $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr58_sgpr59:0x000000000000000F, $sgpr66_sgpr67, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr48_sgpr49, $sgpr54_sgpr55, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr50_sgpr51
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_MOV_B64 0
- ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.50, implicit $vcc
+ ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_MOV_B64 0
+ ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.51, implicit $vcc
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: bb.49:
- ; GFX90A-NEXT: successors: %bb.44(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr52_sgpr53, $sgpr58_sgpr59, $sgpr54_sgpr55
+ ; GFX90A-NEXT: successors: %bb.50(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr48_sgpr49, $sgpr54_sgpr55, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr50_sgpr51
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_MOV_B64 -1
- ; GFX90A-NEXT: S_BRANCH %bb.44
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = COPY renamable $sgpr36_sgpr37
+ ; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr6_vgpr7 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr4_vgpr5 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr0_vgpr1 = IMPLICIT_DEF
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.50.bb68:
- ; GFX90A-NEXT: successors: %bb.54(0x40000000), %bb.51(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr48_sgpr49, $sgpr56_sgpr57:0x000000000000000F, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr46_sgpr47, $sgpr52_sgpr53, $sgpr58_sgpr59, $sgpr54_sgpr55
+ ; GFX90A-NEXT: bb.50.Flow26:
+ ; GFX90A-NEXT: successors: %bb.45(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr58, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8, $sgpr9, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22, $sgpr22_sgpr23, $sgpr24_sgpr25_sgpr26, $sgpr26_sgpr27, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr60, $vgpr61, $vgpr63, $vgpr62, $vgpr40, $vgpr41, $vgpr42, $vgpr43, $vgpr44, $vgpr45, $vgpr46, $vgpr47, $vgpr56, $vgpr57
+ ; GFX90A-NEXT: {{ $}}
+ ; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
+ ; GFX90A-NEXT: S_BRANCH %bb.45
+ ; GFX90A-NEXT: {{ $}}
+ ; GFX90A-NEXT: bb.51.bb68:
+ ; GFX90A-NEXT: successors: %bb.55(0x40000000), %bb.52(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr58_sgpr59:0x000000000000000F, $sgpr66_sgpr67, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr54_sgpr55, $sgpr60_sgpr61, $sgpr56_sgpr57, $sgpr50_sgpr51
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr0 = nuw nsw V_LSHLREV_B32_e32 3, $vgpr30, implicit $exec
; GFX90A-NEXT: renamable $vgpr1 = V_MOV_B32_e32 0, implicit $exec
- ; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, killed renamable $sgpr48_sgpr49, implicit-def dead $scc
- ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.54, implicit $vcc
+ ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, killed renamable $sgpr18_sgpr19, implicit-def dead $scc
+ ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.55, implicit $vcc
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.51:
- ; GFX90A-NEXT: successors: %bb.45(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr52_sgpr53, $sgpr58_sgpr59, $sgpr54_sgpr55
+ ; GFX90A-NEXT: bb.52:
+ ; GFX90A-NEXT: successors: %bb.50(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr54_sgpr55, $sgpr60_sgpr61, $sgpr56_sgpr57
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = COPY renamable $sgpr36_sgpr37
+ ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = COPY renamable $sgpr36_sgpr37
; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr6_vgpr7 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr4_vgpr5 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
- ; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
- ; GFX90A-NEXT: S_BRANCH %bb.45
+ ; GFX90A-NEXT: S_BRANCH %bb.50
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.52.bb80:
- ; GFX90A-NEXT: successors: %bb.59(0x40000000), %bb.53(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr56_sgpr57:0x000000000000000F, $sgpr60_sgpr61, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.53.bb80:
+ ; GFX90A-NEXT: successors: %bb.60(0x40000000), %bb.54(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr58_sgpr59:0x000000000000000F, $sgpr62_sgpr63, $sgpr66_sgpr67, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $sgpr17 = S_BFE_U32 renamable $sgpr20, 65560, implicit-def dead $scc
; GFX90A-NEXT: S_CMP_EQ_U32 killed renamable $sgpr17, 0, implicit-def $scc
; GFX90A-NEXT: renamable $vgpr6 = V_ADD_CO_U32_e32 4096, $vgpr0, implicit-def $vcc, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr7, dead renamable $sgpr50_sgpr51 = V_ADDC_U32_e64 0, 0, killed $vcc, 0, implicit $exec
- ; GFX90A-NEXT: S_CBRANCH_SCC1 %bb.59, implicit killed $scc
+ ; GFX90A-NEXT: renamable $vgpr7, dead renamable $sgpr18_sgpr19 = V_ADDC_U32_e64 0, 0, killed $vcc, 0, implicit $exec
+ ; GFX90A-NEXT: S_CBRANCH_SCC1 %bb.60, implicit killed $scc
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.53:
- ; GFX90A-NEXT: successors: %bb.61(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.54:
+ ; GFX90A-NEXT: successors: %bb.62(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr62_sgpr63, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_MOV_B64 0
- ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = COPY renamable $sgpr36_sgpr37
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = COPY renamable $sgpr36_sgpr37
; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
- ; GFX90A-NEXT: S_BRANCH %bb.61
+ ; GFX90A-NEXT: S_BRANCH %bb.62
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.54.bb73:
- ; GFX90A-NEXT: successors: %bb.52(0x40000000), %bb.55(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr56_sgpr57:0x000000000000000F, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr52_sgpr53
+ ; GFX90A-NEXT: bb.55.bb73:
+ ; GFX90A-NEXT: successors: %bb.53(0x40000000), %bb.56(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr58_sgpr59:0x000000000000000F, $sgpr66_sgpr67, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003F, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr50_sgpr51, $sgpr54_sgpr55, $sgpr60_sgpr61
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr6 = GLOBAL_LOAD_UBYTE renamable $vgpr0_vgpr1, 2048, 0, implicit $exec :: (load (s8) from %ir.i74, addrspace 1)
; GFX90A-NEXT: renamable $vgpr4 = V_ADD_CO_U32_e32 2048, $vgpr0, implicit-def $vcc, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_MOV_B64 0
- ; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_MOV_B64 -1
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = COPY renamable $sgpr36_sgpr37
- ; GFX90A-NEXT: renamable $vgpr5, dead renamable $sgpr58_sgpr59 = V_ADDC_U32_e64 0, 0, killed $vcc, 0, implicit $exec
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = COPY renamable $sgpr36_sgpr37
+ ; GFX90A-NEXT: renamable $vgpr5, dead renamable $sgpr18_sgpr19 = V_ADDC_U32_e64 0, 0, killed $vcc, 0, implicit $exec
; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U16_e64 0, killed $vgpr6, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_MOV_B64 0
; GFX90A-NEXT: renamable $vgpr8_vgpr9 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr6_vgpr7 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $sgpr17 = IMPLICIT_DEF
- ; GFX90A-NEXT: $sgpr60_sgpr61 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
- ; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.52, implicit $exec
+ ; GFX90A-NEXT: $sgpr62_sgpr63 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.53, implicit $exec
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.55.Flow29:
+ ; GFX90A-NEXT: bb.56.Flow29:
; GFX90A-NEXT: successors: %bb.45(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr18, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr13, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr62_sgpr63, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr60_sgpr61, implicit-def $scc
+ ; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr62_sgpr63, implicit-def $scc
; GFX90A-NEXT: S_BRANCH %bb.45
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.56.bb90:
- ; GFX90A-NEXT: successors: %bb.60(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr52_sgpr53, $sgpr56_sgpr57:0x000000000000000F, $sgpr60_sgpr61, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.57.bb90:
+ ; GFX90A-NEXT: successors: %bb.61(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr54_sgpr55, $sgpr58_sgpr59:0x000000000000000F, $sgpr62_sgpr63, $sgpr66_sgpr67, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $vgpr53 = V_CNDMASK_B32_e64 0, 0, 0, 1, killed $sgpr64_sgpr65, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr14 = V_CNDMASK_B32_e64 0, 0, 0, 1, killed $sgpr66_sgpr67, implicit $exec
; GFX90A-NEXT: renamable $vgpr10 = V_MOV_B32_e32 0, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr14_vgpr15 = DS_READ_B64_gfx9 killed renamable $vgpr10, 0, 0, implicit $exec :: (load (s64) from `ptr addrspace(3) null`, addrspace 3)
- ; GFX90A-NEXT: renamable $vgpr10 = COPY renamable $sgpr21, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = DS_READ_B64_gfx9 killed renamable $vgpr10, 0, 0, implicit $exec :: (load (s64) from %ir.7, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr11 = COPY renamable $sgpr21, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr20_vgpr21 = DS_READ_B64_gfx9 killed renamable $vgpr11, 0, 0, implicit $exec :: (load (s64) from %ir.7, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr18_vgpr19 = DS_READ_B64_gfx9 killed renamable $vgpr10, 0, 0, implicit $exec :: (load (s64) from `ptr addrspace(3) null`, addrspace 3)
; GFX90A-NEXT: renamable $vgpr10 = COPY renamable $sgpr22, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr12_vgpr13 = DS_READ_B64_gfx9 killed renamable $vgpr10, 0, 0, implicit $exec :: (load (s64) from %ir.8, addrspace 3)
- ; GFX90A-NEXT: renamable $vgpr10 = COPY renamable $sgpr56, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr11 = V_ALIGNBIT_B32_e64 killed $sgpr57, killed $vgpr10, 1, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr52 = V_ALIGNBIT_B32_e64 $vgpr17, $vgpr16, 1, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr17 = V_CNDMASK_B32_e64 0, 0, 0, 1, $sgpr12_sgpr13, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr15 = V_ALIGNBIT_B32_e64 $vgpr15, $vgpr14, 1, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_XOR_B64 $exec, -1, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = S_OR_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $vgpr10 = COPY renamable $vgpr14, implicit $exec
- ; GFX90A-NEXT: S_BRANCH %bb.60
- ; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.57:
+ ; GFX90A-NEXT: renamable $vgpr16_vgpr17 = DS_READ_B64_gfx9 killed renamable $vgpr10, 0, 0, implicit $exec :: (load (s64) from %ir.8, addrspace 3)
+ ; GFX90A-NEXT: renamable $sgpr18_sgpr19 = S_LSHR_B64 killed renamable $sgpr58_sgpr59, 1, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $vgpr32_vgpr33 = V_LSHRREV_B64_e64 1, $vgpr20_vgpr21, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr15 = V_CNDMASK_B32_e64 0, 0, 0, 1, $sgpr12_sgpr13, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr10_vgpr11 = V_LSHRREV_B64_e64 1, $vgpr18_vgpr19, implicit $exec
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_XOR_B64 $exec, -1, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = S_OR_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $vgpr12 = COPY renamable $vgpr18, implicit $exec
+ ; GFX90A-NEXT: S_BRANCH %bb.61
+ ; GFX90A-NEXT: {{ $}}
+ ; GFX90A-NEXT: bb.58:
; GFX90A-NEXT: successors: %bb.7(0x80000000)
- ; GFX90A-NEXT: liveins: $exec, $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr23, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr18_vgpr19:0x000000000000000F, $vgpr20_vgpr21:0x000000000000000F, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $exec, $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr23, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr36_sgpr37, $sgpr44_sgpr45, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr2_vgpr3:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000F, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $vgpr26_vgpr27:0x000000000000000F, $vgpr28_vgpr29:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $vgpr15 = COPY killed renamable $sgpr23, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr17 = COPY killed renamable $sgpr17, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $vgpr15 = COPY killed renamable $sgpr17, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr10 = COPY killed renamable $sgpr23, implicit $exec
+ ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr48_sgpr49 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr46_sgpr47 = S_MOV_B64 0
- ; GFX90A-NEXT: renamable $sgpr44_sgpr45 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr42_sgpr43 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr40_sgpr41 = S_MOV_B64 0
; GFX90A-NEXT: renamable $sgpr38_sgpr39 = S_MOV_B64 0
@@ -810,201 +829,204 @@ define amdgpu_kernel void @f1(ptr addrspace(1) %arg, ptr addrspace(1) %arg1, i64
; GFX90A-NEXT: renamable $vgpr42_vgpr43 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr40_vgpr41 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr46_vgpr47 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = COPY renamable $vgpr10, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr32 = COPY renamable $vgpr10, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr20 = COPY renamable $vgpr10, implicit $exec
; GFX90A-NEXT: renamable $vgpr14 = COPY renamable $vgpr15, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr52 = COPY renamable $vgpr15, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr16 = COPY renamable $vgpr15, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr53 = COPY renamable $vgpr17, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr13 = COPY renamable $vgpr15, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr12 = COPY renamable $vgpr15, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr17 = COPY renamable $vgpr10, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr16 = COPY renamable $vgpr10, implicit $exec
; GFX90A-NEXT: renamable $sgpr34_sgpr35 = S_MOV_B64 0
; GFX90A-NEXT: S_BRANCH %bb.7
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.58.bb105:
+ ; GFX90A-NEXT: bb.59.bb105:
; GFX90A-NEXT: successors: %bb.3(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr33, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr56_sgpr57:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x00000000000000FF, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000FF, $vgpr2_vgpr3:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $sgpr33, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr44_sgpr45, $sgpr58_sgpr59:0x000000000000000F, $sgpr20_sgpr21_sgpr22_sgpr23:0x00000000000000FF, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000FF, $vgpr2_vgpr3:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr0 = V_MOV_B32_e32 0, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr22_vgpr23 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from `ptr addrspace(3) null`, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr26_vgpr27 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from `ptr addrspace(3) null`, addrspace 3)
; GFX90A-NEXT: renamable $vgpr0 = COPY renamable $sgpr23, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr20_vgpr21 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from %ir.434, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr24_vgpr25 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from %ir.434, addrspace 3)
; GFX90A-NEXT: renamable $vgpr0 = COPY renamable $sgpr21, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr18_vgpr19 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from %ir.7, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr22_vgpr23 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from %ir.7, addrspace 3)
; GFX90A-NEXT: renamable $vgpr0 = COPY killed renamable $sgpr17, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr10_vgpr11 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from %ir.435, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr12_vgpr13 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from %ir.435, addrspace 3)
; GFX90A-NEXT: renamable $vgpr0 = COPY renamable $sgpr22, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr24_vgpr25 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from %ir.8, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr28_vgpr29 = DS_READ_B64_gfx9 killed renamable $vgpr0, 0, 0, implicit $exec :: (load (s64) from %ir.8, addrspace 3)
; GFX90A-NEXT: renamable $sgpr36_sgpr37 = S_MOV_B64 -1
; GFX90A-NEXT: renamable $sgpr23 = S_MOV_B32 0
; GFX90A-NEXT: renamable $sgpr17 = S_MOV_B32 0
; GFX90A-NEXT: S_BRANCH %bb.3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.59.bb85:
- ; GFX90A-NEXT: successors: %bb.56(0x40000000), %bb.60(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr18, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr56_sgpr57:0x000000000000000F, $sgpr60_sgpr61, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.60.bb85:
+ ; GFX90A-NEXT: successors: %bb.57(0x40000000), %bb.61(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr13, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr58_sgpr59:0x000000000000000F, $sgpr62_sgpr63, $sgpr66_sgpr67, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr8 = V_OR_B32_e32 1, $vgpr6, implicit $exec
; GFX90A-NEXT: renamable $vgpr9 = COPY renamable $vgpr7, implicit $exec
; GFX90A-NEXT: renamable $vgpr10 = FLAT_LOAD_UBYTE renamable $vgpr8_vgpr9, 0, 0, implicit $exec, implicit $flat_scr :: (load (s8) from %ir.i86)
; GFX90A-NEXT: renamable $sgpr17 = S_MOV_B32 0
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_MOV_B64 -1
; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U16_e64 0, killed $vgpr10, implicit $exec
- ; GFX90A-NEXT: renamable $sgpr62_sgpr63 = COPY renamable $sgpr36_sgpr37
- ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $sgpr64_sgpr65 = COPY renamable $sgpr36_sgpr37
; GFX90A-NEXT: renamable $vgpr15 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr10 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr32 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr20 = IMPLICIT_DEF
; GFX90A-NEXT: renamable $vgpr14 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr52 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr16 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr53 = IMPLICIT_DEF
- ; GFX90A-NEXT: renamable $vgpr13 = IMPLICIT_DEF implicit-def $vgpr12
- ; GFX90A-NEXT: renamable $vgpr11 = IMPLICIT_DEF implicit-def $vgpr10
- ; GFX90A-NEXT: $sgpr52_sgpr53 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
- ; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.56, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr17 = IMPLICIT_DEF implicit-def $vgpr16
+ ; GFX90A-NEXT: renamable $sgpr18 = IMPLICIT_DEF
+ ; GFX90A-NEXT: renamable $vgpr12 = IMPLICIT_DEF
+ ; GFX90A-NEXT: $sgpr54_sgpr55 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
+ ; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.57, implicit $exec
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.60.Flow31:
- ; GFX90A-NEXT: successors: %bb.61(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr18, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr60_sgpr61, $sgpr62_sgpr63, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.61.Flow31:
+ ; GFX90A-NEXT: successors: %bb.62(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr13, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr62_sgpr63, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr52_sgpr53, implicit-def $scc
- ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_MOV_B64 0
+ ; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr54_sgpr55, implicit-def $scc
+ ; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_MOV_B64 0
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.61.Flow30:
- ; GFX90A-NEXT: successors: %bb.55(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr15, $vgpr17, $vgpr18, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr60_sgpr61, $sgpr62_sgpr63, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.62.Flow30:
+ ; GFX90A-NEXT: successors: %bb.56(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $sgpr17, $vgpr13, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19:0x0000000000000003, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr62_sgpr63, $sgpr64_sgpr65, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x0000000000000003, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x0000000000000003, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_XOR_B64 $exec, -1, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr52_sgpr53, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_AND_B64 killed renamable $sgpr50_sgpr51, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_AND_B64 killed renamable $sgpr62_sgpr63, $exec, implicit-def dead $scc
- ; GFX90A-NEXT: renamable $sgpr50_sgpr51 = S_OR_B64 killed renamable $sgpr50_sgpr51, killed renamable $sgpr56_sgpr57, implicit-def dead $scc
- ; GFX90A-NEXT: S_BRANCH %bb.55
+ ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = S_XOR_B64 $exec, -1, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr60_sgpr61 = S_AND_B64 killed renamable $sgpr54_sgpr55, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr54_sgpr55 = S_AND_B64 killed renamable $sgpr52_sgpr53, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_ANDN2_B64 renamable $sgpr36_sgpr37, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = S_AND_B64 killed renamable $sgpr64_sgpr65, $exec, implicit-def dead $scc
+ ; GFX90A-NEXT: renamable $sgpr52_sgpr53 = S_OR_B64 killed renamable $sgpr52_sgpr53, killed renamable $sgpr58_sgpr59, implicit-def dead $scc
+ ; GFX90A-NEXT: S_BRANCH %bb.56
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.62.bb140:
- ; GFX90A-NEXT: successors: %bb.68(0x40000000), %bb.63(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr18_vgpr19:0x000000000000000F, $vgpr20_vgpr21:0x000000000000000F, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.63.bb140:
+ ; GFX90A-NEXT: successors: %bb.69(0x40000000), %bb.64(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr30_sgpr31, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000F, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $vgpr26_vgpr27:0x000000000000000F, $vgpr28_vgpr29:0x000000000000000F, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr36_sgpr37 = S_MOV_B64 -1
+ ; GFX90A-NEXT: renamable $sgpr24_sgpr25 = S_MOV_B64 -1
; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, killed renamable $sgpr30_sgpr31, implicit-def dead $scc
- ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.68, implicit $vcc
+ ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.69, implicit $vcc
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.63.Flow13:
- ; GFX90A-NEXT: successors: %bb.64(0x40000000), %bb.66(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr36_sgpr37, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000C, $vgpr12_vgpr13:0x000000000000000C, $vgpr18_vgpr19:0x000000000000000C, $vgpr20_vgpr21:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.64.Flow13:
+ ; GFX90A-NEXT: successors: %bb.65(0x40000000), %bb.67(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000C, $vgpr16_vgpr17:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x000000000000000C, $vgpr28_vgpr29:0x000000000000000C, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: $vcc = S_ANDN2_B64 $exec, killed renamable $sgpr36_sgpr37, implicit-def dead $scc
- ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.66, implicit $vcc
+ ; GFX90A-NEXT: $vcc = S_ANDN2_B64 $exec, killed renamable $sgpr24_sgpr25, implicit-def dead $scc
+ ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.67, implicit $vcc
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.64.bb159:
- ; GFX90A-NEXT: successors: %bb.67(0x40000000), %bb.65(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000C, $vgpr12_vgpr13:0x000000000000000C, $vgpr18_vgpr19:0x000000000000000C, $vgpr20_vgpr21:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.65.bb159:
+ ; GFX90A-NEXT: successors: %bb.68(0x40000000), %bb.66(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000C, $vgpr16_vgpr17:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x000000000000000C, $vgpr28_vgpr29:0x000000000000000C, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vcc = V_CMP_NE_U32_e64 0, killed $vgpr30, implicit $exec
; GFX90A-NEXT: $sgpr12_sgpr13 = S_AND_SAVEEXEC_B64 $vcc, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: renamable $sgpr12_sgpr13 = S_XOR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def dead $scc
- ; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.67, implicit $exec
+ ; GFX90A-NEXT: S_CBRANCH_EXECNZ %bb.68, implicit $exec
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.65.Flow10:
- ; GFX90A-NEXT: successors: %bb.66(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.66.Flow10:
+ ; GFX90A-NEXT: successors: %bb.67(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $sgpr12_sgpr13 = S_ANDN2_SAVEEXEC_B64 $sgpr12_sgpr13, implicit-def $exec, implicit-def $scc, implicit $exec
; GFX90A-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def $scc
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.66.Flow14:
+ ; GFX90A-NEXT: bb.67.Flow14:
; GFX90A-NEXT: successors: %bb.8(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr56_sgpr57 = COPY $exec
+ ; GFX90A-NEXT: renamable $sgpr58_sgpr59 = COPY $exec
; GFX90A-NEXT: S_BRANCH %bb.8
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.67.bb161:
- ; GFX90A-NEXT: successors: %bb.65(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000C, $vgpr12_vgpr13:0x000000000000000C, $vgpr18_vgpr19:0x000000000000000C, $vgpr20_vgpr21:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.68.bb161:
+ ; GFX90A-NEXT: successors: %bb.66(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000C, $vgpr16_vgpr17:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x000000000000000C, $vgpr28_vgpr29:0x000000000000000C, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr21, killed $vgpr23, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr2, killed $vgpr25, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr3 = V_OR_B32_e32 killed $vgpr11, killed $vgpr19, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr25, killed $vgpr27, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr2, killed $vgpr29, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr3 = V_OR_B32_e32 killed $vgpr13, killed $vgpr23, implicit $exec
; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr3, killed $vgpr2, implicit $exec
; GFX90A-NEXT: renamable $vgpr3 = V_MOV_B32_e32 0, implicit $exec
- ; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U32_sdwa 0, killed $vgpr53, 0, $vgpr3, 0, 0, 6, implicit $exec
+ ; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U32_sdwa 0, killed $vgpr14, 0, $vgpr3, 0, 0, 6, implicit $exec
; GFX90A-NEXT: renamable $vgpr2 = V_CNDMASK_B32_e64 0, 0, 0, killed $vgpr2, killed $vcc, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr10 = V_OR_B32_e32 killed $vgpr52, killed $vgpr13, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr10, killed $vgpr2, implicit $exec
- ; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U32_sdwa 0, killed $vgpr17, 0, $vgpr3, 0, 0, 6, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr11 = V_OR_B32_e32 killed $vgpr32, killed $vgpr17, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr11, killed $vgpr2, implicit $exec
+ ; GFX90A-NEXT: renamable $vcc = V_CMP_EQ_U32_sdwa 0, killed $vgpr15, 0, $vgpr3, 0, 0, 6, implicit $exec
; GFX90A-NEXT: renamable $vgpr2 = V_CNDMASK_B32_e64 0, 0, 0, killed $vgpr2, killed $vcc, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr2, killed $vgpr15, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr2 = V_OR_B32_e32 killed $vgpr2, killed $vgpr10, implicit $exec
; GFX90A-NEXT: DS_WRITE2_B32_gfx9 killed renamable $vgpr3, killed renamable $vgpr2, renamable $vgpr3, 0, 1, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, align 4, addrspace 3)
- ; GFX90A-NEXT: S_BRANCH %bb.65
- ; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.68.bb174:
- ; GFX90A-NEXT: successors: %bb.72(0x40000000), %bb.69(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr28_sgpr29, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000F, $vgpr12_vgpr13:0x000000000000000F, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr18_vgpr19:0x000000000000000F, $vgpr20_vgpr21:0x000000000000000F, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
- ; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $vgpr26 = V_OR_B32_e32 1, $vgpr24, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr48 = V_OR_B32_e32 $vgpr26, $vgpr22, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr34 = V_OR_B32_e32 $vgpr48, $vgpr20, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr28 = V_CNDMASK_B32_e64 0, $vgpr34, 0, 0, $sgpr12_sgpr13, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr38 = V_OR_B32_e32 $vgpr28, $vgpr18, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr36 = V_OR_B32_e32 $vgpr38, $vgpr10, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr32 = V_OR_B32_e32 $vgpr36, $vgpr12, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr50 = V_CNDMASK_B32_e64 0, 0, 0, $vgpr32, killed $sgpr12_sgpr13, implicit $exec
+ ; GFX90A-NEXT: S_BRANCH %bb.66
+ ; GFX90A-NEXT: {{ $}}
+ ; GFX90A-NEXT: bb.69.bb174:
+ ; GFX90A-NEXT: successors: %bb.73(0x40000000), %bb.70(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr28_sgpr29, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000F, $vgpr16_vgpr17:0x000000000000000F, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr22_vgpr23:0x000000000000000F, $vgpr24_vgpr25:0x000000000000000F, $vgpr26_vgpr27:0x000000000000000F, $vgpr28_vgpr29:0x000000000000000F, $vgpr32_vgpr33:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: {{ $}}
+ ; GFX90A-NEXT: renamable $agpr0 = COPY killed renamable $vgpr32, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr32 = V_OR_B32_e32 1, $vgpr28, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr52 = V_OR_B32_e32 $vgpr32, $vgpr26, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr38 = V_OR_B32_e32 $vgpr52, $vgpr24, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr34 = V_CNDMASK_B32_e64 0, $vgpr38, 0, 0, $sgpr12_sgpr13, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr50 = V_OR_B32_e32 $vgpr34, $vgpr22, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr48 = V_OR_B32_e32 $vgpr50, $vgpr12, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr36 = V_OR_B32_e32 $vgpr48, $vgpr16, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr54 = V_CNDMASK_B32_e64 0, 0, 0, $vgpr36, killed $sgpr12_sgpr13, implicit $exec
; GFX90A-NEXT: renamable $sgpr12_sgpr13 = S_MOV_B64 -1
; GFX90A-NEXT: renamable $vcc = S_AND_B64 $exec, killed renamable $sgpr28_sgpr29, implicit-def dead $scc
- ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.72, implicit $vcc
+ ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.73, implicit $vcc
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.69.Flow:
- ; GFX90A-NEXT: successors: %bb.70(0x40000000), %bb.71(0x40000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000C, $vgpr12_vgpr13:0x000000000000000C, $vgpr18_vgpr19:0x000000000000000C, $vgpr20_vgpr21:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x0000000000000003, $vgpr28_vgpr29:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr34_vgpr35:0x0000000000000003, $vgpr36_vgpr37:0x0000000000000003, $vgpr38_vgpr39:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr48_vgpr49:0x0000000000000003, $vgpr50_vgpr51:0x0000000000000003, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.70.Flow:
+ ; GFX90A-NEXT: successors: %bb.71(0x40000000), %bb.72(0x40000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $agpr0_agpr1:0x0000000000000003, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr12_sgpr13, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000C, $vgpr16_vgpr17:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x000000000000000C, $vgpr28_vgpr29:0x000000000000000C, $vgpr32_vgpr33:0x0000000000000003, $vgpr34_vgpr35:0x0000000000000003, $vgpr36_vgpr37:0x0000000000000003, $vgpr38_vgpr39:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr48_vgpr49:0x0000000000000003, $vgpr50_vgpr51:0x0000000000000003, $vgpr52_vgpr53:0x0000000000000003, $vgpr54_vgpr55:0x0000000000000003, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: $vcc = S_ANDN2_B64 $exec, killed renamable $sgpr12_sgpr13, implicit-def dead $scc
- ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.71, implicit $vcc
+ ; GFX90A-NEXT: S_CBRANCH_VCCNZ %bb.72, implicit $vcc
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.70.bb186:
- ; GFX90A-NEXT: successors: %bb.71(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000C, $vgpr12_vgpr13:0x000000000000000C, $vgpr18_vgpr19:0x000000000000000C, $vgpr20_vgpr21:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x0000000000000003, $vgpr28_vgpr29:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr34_vgpr35:0x0000000000000003, $vgpr36_vgpr37:0x0000000000000003, $vgpr38_vgpr39:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr48_vgpr49:0x0000000000000003, $vgpr50_vgpr51:0x0000000000000003, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.71.bb186:
+ ; GFX90A-NEXT: successors: %bb.72(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $agpr0_agpr1:0x0000000000000003, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000C, $vgpr16_vgpr17:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x000000000000000C, $vgpr28_vgpr29:0x000000000000000C, $vgpr32_vgpr33:0x0000000000000003, $vgpr34_vgpr35:0x0000000000000003, $vgpr36_vgpr37:0x0000000000000003, $vgpr38_vgpr39:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr48_vgpr49:0x0000000000000003, $vgpr50_vgpr51:0x0000000000000003, $vgpr52_vgpr53:0x0000000000000003, $vgpr54_vgpr55:0x0000000000000003, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
; GFX90A-NEXT: renamable $vgpr2_vgpr3 = V_LSHLREV_B64_e64 3, killed $vgpr2_vgpr3, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr10 = COPY renamable $sgpr27, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr11 = COPY renamable $sgpr27, implicit $exec
; GFX90A-NEXT: renamable $vgpr2, renamable $vcc = V_ADD_CO_U32_e64 killed $sgpr26, $vgpr2, 0, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr3, dead renamable $vcc = V_ADDC_U32_e64 killed $vgpr10, killed $vgpr3, killed $vcc, 0, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr27 = V_MOV_B32_e32 0, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr49 = COPY renamable $vgpr27, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr35 = COPY renamable $vgpr27, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr39 = COPY renamable $vgpr27, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr37 = COPY renamable $vgpr27, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr29 = COPY renamable $vgpr27, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr51 = COPY renamable $vgpr27, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr33 = COPY renamable $vgpr27, implicit $exec
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr27, renamable $vgpr26_vgpr27, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
- ; GFX90A-NEXT: renamable $vgpr10 = COPY renamable $sgpr21, implicit $exec
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr10, killed renamable $vgpr48_vgpr49, 0, 0, implicit $exec :: (store (s64) into %ir.7, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr3, dead renamable $vcc = V_ADDC_U32_e64 killed $vgpr11, killed $vgpr3, killed $vcc, 0, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr33 = V_MOV_B32_e32 0, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr53 = COPY renamable $vgpr33, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr39 = COPY renamable $vgpr33, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr51 = COPY renamable $vgpr33, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr49 = COPY renamable $vgpr33, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr35 = COPY renamable $vgpr33, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr55 = COPY renamable $vgpr33, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr37 = COPY renamable $vgpr33, implicit $exec
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr33, renamable $vgpr32_vgpr33, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr11 = COPY renamable $sgpr21, implicit $exec
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr11, killed renamable $vgpr52_vgpr53, 0, 0, implicit $exec :: (store (s64) into %ir.7, addrspace 3)
; GFX90A-NEXT: renamable $vgpr12 = COPY killed renamable $sgpr22, implicit $exec
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 killed renamable $vgpr12, killed renamable $vgpr34_vgpr35, 0, 0, implicit $exec :: (store (s64) into %ir.8, addrspace 3)
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr27, killed renamable $vgpr38_vgpr39, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr10, killed renamable $vgpr36_vgpr37, 0, 0, implicit $exec :: (store (s64) into %ir.7, addrspace 3)
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr27, killed renamable $vgpr28_vgpr29, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 killed renamable $vgpr10, killed renamable $vgpr50_vgpr51, 0, 0, implicit $exec :: (store (s64) into %ir.7, addrspace 3)
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 killed renamable $vgpr27, killed renamable $vgpr32_vgpr33, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 killed renamable $vgpr12, killed renamable $vgpr38_vgpr39, 0, 0, implicit $exec :: (store (s64) into %ir.8, addrspace 3)
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr33, killed renamable $vgpr50_vgpr51, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr11, killed renamable $vgpr48_vgpr49, 0, 0, implicit $exec :: (store (s64) into %ir.7, addrspace 3)
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 renamable $vgpr33, killed renamable $vgpr34_vgpr35, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 killed renamable $vgpr11, killed renamable $vgpr54_vgpr55, 0, 0, implicit $exec :: (store (s64) into %ir.7, addrspace 3)
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 killed renamable $vgpr33, killed renamable $vgpr36_vgpr37, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr3, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null` + 4, basealign 8, addrspace 5)
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed renamable $vgpr2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32) into `ptr addrspace(5) null`, align 8, addrspace 5)
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.71.Flow9:
- ; GFX90A-NEXT: successors: %bb.63(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000C, $vgpr12_vgpr13:0x000000000000000C, $vgpr18_vgpr19:0x000000000000000C, $vgpr20_vgpr21:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.72.Flow9:
+ ; GFX90A-NEXT: successors: %bb.64(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $agpr0_agpr1:0x0000000000000003, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $vgpr0_vgpr1:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000C, $vgpr16_vgpr17:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x000000000000000C, $vgpr28_vgpr29:0x000000000000000C, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $sgpr36_sgpr37 = S_MOV_B64 0
- ; GFX90A-NEXT: S_BRANCH %bb.63
+ ; GFX90A-NEXT: renamable $sgpr24_sgpr25 = S_MOV_B64 0
+ ; GFX90A-NEXT: renamable $vgpr32 = COPY killed renamable $agpr0, implicit $exec
+ ; GFX90A-NEXT: S_BRANCH %bb.64
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: bb.72.bb196:
- ; GFX90A-NEXT: successors: %bb.69(0x80000000)
- ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr15, $vgpr17, $vgpr30, $vgpr31, $vgpr52, $vgpr53, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr24_sgpr25, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr58_sgpr59, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x000000000000000C, $vgpr12_vgpr13:0x000000000000000C, $vgpr14_vgpr15:0x0000000000000003, $vgpr16_vgpr17:0x0000000000000003, $vgpr18_vgpr19:0x000000000000000C, $vgpr20_vgpr21:0x000000000000000C, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x0000000000000003, $vgpr28_vgpr29:0x0000000000000003, $vgpr32_vgpr33:0x0000000000000003, $vgpr34_vgpr35:0x0000000000000003, $vgpr36_vgpr37:0x0000000000000003, $vgpr38_vgpr39:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr48_vgpr49:0x0000000000000003, $vgpr50_vgpr51:0x0000000000000003, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
+ ; GFX90A-NEXT: bb.73.bb196:
+ ; GFX90A-NEXT: successors: %bb.70(0x80000000)
+ ; GFX90A-NEXT: liveins: $sgpr14, $sgpr15, $sgpr16, $vgpr14, $vgpr15, $vgpr30, $vgpr31, $agpr0_agpr1:0x0000000000000003, $sgpr4_sgpr5, $sgpr6_sgpr7, $sgpr8_sgpr9:0x000000000000000F, $sgpr10_sgpr11, $sgpr18_sgpr19, $sgpr34_sgpr35, $sgpr38_sgpr39, $sgpr40_sgpr41, $sgpr42_sgpr43, $sgpr44_sgpr45, $sgpr46_sgpr47, $sgpr48_sgpr49, $sgpr50_sgpr51, $sgpr52_sgpr53, $sgpr54_sgpr55, $sgpr56_sgpr57, $sgpr60_sgpr61, $sgpr20_sgpr21_sgpr22_sgpr23:0x000000000000003C, $sgpr24_sgpr25_sgpr26_sgpr27:0x00000000000000F0, $vgpr0_vgpr1:0x000000000000000F, $vgpr2_vgpr3:0x000000000000000F, $vgpr4_vgpr5:0x000000000000000F, $vgpr6_vgpr7:0x000000000000000F, $vgpr8_vgpr9:0x000000000000000F, $vgpr10_vgpr11:0x0000000000000003, $vgpr12_vgpr13:0x000000000000000C, $vgpr16_vgpr17:0x000000000000000C, $vgpr18_vgpr19:0x0000000000000003, $vgpr20_vgpr21:0x0000000000000003, $vgpr22_vgpr23:0x000000000000000C, $vgpr24_vgpr25:0x000000000000000C, $vgpr26_vgpr27:0x000000000000000C, $vgpr28_vgpr29:0x000000000000000C, $vgpr32_vgpr33:0x0000000000000003, $vgpr34_vgpr35:0x0000000000000003, $vgpr36_vgpr37:0x0000000000000003, $vgpr38_vgpr39:0x0000000000000003, $vgpr40_vgpr41:0x000000000000000F, $vgpr42_vgpr43:0x000000000000000F, $vgpr44_vgpr45:0x000000000000000F, $vgpr46_vgpr47:0x000000000000000F, $vgpr48_vgpr49:0x0000000000000003, $vgpr50_vgpr51:0x0000000000000003, $vgpr52_vgpr53:0x0000000000000003, $vgpr54_vgpr55:0x0000000000000003, $vgpr56_vgpr57:0x000000000000000F, $vgpr58_vgpr59:0x000000000000000F, $vgpr60_vgpr61:0x000000000000000F, $vgpr62_vgpr63:0x000000000000000F, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX90A-NEXT: {{ $}}
- ; GFX90A-NEXT: renamable $vgpr10 = V_OR_B32_e32 $vgpr50, killed $vgpr16, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr54 = V_OR_B32_e32 killed $vgpr10, killed $vgpr14, implicit $exec
- ; GFX90A-NEXT: renamable $vgpr55 = V_MOV_B32_e32 0, implicit $exec
- ; GFX90A-NEXT: DS_WRITE_B64_gfx9 killed renamable $vgpr55, renamable $vgpr54_vgpr55, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
+ ; GFX90A-NEXT: renamable $vgpr11 = V_OR_B32_e32 $vgpr54, killed $vgpr20, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr18 = V_OR_B32_e32 killed $vgpr11, killed $vgpr18, implicit $exec
+ ; GFX90A-NEXT: renamable $vgpr19 = V_MOV_B32_e32 0, implicit $exec
+ ; GFX90A-NEXT: DS_WRITE_B64_gfx9 killed renamable $vgpr19, renamable $vgpr18_vgpr19, 0, 0, implicit $exec :: (store (s64) into `ptr addrspace(3) null`, addrspace 3)
; GFX90A-NEXT: renamable $sgpr12_sgpr13 = S_MOV_B64 0
- ; GFX90A-NEXT: S_BRANCH %bb.69
+ ; GFX90A-NEXT: S_BRANCH %bb.70
bb:
%i = tail call i32 @llvm.amdgcn.workitem.id.x()
%i11 = icmp eq i32 %i, 0
diff --git a/llvm/test/CodeGen/AMDGPU/call-argument-types.ll b/llvm/test/CodeGen/AMDGPU/call-argument-types.ll
index 9bbecacd6c774..dd77c419d6fd8 100644
--- a/llvm/test/CodeGen/AMDGPU/call-argument-types.ll
+++ b/llvm/test/CodeGen/AMDGPU/call-argument-types.ll
@@ -2489,7 +2489,7 @@ define amdgpu_kernel void @test_call_external_void_func_v3i16() #0 {
; CI-NEXT: s_add_u32 s36, s36, s3
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
-; CI-NEXT: buffer_load_dwordx2 v[2:3], off, s[0:3], 0
+; CI-NEXT: buffer_load_dwordx2 v[3:4], off, s[0:3], 0
; CI-NEXT: s_addc_u32 s37, s37, 0
; CI-NEXT: s_mov_b64 s[6:7], s[0:1]
; CI-NEXT: s_mov_b64 s[0:1], s[36:37]
@@ -2499,9 +2499,9 @@ define amdgpu_kernel void @test_call_external_void_func_v3i16() #0 {
; CI-NEXT: s_mov_b64 s[2:3], s[38:39]
; CI-NEXT: s_mov_b32 s32, 0
; CI-NEXT: s_waitcnt vmcnt(0)
-; CI-NEXT: v_alignbit_b32 v1, v3, v2, 16
-; CI-NEXT: v_mov_b32_e32 v0, v2
-; CI-NEXT: v_mov_b32_e32 v2, v3
+; CI-NEXT: v_lshr_b64 v[1:2], v[3:4], 16
+; CI-NEXT: v_mov_b32_e32 v0, v3
+; CI-NEXT: v_mov_b32_e32 v2, v4
; CI-NEXT: s_swappc_b64 s[30:31], s[4:5]
; CI-NEXT: s_endpgm
;
diff --git a/llvm/test/CodeGen/AMDGPU/cgp-bitfield-extract.ll b/llvm/test/CodeGen/AMDGPU/cgp-bitfield-extract.ll
index 14a96ac5c6338..e120b9791084b 100644
--- a/llvm/test/CodeGen/AMDGPU/cgp-bitfield-extract.ll
+++ b/llvm/test/CodeGen/AMDGPU/cgp-bitfield-extract.ll
@@ -176,11 +176,10 @@ ret:
; GCN-LABEL: {{^}}sink_ubfe_i64_span_midpoint:
; GCN: s_cbranch_scc{{[0-1]}} .LBB3_2
-; GCN: v_alignbit_b32 v[[LO:[0-9]+]], s{{[0-9]+}}, v{{[0-9]+}}, 30
-; GCN: v_and_b32_e32 v{{[0-9]+}}, 0x7f, v[[LO]]
+; GCN: s_lshr_b64
; GCN: .LBB3_3:
-; GCN: v_and_b32_e32 v{{[0-9]+}}, 0xff, v[[LO]]
+; GCN: s_lshr_b64
; GCN: buffer_store_dwordx2
define amdgpu_kernel void @sink_ubfe_i64_span_midpoint(ptr addrspace(1) %out, i64 %arg1, i1 %arg) #0 {
diff --git a/llvm/test/CodeGen/AMDGPU/fmed3.ll b/llvm/test/CodeGen/AMDGPU/fmed3.ll
index 57c692621c121..c70b921c97999 100644
--- a/llvm/test/CodeGen/AMDGPU/fmed3.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmed3.ll
@@ -7582,7 +7582,7 @@ define amdgpu_kernel void @v_test_nnan_input_fmed3_r_i_i_f16(ptr addrspace(1) %o
; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_1)
; GFX11-GISEL-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 1, v0
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v0, v1, s[2:3]
+; GFX11-GISEL-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[2:3]
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, 1.0, v0.l
; GFX11-GISEL-TRUE16-NEXT: v_med3_f16 v0.l, v0.l, 2.0, 4.0
@@ -7837,16 +7837,13 @@ define amdgpu_kernel void @v_nnan_inputs_med3_f16_pat0(ptr addrspace(1) %out, pt
; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-TRUE16-NEXT: v_lshlrev_b32_e32 v2, 1, v0
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v0, v2, s[2:3] glc dlc
+; GFX11-GISEL-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3] glc dlc
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v1, v2, s[4:5] glc dlc
+; GFX11-GISEL-TRUE16-NEXT: global_load_d16_hi_b16 v0, v2, s[4:5] glc dlc
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v3, v2, s[6:7] glc dlc
+; GFX11-GISEL-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[6:7] glc dlc
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, 1.0, v0.l
-; GFX11-GISEL-TRUE16-NEXT: v_mov_b16_e32 v0.h, v1.l
-; GFX11-GISEL-TRUE16-NEXT: v_mov_b16_e32 v1.l, v3.l
-; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.h, 2.0, v0.h
; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v1.l, 4.0, v1.l
; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
diff --git a/llvm/test/CodeGen/AMDGPU/function-returns.ll b/llvm/test/CodeGen/AMDGPU/function-returns.ll
index ac9f56d1ee7b1..a13d36e6992d4 100644
--- a/llvm/test/CodeGen/AMDGPU/function-returns.ll
+++ b/llvm/test/CodeGen/AMDGPU/function-returns.ll
@@ -894,11 +894,11 @@ define <3 x i16> @v3i16_func_void() #0 {
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: s_mov_b32 s7, 0xf000
; CI-NEXT: s_mov_b32 s6, -1
-; CI-NEXT: buffer_load_dwordx2 v[2:3], off, s[4:7], 0
+; CI-NEXT: buffer_load_dwordx2 v[3:4], off, s[4:7], 0
; CI-NEXT: s_waitcnt vmcnt(0)
-; CI-NEXT: v_alignbit_b32 v1, v3, v2, 16
-; CI-NEXT: v_mov_b32_e32 v0, v2
-; CI-NEXT: v_mov_b32_e32 v2, v3
+; CI-NEXT: v_lshr_b64 v[1:2], v[3:4], 16
+; CI-NEXT: v_mov_b32_e32 v0, v3
+; CI-NEXT: v_mov_b32_e32 v2, v4
; CI-NEXT: s_setpc_b64 s[30:31]
;
; GFX89-LABEL: v3i16_func_void:
@@ -1007,7 +1007,7 @@ define <5 x i16> @v5i16_func_void() #0 {
; CI-NEXT: buffer_load_dwordx2 v[0:1], off, s[4:7], 0
; CI-NEXT: buffer_load_sshort v4, off, s[4:7], 0 offset:8
; CI-NEXT: s_waitcnt vmcnt(1)
-; CI-NEXT: v_alignbit_b32 v5, v1, v0, 16
+; CI-NEXT: v_lshr_b64 v[5:6], v[0:1], 16
; CI-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; CI-NEXT: v_mov_b32_e32 v2, v1
; CI-NEXT: v_mov_b32_e32 v1, v5
diff --git a/llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll b/llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll
index ea3d57d127151..d0500fc945077 100644
--- a/llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll
+++ b/llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll
@@ -110,15 +110,13 @@ define amdgpu_kernel void @f2(i32 %arg, i32 %arg1, i32 %arg2, i1 %arg3, i32 %arg
; GFX11-NEXT: s_cbranch_scc0 .LBB2_8
; GFX11-NEXT: ; %bb.5: ; %bb18.preheader
; GFX11-NEXT: s_load_b128 s[28:31], s[16:17], 0x44
+; GFX11-NEXT: v_mov_b32_e32 v0, 0
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-NEXT: s_mul_hi_u32 s0, s29, s28
-; GFX11-NEXT: s_mul_i32 s1, s29, s28
-; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GFX11-NEXT: v_alignbit_b32 v0, s0, s1, 1
+; GFX11-NEXT: s_mul_hi_u32 s1, s29, s28
+; GFX11-NEXT: s_mul_i32 s0, s29, s28
+; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT: s_lshr_b64 s[0:1], s[0:1], 1
; GFX11-NEXT: s_mov_b32 s1, 0
-; GFX11-NEXT: v_readfirstlane_b32 s0, v0
-; GFX11-NEXT: v_mov_b32_e32 v0, 0
-; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_or_b32 s0, s0, 1
; GFX11-NEXT: s_lshr_b32 s0, s0, s30
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
diff --git a/llvm/test/CodeGen/AMDGPU/minimummaximum.ll b/llvm/test/CodeGen/AMDGPU/minimummaximum.ll
index a9d760d19ec04..deedb36a69cbd 100644
--- a/llvm/test/CodeGen/AMDGPU/minimummaximum.ll
+++ b/llvm/test/CodeGen/AMDGPU/minimummaximum.ll
@@ -181,17 +181,29 @@ define amdgpu_ps void @s_test_minmax_f16(half inreg %a, half inreg %b, half inre
; SDAG-FAKE16-NEXT: global_store_b16 v0, v1, s[4:5]
; SDAG-FAKE16-NEXT: s_endpgm
;
-; GISEL-LABEL: s_test_minmax_f16:
-; GISEL: ; %bb.0:
-; GISEL-NEXT: s_maximum_f16 s0, s0, s1
-; GISEL-NEXT: s_mov_b32 s6, s3
-; GISEL-NEXT: s_mov_b32 s7, s4
-; GISEL-NEXT: v_mov_b32_e32 v1, 0
-; GISEL-NEXT: s_minimum_f16 s0, s0, s2
-; GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GISEL-NEXT: global_store_b16 v1, v0, s[6:7]
-; GISEL-NEXT: s_endpgm
+; GISEL-TRUE16-LABEL: s_test_minmax_f16:
+; GISEL-TRUE16: ; %bb.0:
+; GISEL-TRUE16-NEXT: s_maximum_f16 s0, s0, s1
+; GISEL-TRUE16-NEXT: v_mov_b32_e32 v1, 0
+; GISEL-TRUE16-NEXT: s_mov_b32 s6, s3
+; GISEL-TRUE16-NEXT: s_mov_b32 s7, s4
+; GISEL-TRUE16-NEXT: s_minimum_f16 s0, s0, s2
+; GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GISEL-TRUE16-NEXT: v_mov_b16_e32 v0.l, s0
+; GISEL-TRUE16-NEXT: global_store_b16 v1, v0, s[6:7]
+; GISEL-TRUE16-NEXT: s_endpgm
+;
+; GISEL-FAKE16-LABEL: s_test_minmax_f16:
+; GISEL-FAKE16: ; %bb.0:
+; GISEL-FAKE16-NEXT: s_maximum_f16 s0, s0, s1
+; GISEL-FAKE16-NEXT: s_mov_b32 s6, s3
+; GISEL-FAKE16-NEXT: s_mov_b32 s7, s4
+; GISEL-FAKE16-NEXT: v_mov_b32_e32 v1, 0
+; GISEL-FAKE16-NEXT: s_minimum_f16 s0, s0, s2
+; GISEL-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GISEL-FAKE16-NEXT: v_mov_b32_e32 v0, s0
+; GISEL-FAKE16-NEXT: global_store_b16 v1, v0, s[6:7]
+; GISEL-FAKE16-NEXT: s_endpgm
%smax = call half @llvm.maximum.f16(half %a, half %b)
%sminmax = call half @llvm.minimum.f16(half %smax, half %c)
store half %sminmax, ptr addrspace(1) %out
diff --git a/llvm/test/CodeGen/AMDGPU/minmax.ll b/llvm/test/CodeGen/AMDGPU/minmax.ll
index 954dab3d0fc6f..b16d0b7e1213b 100644
--- a/llvm/test/CodeGen/AMDGPU/minmax.ll
+++ b/llvm/test/CodeGen/AMDGPU/minmax.ll
@@ -577,16 +577,27 @@ define amdgpu_ps void @s_test_minmax_f16_ieee_false(half inreg %a, half inreg %b
; SDAG-GFX12-FAKE16-NEXT: global_store_b16 v1, v0, s[4:5]
; SDAG-GFX12-FAKE16-NEXT: s_endpgm
;
-; GISEL-GFX12-LABEL: s_test_minmax_f16_ieee_false:
-; GISEL-GFX12: ; %bb.0:
-; GISEL-GFX12-NEXT: s_max_num_f16 s0, s0, s1
-; GISEL-GFX12-NEXT: s_mov_b32 s6, s3
-; GISEL-GFX12-NEXT: s_mov_b32 s7, s4
-; GISEL-GFX12-NEXT: v_mov_b32_e32 v1, 0
-; GISEL-GFX12-NEXT: s_min_num_f16 s0, s0, s2
-; GISEL-GFX12-NEXT: v_mov_b32_e32 v0, s0
-; GISEL-GFX12-NEXT: global_store_b16 v1, v0, s[6:7]
-; GISEL-GFX12-NEXT: s_endpgm
+; GISEL-GFX12-TRUE16-LABEL: s_test_minmax_f16_ieee_false:
+; GISEL-GFX12-TRUE16: ; %bb.0:
+; GISEL-GFX12-TRUE16-NEXT: s_max_num_f16 s0, s0, s1
+; GISEL-GFX12-TRUE16-NEXT: v_mov_b32_e32 v1, 0
+; GISEL-GFX12-TRUE16-NEXT: s_mov_b32 s6, s3
+; GISEL-GFX12-TRUE16-NEXT: s_mov_b32 s7, s4
+; GISEL-GFX12-TRUE16-NEXT: s_min_num_f16 s0, s0, s2
+; GISEL-GFX12-TRUE16-NEXT: v_mov_b16_e32 v0.l, s0
+; GISEL-GFX12-TRUE16-NEXT: global_store_b16 v1, v0, s[6:7]
+; GISEL-GFX12-TRUE16-NEXT: s_endpgm
+;
+; GISEL-GFX12-FAKE16-LABEL: s_test_minmax_f16_ieee_false:
+; GISEL-GFX12-FAKE16: ; %bb.0:
+; GISEL-GFX12-FAKE16-NEXT: s_max_num_f16 s0, s0, s1
+; GISEL-GFX12-FAKE16-NEXT: s_mov_b32 s6, s3
+; GISEL-GFX12-FAKE16-NEXT: s_mov_b32 s7, s4
+; GISEL-GFX12-FAKE16-NEXT: v_mov_b32_e32 v1, 0
+; GISEL-GFX12-FAKE16-NEXT: s_min_num_f16 s0, s0, s2
+; GISEL-GFX12-FAKE16-NEXT: v_mov_b32_e32 v0, s0
+; GISEL-GFX12-FAKE16-NEXT: global_store_b16 v1, v0, s[6:7]
+; GISEL-GFX12-FAKE16-NEXT: s_endpgm
%smax = call half @llvm.maxnum.f16(half %a, half %b)
%sminmax = call half @llvm.minnum.f16(half %smax, half %c)
store half %sminmax, ptr addrspace(1) %out
diff --git a/llvm/test/CodeGen/AMDGPU/mul_int24.ll b/llvm/test/CodeGen/AMDGPU/mul_int24.ll
index ad3f36dc3f216..38a1be57be015 100644
--- a/llvm/test/CodeGen/AMDGPU/mul_int24.ll
+++ b/llvm/test/CodeGen/AMDGPU/mul_int24.ll
@@ -819,9 +819,10 @@ define amdgpu_kernel void @test_umul_i24(ptr addrspace(1) %out, i32 %arg) {
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_waitcnt lgkmcnt(0)
; SI-NEXT: s_lshr_b32 s2, s2, 9
-; SI-NEXT: v_mul_hi_u32 v0, s2, v0
-; SI-NEXT: s_mul_i32 s2, s2, 0xff803fe1
-; SI-NEXT: v_alignbit_b32 v0, v0, s2, 1
+; SI-NEXT: s_mul_i32 s4, s2, 0xff803fe1
+; SI-NEXT: v_mul_hi_u32 v1, s2, v0
+; SI-NEXT: v_mov_b32_e32 v0, s4
+; SI-NEXT: v_lshr_b64 v[0:1], v[0:1], 1
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; SI-NEXT: s_endpgm
@@ -836,7 +837,7 @@ define amdgpu_kernel void @test_umul_i24(ptr addrspace(1) %out, i32 %arg) {
; VI-NEXT: s_lshr_b32 s0, s0, 9
; VI-NEXT: v_mad_u64_u32 v[0:1], s[0:1], s0, v0, 0
; VI-NEXT: s_mov_b64 s[0:1], 0
-; VI-NEXT: v_alignbit_b32 v0, v1, v0, 1
+; VI-NEXT: v_lshrrev_b64 v[0:1], 1, v[0:1]
; VI-NEXT: s_nop 2
; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT: s_endpgm
@@ -848,11 +849,11 @@ define amdgpu_kernel void @test_umul_i24(ptr addrspace(1) %out, i32 %arg) {
; GFX9-NEXT: s_mov_b32 s3, 0xf000
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_lshr_b32 s2, s2, 9
-; GFX9-NEXT: s_mul_hi_u32 s4, s2, 0xff803fe1
-; GFX9-NEXT: s_mul_i32 s2, s2, 0xff803fe1
-; GFX9-NEXT: v_mov_b32_e32 v0, s2
-; GFX9-NEXT: v_alignbit_b32 v0, s4, v0, 1
+; GFX9-NEXT: s_mul_hi_u32 s5, s2, 0xff803fe1
+; GFX9-NEXT: s_mul_i32 s4, s2, 0xff803fe1
+; GFX9-NEXT: s_lshr_b64 s[4:5], s[4:5], 1
; GFX9-NEXT: s_mov_b32 s2, -1
+; GFX9-NEXT: v_mov_b32_e32 v0, s4
; GFX9-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX9-NEXT: s_endpgm
;
diff --git a/llvm/test/CodeGen/AMDGPU/sdiv64.ll b/llvm/test/CodeGen/AMDGPU/sdiv64.ll
index 96dd6276f7e38..ea71736ceac43 100644
--- a/llvm/test/CodeGen/AMDGPU/sdiv64.ll
+++ b/llvm/test/CodeGen/AMDGPU/sdiv64.ll
@@ -1037,27 +1037,26 @@ define amdgpu_kernel void @s_test_sdiv24_48(ptr addrspace(1) %out, i48 %x, i48 %
; GCN-NEXT: s_mov_b32 s7, 0xf000
; GCN-NEXT: s_mov_b32 s6, -1
; GCN-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-NEXT: s_mov_b32 s5, s1
-; GCN-NEXT: s_sext_i32_i16 s1, s9
-; GCN-NEXT: v_mov_b32_e32 v0, s8
-; GCN-NEXT: v_alignbit_b32 v0, s1, v0, 24
-; GCN-NEXT: v_cvt_f32_i32_e32 v1, v0
; GCN-NEXT: s_mov_b32 s4, s0
-; GCN-NEXT: s_sext_i32_i16 s0, s3
-; GCN-NEXT: v_mov_b32_e32 v2, s2
-; GCN-NEXT: v_alignbit_b32 v2, s0, v2, 24
-; GCN-NEXT: v_cvt_f32_i32_e32 v3, v2
-; GCN-NEXT: v_rcp_iflag_f32_e32 v4, v1
-; GCN-NEXT: v_xor_b32_e32 v0, v2, v0
-; GCN-NEXT: v_ashrrev_i32_e32 v0, 30, v0
-; GCN-NEXT: v_or_b32_e32 v0, 1, v0
-; GCN-NEXT: v_mul_f32_e32 v2, v3, v4
+; GCN-NEXT: s_sext_i32_i16 s9, s9
+; GCN-NEXT: s_mov_b32 s5, s1
+; GCN-NEXT: s_lshr_b64 s[0:1], s[8:9], 24
+; GCN-NEXT: v_cvt_f32_i32_e32 v0, s0
+; GCN-NEXT: s_sext_i32_i16 s3, s3
+; GCN-NEXT: s_lshr_b64 s[2:3], s[2:3], 24
+; GCN-NEXT: v_cvt_f32_i32_e32 v1, s2
+; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0
+; GCN-NEXT: s_xor_b32 s0, s2, s0
+; GCN-NEXT: s_ashr_i32 s0, s0, 30
+; GCN-NEXT: s_or_b32 s2, s0, 1
+; GCN-NEXT: v_mul_f32_e32 v2, v1, v2
; GCN-NEXT: v_trunc_f32_e32 v2, v2
-; GCN-NEXT: v_mad_f32 v3, -v2, v1, v3
+; GCN-NEXT: v_mad_f32 v1, -v2, v0, v1
; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v1|
-; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
-; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2
+; GCN-NEXT: v_cmp_ge_f32_e64 s[0:1], |v1|, |v0|
+; GCN-NEXT: s_and_b64 s[0:1], s[0:1], exec
+; GCN-NEXT: s_cselect_b32 s0, s2, 0
+; GCN-NEXT: v_add_i32_e32 v0, vcc, s0, v2
; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24
; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0
; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0
@@ -1071,27 +1070,26 @@ define amdgpu_kernel void @s_test_sdiv24_48(ptr addrspace(1) %out, i48 %x, i48 %
; GCN-IR-NEXT: s_mov_b32 s7, 0xf000
; GCN-IR-NEXT: s_mov_b32 s6, -1
; GCN-IR-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT: s_mov_b32 s5, s1
-; GCN-IR-NEXT: s_sext_i32_i16 s1, s9
-; GCN-IR-NEXT: v_mov_b32_e32 v0, s8
-; GCN-IR-NEXT: v_alignbit_b32 v0, s1, v0, 24
-; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, v0
; GCN-IR-NEXT: s_mov_b32 s4, s0
-; GCN-IR-NEXT: s_sext_i32_i16 s0, s3
-; GCN-IR-NEXT: v_mov_b32_e32 v2, s2
-; GCN-IR-NEXT: v_alignbit_b32 v2, s0, v2, 24
-; GCN-IR-NEXT: v_cvt_f32_i32_e32 v3, v2
-; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v4, v1
-; GCN-IR-NEXT: v_xor_b32_e32 v0, v2, v0
-; GCN-IR-NEXT: v_ashrrev_i32_e32 v0, 30, v0
-; GCN-IR-NEXT: v_or_b32_e32 v0, 1, v0
-; GCN-IR-NEXT: v_mul_f32_e32 v2, v3, v4
+; GCN-IR-NEXT: s_sext_i32_i16 s9, s9
+; GCN-IR-NEXT: s_mov_b32 s5, s1
+; GCN-IR-NEXT: s_lshr_b64 s[0:1], s[8:9], 24
+; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s0
+; GCN-IR-NEXT: s_sext_i32_i16 s3, s3
+; GCN-IR-NEXT: s_lshr_b64 s[2:3], s[2:3], 24
+; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s2
+; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v0
+; GCN-IR-NEXT: s_xor_b32 s0, s2, s0
+; GCN-IR-NEXT: s_ashr_i32 s0, s0, 30
+; GCN-IR-NEXT: s_or_b32 s2, s0, 1
+; GCN-IR-NEXT: v_mul_f32_e32 v2, v1, v2
; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2
-; GCN-IR-NEXT: v_mad_f32 v3, -v2, v1, v3
+; GCN-IR-NEXT: v_mad_f32 v1, -v2, v0, v1
; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v1|
-; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
-; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT: v_cmp_ge_f32_e64 s[0:1], |v1|, |v0|
+; GCN-IR-NEXT: s_and_b64 s[0:1], s[0:1], exec
+; GCN-IR-NEXT: s_cselect_b32 s0, s2, 0
+; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, s0, v2
; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24
; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0
; GCN-IR-NEXT: buffer_store_dword v0, off, s[4:7], 0
diff --git a/llvm/test/CodeGen/AMDGPU/shift-and-i64-ubfe.ll b/llvm/test/CodeGen/AMDGPU/shift-and-i64-ubfe.ll
index 6de015c6de79b..8ea51b63fd1be 100644
--- a/llvm/test/CodeGen/AMDGPU/shift-and-i64-ubfe.ll
+++ b/llvm/test/CodeGen/AMDGPU/shift-and-i64-ubfe.ll
@@ -156,8 +156,8 @@ define amdgpu_kernel void @v_uextract_bit_1_31_i64(ptr addrspace(1) %out, ptr ad
; Truncated after the shift, so only low shift result is used.
; GCN-LABEL: {{^}}v_uextract_bit_31_32_i64:
; GCN: buffer_load_dwordx2 v[[[VALLO:[0-9]+]]:[[VALHI:[0-9]+]]]
-; GCN: v_alignbit_b32 v[[SHRLO:[0-9]+]], v[[VALHI]], v[[VALLO]], 31
-; GCN-DAG: v_and_b32_e32 v[[AND:[0-9]+]], 3, v[[SHRLO]]{{$}}
+; GCN: s_mov_b64
+; GCN-DAG: v_and_b32_e32 v[[AND:[0-9]+]], 3, v[[SHRLO:[0-9]+]]{{$}}
; GCN-DAG: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0{{$}}
; GCN: buffer_store_dwordx2 v[[[AND]]:[[ZERO]]]
define amdgpu_kernel void @v_uextract_bit_31_32_i64(ptr addrspace(1) %out, ptr addrspace(1) %in) #1 {
diff --git a/llvm/test/CodeGen/AMDGPU/shrink-add-sub-constant.ll b/llvm/test/CodeGen/AMDGPU/shrink-add-sub-constant.ll
index cc08708b63005..fe45d7d0284bb 100644
--- a/llvm/test/CodeGen/AMDGPU/shrink-add-sub-constant.ll
+++ b/llvm/test/CodeGen/AMDGPU/shrink-add-sub-constant.ll
@@ -1366,7 +1366,7 @@ define amdgpu_kernel void @v_test_i16_x_sub_64(ptr addrspace(1) %out, ptr addrsp
; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 1, v0
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v0, v1, s[2:3]
+; GFX11-GISEL-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[2:3]
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-GISEL-TRUE16-NEXT: v_add_nc_u16 v0.l, 0xffc0, v0.l
; GFX11-GISEL-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1]
@@ -1559,7 +1559,7 @@ define amdgpu_kernel void @v_test_i16_x_sub_64_zext_to_i32(ptr addrspace(1) %out
; GFX11-GISEL-TRUE16-NEXT: v_lshlrev_b32_e32 v0, 1, v1
; GFX11-GISEL-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v0, v0, s[2:3]
+; GFX11-GISEL-TRUE16-NEXT: global_load_d16_b16 v0, v0, s[2:3]
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-GISEL-TRUE16-NEXT: v_add_nc_u16 v0.l, 0xffc0, v0.l
; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
@@ -1799,21 +1799,16 @@ define amdgpu_kernel void @v_test_i16_x_sub_64_multi_use(ptr addrspace(1) %out,
; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-TRUE16-NEXT: v_lshlrev_b32_e32 v1, 1, v0
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v0, v1, s[2:3] glc dlc
+; GFX11-GISEL-TRUE16-NEXT: global_load_d16_b16 v0, v1, s[2:3] glc dlc
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-TRUE16-NEXT: global_load_u16 v2, v1, s[2:3] glc dlc
+; GFX11-GISEL-TRUE16-NEXT: global_load_d16_hi_b16 v0, v1, s[2:3] glc dlc
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-GISEL-TRUE16-NEXT: v_add_nc_u16 v0.l, 0xffc0, v0.l
-; GFX11-GISEL-TRUE16-NEXT: v_mov_b16_e32 v0.h, v2.l
-; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-GISEL-TRUE16-NEXT: v_mov_b16_e32 v2.l, v0.l
; GFX11-GISEL-TRUE16-NEXT: v_add_nc_u16 v0.h, 0xffc0, v0.h
-; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-GISEL-TRUE16-NEXT: v_mov_b16_e32 v0.l, v0.h
-; GFX11-GISEL-TRUE16-NEXT: global_store_b16 v1, v2, s[0:1] dlc
-; GFX11-GISEL-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0
; GFX11-GISEL-TRUE16-NEXT: global_store_b16 v1, v0, s[0:1] dlc
; GFX11-GISEL-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0
+; GFX11-GISEL-TRUE16-NEXT: global_store_d16_hi_b16 v1, v0, s[0:1] dlc
+; GFX11-GISEL-TRUE16-NEXT: s_waitcnt_vscnt null, 0x0
; GFX11-GISEL-TRUE16-NEXT: s_endpgm
;
; GFX11-GISEL-FAKE16-LABEL: v_test_i16_x_sub_64_multi_use:
diff --git a/llvm/test/CodeGen/AMDGPU/srem64.ll b/llvm/test/CodeGen/AMDGPU/srem64.ll
index 23364e860d154..169ee79a6ac37 100644
--- a/llvm/test/CodeGen/AMDGPU/srem64.ll
+++ b/llvm/test/CodeGen/AMDGPU/srem64.ll
@@ -1153,34 +1153,33 @@ define amdgpu_kernel void @s_test_srem33_64(ptr addrspace(1) %out, i64 %x, i64 %
define amdgpu_kernel void @s_test_srem24_48(ptr addrspace(1) %out, i48 %x, i48 %y) {
; GCN-LABEL: s_test_srem24_48:
; GCN: ; %bb.0:
+; GCN-NEXT: s_load_dwordx2 s[8:9], s[4:5], 0xd
; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
-; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
; GCN-NEXT: s_mov_b32 s7, 0xf000
; GCN-NEXT: s_mov_b32 s6, -1
; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: s_sext_i32_i16 s9, s9
+; GCN-NEXT: s_lshr_b64 s[4:5], s[8:9], 24
+; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4
; GCN-NEXT: s_sext_i32_i16 s3, s3
-; GCN-NEXT: s_sext_i32_i16 s5, s5
-; GCN-NEXT: v_mov_b32_e32 v0, s4
-; GCN-NEXT: v_alignbit_b32 v0, s5, v0, 24
-; GCN-NEXT: v_cvt_f32_i32_e32 v1, v0
-; GCN-NEXT: v_mov_b32_e32 v2, s2
-; GCN-NEXT: v_alignbit_b32 v2, s3, v2, 24
-; GCN-NEXT: v_cvt_f32_i32_e32 v3, v2
-; GCN-NEXT: v_rcp_iflag_f32_e32 v4, v1
-; GCN-NEXT: v_xor_b32_e32 v5, v2, v0
-; GCN-NEXT: v_ashrrev_i32_e32 v5, 30, v5
-; GCN-NEXT: v_or_b32_e32 v5, 1, v5
-; GCN-NEXT: v_mul_f32_e32 v4, v3, v4
-; GCN-NEXT: v_trunc_f32_e32 v4, v4
-; GCN-NEXT: v_mad_f32 v3, -v4, v1, v3
-; GCN-NEXT: v_cvt_i32_f32_e32 v4, v4
-; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v1|
-; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v5, vcc
+; GCN-NEXT: s_lshr_b64 s[2:3], s[2:3], 24
+; GCN-NEXT: v_cvt_f32_i32_e32 v1, s2
+; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0
+; GCN-NEXT: s_xor_b32 s3, s2, s4
+; GCN-NEXT: s_ashr_i32 s3, s3, 30
+; GCN-NEXT: s_or_b32 s3, s3, 1
+; GCN-NEXT: v_mul_f32_e32 v2, v1, v2
+; GCN-NEXT: v_trunc_f32_e32 v2, v2
+; GCN-NEXT: v_mad_f32 v1, -v2, v0, v1
+; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2
+; GCN-NEXT: v_cmp_ge_f32_e64 s[8:9], |v1|, |v0|
+; GCN-NEXT: s_and_b64 s[8:9], s[8:9], exec
+; GCN-NEXT: s_cselect_b32 s3, s3, 0
+; GCN-NEXT: v_add_i32_e32 v0, vcc, s3, v2
+; GCN-NEXT: v_mul_lo_u32 v0, v0, s4
; GCN-NEXT: s_mov_b32 s4, s0
-; GCN-NEXT: v_add_i32_e32 v1, vcc, v1, v4
-; GCN-NEXT: v_mul_lo_u32 v0, v1, v0
; GCN-NEXT: s_mov_b32 s5, s1
-; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v0, v2
+; GCN-NEXT: v_sub_i32_e32 v0, vcc, s2, v0
; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24
; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0
; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0
@@ -1189,34 +1188,33 @@ define amdgpu_kernel void @s_test_srem24_48(ptr addrspace(1) %out, i48 %x, i48 %
;
; GCN-IR-LABEL: s_test_srem24_48:
; GCN-IR: ; %bb.0:
+; GCN-IR-NEXT: s_load_dwordx2 s[8:9], s[4:5], 0xd
; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x9
-; GCN-IR-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0xd
; GCN-IR-NEXT: s_mov_b32 s7, 0xf000
; GCN-IR-NEXT: s_mov_b32 s6, -1
; GCN-IR-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-IR-NEXT: s_sext_i32_i16 s9, s9
+; GCN-IR-NEXT: s_lshr_b64 s[4:5], s[8:9], 24
+; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4
; GCN-IR-NEXT: s_sext_i32_i16 s3, s3
-; GCN-IR-NEXT: s_sext_i32_i16 s5, s5
-; GCN-IR-NEXT: v_mov_b32_e32 v0, s4
-; GCN-IR-NEXT: v_alignbit_b32 v0, s5, v0, 24
-; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, v0
-; GCN-IR-NEXT: v_mov_b32_e32 v2, s2
-; GCN-IR-NEXT: v_alignbit_b32 v2, s3, v2, 24
-; GCN-IR-NEXT: v_cvt_f32_i32_e32 v3, v2
-; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v4, v1
-; GCN-IR-NEXT: v_xor_b32_e32 v5, v2, v0
-; GCN-IR-NEXT: v_ashrrev_i32_e32 v5, 30, v5
-; GCN-IR-NEXT: v_or_b32_e32 v5, 1, v5
-; GCN-IR-NEXT: v_mul_f32_e32 v4, v3, v4
-; GCN-IR-NEXT: v_trunc_f32_e32 v4, v4
-; GCN-IR-NEXT: v_mad_f32 v3, -v4, v1, v3
-; GCN-IR-NEXT: v_cvt_i32_f32_e32 v4, v4
-; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v1|
-; GCN-IR-NEXT: v_cndmask_b32_e32 v1, 0, v5, vcc
+; GCN-IR-NEXT: s_lshr_b64 s[2:3], s[2:3], 24
+; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s2
+; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v0
+; GCN-IR-NEXT: s_xor_b32 s3, s2, s4
+; GCN-IR-NEXT: s_ashr_i32 s3, s3, 30
+; GCN-IR-NEXT: s_or_b32 s3, s3, 1
+; GCN-IR-NEXT: v_mul_f32_e32 v2, v1, v2
+; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2
+; GCN-IR-NEXT: v_mad_f32 v1, -v2, v0, v1
+; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2
+; GCN-IR-NEXT: v_cmp_ge_f32_e64 s[8:9], |v1|, |v0|
+; GCN-IR-NEXT: s_and_b64 s[8:9], s[8:9], exec
+; GCN-IR-NEXT: s_cselect_b32 s3, s3, 0
+; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, s3, v2
+; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s4
; GCN-IR-NEXT: s_mov_b32 s4, s0
-; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v1, v4
-; GCN-IR-NEXT: v_mul_lo_u32 v0, v1, v0
; GCN-IR-NEXT: s_mov_b32 s5, s1
-; GCN-IR-NEXT: v_subrev_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s2, v0
; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24
; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0
; GCN-IR-NEXT: buffer_store_dword v0, off, s[4:7], 0
diff --git a/llvm/test/CodeGen/AMDGPU/udiv64.ll b/llvm/test/CodeGen/AMDGPU/udiv64.ll
index db7d816386a70..bea1148df4738 100644
--- a/llvm/test/CodeGen/AMDGPU/udiv64.ll
+++ b/llvm/test/CodeGen/AMDGPU/udiv64.ll
@@ -722,16 +722,14 @@ define amdgpu_kernel void @s_test_udiv24_i48(ptr addrspace(1) %out, i48 %x, i48
; GCN-NEXT: s_mov_b32 s6, -1
; GCN-NEXT: v_mov_b32_e32 v3, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-NEXT: s_and_b32 s2, s2, 0xff000000
-; GCN-NEXT: s_and_b32 s4, s4, 0xff000000
-; GCN-NEXT: s_and_b32 s5, s5, 0xffff
-; GCN-NEXT: v_mov_b32_e32 v0, s4
-; GCN-NEXT: v_alignbit_b32 v0, s5, v0, 24
-; GCN-NEXT: v_cvt_f32_u32_e32 v0, v0
; GCN-NEXT: s_and_b32 s3, s3, 0xffff
-; GCN-NEXT: v_mov_b32_e32 v1, s2
-; GCN-NEXT: v_alignbit_b32 v1, s3, v1, 24
-; GCN-NEXT: v_cvt_f32_u32_e32 v1, v1
+; GCN-NEXT: s_and_b32 s5, s5, 0xffff
+; GCN-NEXT: s_and_b32 s4, s4, 0xff000000
+; GCN-NEXT: s_lshr_b64 s[4:5], s[4:5], 24
+; GCN-NEXT: v_cvt_f32_u32_e32 v0, s4
+; GCN-NEXT: s_and_b32 s2, s2, 0xff000000
+; GCN-NEXT: s_lshr_b64 s[2:3], s[2:3], 24
+; GCN-NEXT: v_cvt_f32_u32_e32 v1, s2
; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0
; GCN-NEXT: s_mov_b32 s4, s0
; GCN-NEXT: s_mov_b32 s5, s1
@@ -754,16 +752,14 @@ define amdgpu_kernel void @s_test_udiv24_i48(ptr addrspace(1) %out, i48 %x, i48
; GCN-IR-NEXT: s_mov_b32 s6, -1
; GCN-IR-NEXT: v_mov_b32_e32 v3, 0
; GCN-IR-NEXT: s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT: s_and_b32 s2, s2, 0xff000000
-; GCN-IR-NEXT: s_and_b32 s4, s4, 0xff000000
-; GCN-IR-NEXT: s_and_b32 s5, s5, 0xffff
-; GCN-IR-NEXT: v_mov_b32_e32 v0, s4
-; GCN-IR-NEXT: v_alignbit_b32 v0, s5, v0, 24
-; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, v0
; GCN-IR-NEXT: s_and_b32 s3, s3, 0xffff
-; GCN-IR-NEXT: v_mov_b32_e32 v1, s2
-; GCN-IR-NEXT: v_alignbit_b32 v1, s3, v1, 24
-; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, v1
+; GCN-IR-NEXT: s_and_b32 s5, s5, 0xffff
+; GCN-IR-NEXT: s_and_b32 s4, s4, 0xff000000
+; GCN-IR-NEXT: s_lshr_b64 s[4:5], s[4:5], 24
+; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s4
+; GCN-IR-NEXT: s_and_b32 s2, s2, 0xff000000
+; GCN-IR-NEXT: s_lshr_b64 s[2:3], s[2:3], 24
+; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, s2
; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v0
; GCN-IR-NEXT: s_mov_b32 s4, s0
; GCN-IR-NEXT: s_mov_b32 s5, s1
diff --git a/llvm/test/CodeGen/AMDGPU/uniform-alignbit.ll b/llvm/test/CodeGen/AMDGPU/uniform-alignbit.ll
new file mode 100644
index 0000000000000..fa0342bed513b
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/uniform-alignbit.ll
@@ -0,0 +1,38 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 < %s | FileCheck -check-prefixes=GCN %s
+
+define amdgpu_kernel void @uniform_build_vector(i64 %in, ptr addrspace(1) %out) {
+; GCN-LABEL: uniform_build_vector:
+; GCN: ; %bb.0: ; %entry
+; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: s_lshr_b64 s[4:5], s[0:1], 1
+; GCN-NEXT: s_mov_b32 s5, 0
+; GCN-NEXT: s_mov_b32 s6, s5
+; GCN-NEXT: s_mov_b32 s7, s5
+; GCN-NEXT: buffer_load_dwordx4 v[0:3], off, s[4:7], 0
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: v_mov_b32_e32 v1, 0
+; GCN-NEXT: ; sched_barrier mask(0x00000000)
+; GCN-NEXT: v_and_b32_e32 v0, 1, v0
+; GCN-NEXT: v_cvt_f32_ubyte0_e32 v0, v0
+; GCN-NEXT: global_store_dword v1, v0, s[2:3]
+; GCN-NEXT: s_endpgm
+entry:
+ %shifted = lshr i64 %in, 1
+ %trunc = trunc i64 %shifted to i32
+ %insert = insertelement <4 x i32> zeroinitializer, i32 %trunc, i64 0
+ %load = tail call <4 x i32> @llvm.amdgcn.raw.buffer.load.v4i32(<4 x i32> %insert, i32 0, i32 0, i32 0)
+ tail call void @llvm.amdgcn.sched.barrier(i32 0)
+ %extract = extractelement <4 x i32> %load, i64 0
+ %and = and i32 %extract, 1
+ %convert = sitofp i32 %and to float
+ store float %convert, ptr addrspace(1) %out
+ ret void
+}
+
+; Function Attrs: convergent nocallback nofree nounwind willreturn
+declare void @llvm.amdgcn.sched.barrier(i32 immarg) #0
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(read)
+declare <4 x i32> @llvm.amdgcn.raw.buffer.load.v4i32(<4 x i32>, i32, i32, i32 immarg) #1
diff --git a/llvm/test/CodeGen/AMDGPU/urem64.ll b/llvm/test/CodeGen/AMDGPU/urem64.ll
index a794d139063d5..5d0631488b93d 100644
--- a/llvm/test/CodeGen/AMDGPU/urem64.ll
+++ b/llvm/test/CodeGen/AMDGPU/urem64.ll
@@ -959,30 +959,30 @@ define amdgpu_kernel void @s_test_urem_k_den_i64(ptr addrspace(1) %out, i64 %x)
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_mul_hi_u32 v3, s2, v2
; GCN-NEXT: v_mul_hi_u32 v2, s3, v2
-; GCN-NEXT: v_mul_hi_u32 v1, s2, v0
; GCN-NEXT: s_mul_i32 s5, s3, 0xaaaaaaab
+; GCN-NEXT: v_mul_hi_u32 v1, s2, v0
; GCN-NEXT: v_add_i32_e32 v3, vcc, s5, v3
; GCN-NEXT: s_mov_b32 s4, s0
; GCN-NEXT: s_mul_i32 s0, s2, 0xaaaaaaaa
; GCN-NEXT: v_addc_u32_e32 v2, vcc, 0, v2, vcc
; GCN-NEXT: v_add_i32_e32 v3, vcc, s0, v3
-; GCN-NEXT: v_mul_hi_u32 v0, s3, v0
+; GCN-NEXT: v_mul_hi_u32 v3, s3, v0
; GCN-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; GCN-NEXT: v_add_i32_e32 v1, vcc, v2, v1
; GCN-NEXT: s_mul_i32 s0, s3, 0xaaaaaaaa
; GCN-NEXT: v_addc_u32_e64 v2, s[8:9], 0, 0, vcc
-; GCN-NEXT: v_add_i32_e32 v1, vcc, s0, v1
-; GCN-NEXT: v_addc_u32_e32 v0, vcc, v0, v2, vcc
-; GCN-NEXT: v_alignbit_b32 v1, v0, v1, 4
-; GCN-NEXT: v_lshrrev_b32_e32 v0, 4, v0
-; GCN-NEXT: v_mul_hi_u32 v2, v1, 24
+; GCN-NEXT: v_add_i32_e32 v0, vcc, s0, v1
+; GCN-NEXT: v_addc_u32_e32 v1, vcc, v3, v2, vcc
+; GCN-NEXT: v_lshrrev_b32_e32 v2, 4, v1
+; GCN-NEXT: v_lshr_b64 v[0:1], v[0:1], 4
+; GCN-NEXT: v_mul_lo_u32 v1, v2, 24
+; GCN-NEXT: v_mul_hi_u32 v2, v0, 24
; GCN-NEXT: v_mul_lo_u32 v0, v0, 24
-; GCN-NEXT: v_mul_lo_u32 v1, v1, 24
-; GCN-NEXT: v_mov_b32_e32 v3, s3
; GCN-NEXT: s_mov_b32 s5, s1
-; GCN-NEXT: v_add_i32_e32 v2, vcc, v0, v2
-; GCN-NEXT: v_sub_i32_e32 v0, vcc, s2, v1
-; GCN-NEXT: v_subb_u32_e32 v1, vcc, v3, v2, vcc
+; GCN-NEXT: v_add_i32_e32 v1, vcc, v1, v2
+; GCN-NEXT: v_mov_b32_e32 v2, s3
+; GCN-NEXT: v_sub_i32_e32 v0, vcc, s2, v0
+; GCN-NEXT: v_subb_u32_e32 v1, vcc, v2, v1, vcc
; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; GCN-NEXT: s_endpgm
;
diff --git a/llvm/test/CodeGen/AMDGPU/v_pack.ll b/llvm/test/CodeGen/AMDGPU/v_pack.ll
index 24f5821cbee54..7b60ea100a51c 100644
--- a/llvm/test/CodeGen/AMDGPU/v_pack.ll
+++ b/llvm/test/CodeGen/AMDGPU/v_pack.ll
@@ -108,15 +108,12 @@ define amdgpu_kernel void @v_pack_b32_v2f16(ptr addrspace(1) %in0, ptr addrspace
; GFX11-GISEL-REAL16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-REAL16-NEXT: v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-GISEL-REAL16-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-GISEL-REAL16-NEXT: v_lshlrev_b32_e32 v1, 1, v0
; GFX11-GISEL-REAL16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: global_load_u16 v1, v0, s[0:1] glc dlc
+; GFX11-GISEL-REAL16-NEXT: global_load_d16_b16 v0, v1, s[0:1] glc dlc
; GFX11-GISEL-REAL16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: global_load_u16 v2, v0, s[2:3] glc dlc
+; GFX11-GISEL-REAL16-NEXT: global_load_d16_hi_b16 v0, v1, s[2:3] glc dlc
; GFX11-GISEL-REAL16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: v_mov_b16_e32 v0.l, v1.l
-; GFX11-GISEL-REAL16-NEXT: v_mov_b16_e32 v0.h, v2.l
-; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-REAL16-NEXT: v_add_f16_e32 v0.l, 2.0, v0.l
; GFX11-GISEL-REAL16-NEXT: v_add_f16_e32 v0.h, 2.0, v0.h
; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
@@ -240,15 +237,12 @@ define amdgpu_kernel void @v_pack_b32_v2f16_sub(ptr addrspace(1) %in0, ptr addrs
; GFX11-GISEL-REAL16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-REAL16-NEXT: v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-GISEL-REAL16-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-GISEL-REAL16-NEXT: v_lshlrev_b32_e32 v1, 1, v0
; GFX11-GISEL-REAL16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: global_load_u16 v1, v0, s[0:1] glc dlc
+; GFX11-GISEL-REAL16-NEXT: global_load_d16_b16 v0, v1, s[0:1] glc dlc
; GFX11-GISEL-REAL16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: global_load_u16 v2, v0, s[2:3] glc dlc
+; GFX11-GISEL-REAL16-NEXT: global_load_d16_hi_b16 v0, v1, s[2:3] glc dlc
; GFX11-GISEL-REAL16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: v_mov_b16_e32 v0.l, v1.l
-; GFX11-GISEL-REAL16-NEXT: v_mov_b16_e32 v0.h, v2.l
-; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-REAL16-NEXT: v_subrev_f16_e32 v0.l, 2.0, v0.l
; GFX11-GISEL-REAL16-NEXT: v_add_f16_e32 v0.h, 2.0, v0.h
; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
@@ -486,15 +480,12 @@ define amdgpu_kernel void @v_pack_b32.fabs(ptr addrspace(1) %in0, ptr addrspace(
; GFX11-GISEL-REAL16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-REAL16-NEXT: v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-GISEL-REAL16-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-GISEL-REAL16-NEXT: v_lshlrev_b32_e32 v1, 1, v0
; GFX11-GISEL-REAL16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: global_load_u16 v1, v0, s[0:1] glc dlc
+; GFX11-GISEL-REAL16-NEXT: global_load_d16_b16 v0, v1, s[0:1] glc dlc
; GFX11-GISEL-REAL16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: global_load_u16 v2, v0, s[2:3] glc dlc
+; GFX11-GISEL-REAL16-NEXT: global_load_d16_hi_b16 v0, v1, s[2:3] glc dlc
; GFX11-GISEL-REAL16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: v_mov_b16_e32 v0.l, v1.l
-; GFX11-GISEL-REAL16-NEXT: v_mov_b16_e32 v0.h, v2.l
-; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-REAL16-NEXT: v_add_f16_e32 v0.l, 2.0, v0.l
; GFX11-GISEL-REAL16-NEXT: v_add_f16_e32 v0.h, 2.0, v0.h
; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
@@ -620,15 +611,12 @@ define amdgpu_kernel void @v_pack_b32.fneg(ptr addrspace(1) %in0, ptr addrspace(
; GFX11-GISEL-REAL16-NEXT: s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-REAL16-NEXT: v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-GISEL-REAL16-NEXT: v_lshlrev_b32_e32 v0, 1, v0
+; GFX11-GISEL-REAL16-NEXT: v_lshlrev_b32_e32 v1, 1, v0
; GFX11-GISEL-REAL16-NEXT: s_waitcnt lgkmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: global_load_u16 v1, v0, s[0:1] glc dlc
+; GFX11-GISEL-REAL16-NEXT: global_load_d16_b16 v0, v1, s[0:1] glc dlc
; GFX11-GISEL-REAL16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: global_load_u16 v2, v0, s[2:3] glc dlc
+; GFX11-GISEL-REAL16-NEXT: global_load_d16_hi_b16 v0, v1, s[2:3] glc dlc
; GFX11-GISEL-REAL16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-GISEL-REAL16-NEXT: v_mov_b16_e32 v0.l, v1.l
-; GFX11-GISEL-REAL16-NEXT: v_mov_b16_e32 v0.h, v2.l
-; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-REAL16-NEXT: v_add_f16_e32 v0.l, 2.0, v0.l
; GFX11-GISEL-REAL16-NEXT: v_add_f16_e32 v0.h, 2.0, v0.h
; GFX11-GISEL-REAL16-NEXT: s_delay_alu instid0(VALU_DEP_1)
More information about the llvm-commits
mailing list