summaryrefslogtreecommitdiff
path: root/llvm/test/CodeGen/AMDGPU/fptoi.i128.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/fptoi.i128.ll')
-rw-r--r--llvm/test/CodeGen/AMDGPU/fptoi.i128.ll390
1 files changed, 196 insertions, 194 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/fptoi.i128.ll b/llvm/test/CodeGen/AMDGPU/fptoi.i128.ll
index cdd34cbde6dd..e7af7467171c 100644
--- a/llvm/test/CodeGen/AMDGPU/fptoi.i128.ll
+++ b/llvm/test/CodeGen/AMDGPU/fptoi.i128.ll
@@ -11,8 +11,8 @@ define i128 @fptosi_f64_to_i128(double %x) {
; SDAG-NEXT: v_mov_b32_e32 v7, 0
; SDAG-NEXT: s_mov_b64 s[4:5], 0x3fe
; SDAG-NEXT: v_mov_b32_e32 v4, v0
-; SDAG-NEXT: v_cmp_lt_u64_e32 vcc, s[4:5], v[6:7]
; SDAG-NEXT: v_mov_b32_e32 v0, 0
+; SDAG-NEXT: v_cmp_lt_u64_e32 vcc, s[4:5], v[6:7]
; SDAG-NEXT: v_mov_b32_e32 v2, 0
; SDAG-NEXT: v_mov_b32_e32 v1, 0
; SDAG-NEXT: v_mov_b32_e32 v3, 0
@@ -57,33 +57,34 @@ define i128 @fptosi_f64_to_i128(double %x) {
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v7
; SDAG-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5]
; SDAG-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v7
-; SDAG-NEXT: v_cndmask_b32_e64 v6, 0, v1, s[6:7]
+; SDAG-NEXT: v_cndmask_b32_e64 v3, 0, v1, s[6:7]
; SDAG-NEXT: v_cndmask_b32_e64 v2, v2, v0, s[4:5]
; SDAG-NEXT: v_lshlrev_b64 v[0:1], v7, v[4:5]
-; SDAG-NEXT: v_cndmask_b32_e64 v2, 0, v2, s[6:7]
+; SDAG-NEXT: v_mov_b32_e32 v5, 0
; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v0, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e64 v11, 0, v1, s[4:5]
; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v12, v10, 0
-; SDAG-NEXT: v_mov_b32_e32 v3, 0
+; SDAG-NEXT: v_cndmask_b32_e64 v2, 0, v2, s[6:7]
; SDAG-NEXT: v_mul_lo_u32 v13, v8, v2
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v2, v1
-; SDAG-NEXT: v_mul_lo_u32 v14, v10, v6
-; SDAG-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v11, v10, v[2:3]
-; SDAG-NEXT: ; implicit-def: $vgpr10
-; SDAG-NEXT: v_add3_u32 v5, v5, v14, v13
-; SDAG-NEXT: v_mov_b32_e32 v2, v6
-; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v12, v8, v[2:3]
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v9, v12, v[4:5]
-; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v7, v2
-; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], 0, 0, s[4:5]
-; SDAG-NEXT: v_mul_lo_u32 v6, v9, v11
+; SDAG-NEXT: v_mov_b32_e32 v4, v1
+; SDAG-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v11, v10, v[4:5]
+; SDAG-NEXT: v_mul_lo_u32 v14, v10, v3
+; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v10, v2, 0
+; SDAG-NEXT: v_mov_b32_e32 v4, v6
+; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v12, v8, v[4:5]
+; SDAG-NEXT: v_add3_u32 v3, v3, v14, v13
+; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v9, v12, v[2:3]
+; SDAG-NEXT: v_add_co_u32_e64 v5, s[4:5], v7, v5
+; SDAG-NEXT: v_addc_co_u32_e64 v6, s[4:5], 0, 0, s[4:5]
+; SDAG-NEXT: v_mul_lo_u32 v3, v9, v11
; SDAG-NEXT: v_mul_lo_u32 v9, v9, v12
-; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v11, v8, v[2:3]
+; SDAG-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v8, v[5:6]
+; SDAG-NEXT: ; implicit-def: $vgpr10
; SDAG-NEXT: ; implicit-def: $vgpr8
-; SDAG-NEXT: v_add3_u32 v5, v9, v5, v6
-; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v2, v4
-; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v3, v5, s[4:5]
+; SDAG-NEXT: v_add3_u32 v3, v9, v2, v3
+; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v5, v1
+; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v6, v3, s[4:5]
+; SDAG-NEXT: v_mov_b32_e32 v1, v4
; SDAG-NEXT: ; implicit-def: $vgpr6_vgpr7
; SDAG-NEXT: ; implicit-def: $vgpr4_vgpr5
; SDAG-NEXT: ; implicit-def: $vgpr9
@@ -376,8 +377,8 @@ define i128 @fptoui_f64_to_i128(double %x) {
; SDAG-NEXT: v_mov_b32_e32 v7, 0
; SDAG-NEXT: s_mov_b64 s[4:5], 0x3fe
; SDAG-NEXT: v_mov_b32_e32 v4, v0
-; SDAG-NEXT: v_cmp_lt_u64_e32 vcc, s[4:5], v[6:7]
; SDAG-NEXT: v_mov_b32_e32 v0, 0
+; SDAG-NEXT: v_cmp_lt_u64_e32 vcc, s[4:5], v[6:7]
; SDAG-NEXT: v_mov_b32_e32 v2, 0
; SDAG-NEXT: v_mov_b32_e32 v1, 0
; SDAG-NEXT: v_mov_b32_e32 v3, 0
@@ -422,33 +423,34 @@ define i128 @fptoui_f64_to_i128(double %x) {
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v7
; SDAG-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5]
; SDAG-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v7
-; SDAG-NEXT: v_cndmask_b32_e64 v6, 0, v1, s[6:7]
+; SDAG-NEXT: v_cndmask_b32_e64 v3, 0, v1, s[6:7]
; SDAG-NEXT: v_cndmask_b32_e64 v2, v2, v0, s[4:5]
; SDAG-NEXT: v_lshlrev_b64 v[0:1], v7, v[4:5]
-; SDAG-NEXT: v_cndmask_b32_e64 v2, 0, v2, s[6:7]
+; SDAG-NEXT: v_mov_b32_e32 v5, 0
; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v0, s[4:5]
; SDAG-NEXT: v_cndmask_b32_e64 v11, 0, v1, s[4:5]
; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v12, v10, 0
-; SDAG-NEXT: v_mov_b32_e32 v3, 0
+; SDAG-NEXT: v_cndmask_b32_e64 v2, 0, v2, s[6:7]
; SDAG-NEXT: v_mul_lo_u32 v13, v8, v2
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v2, v1
-; SDAG-NEXT: v_mul_lo_u32 v14, v10, v6
-; SDAG-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v11, v10, v[2:3]
-; SDAG-NEXT: ; implicit-def: $vgpr10
-; SDAG-NEXT: v_add3_u32 v5, v5, v14, v13
-; SDAG-NEXT: v_mov_b32_e32 v2, v6
-; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v12, v8, v[2:3]
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v9, v12, v[4:5]
-; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v7, v2
-; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], 0, 0, s[4:5]
-; SDAG-NEXT: v_mul_lo_u32 v6, v9, v11
+; SDAG-NEXT: v_mov_b32_e32 v4, v1
+; SDAG-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v11, v10, v[4:5]
+; SDAG-NEXT: v_mul_lo_u32 v14, v10, v3
+; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v10, v2, 0
+; SDAG-NEXT: v_mov_b32_e32 v4, v6
+; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v12, v8, v[4:5]
+; SDAG-NEXT: v_add3_u32 v3, v3, v14, v13
+; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v9, v12, v[2:3]
+; SDAG-NEXT: v_add_co_u32_e64 v5, s[4:5], v7, v5
+; SDAG-NEXT: v_addc_co_u32_e64 v6, s[4:5], 0, 0, s[4:5]
+; SDAG-NEXT: v_mul_lo_u32 v3, v9, v11
; SDAG-NEXT: v_mul_lo_u32 v9, v9, v12
-; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v11, v8, v[2:3]
+; SDAG-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v8, v[5:6]
+; SDAG-NEXT: ; implicit-def: $vgpr10
; SDAG-NEXT: ; implicit-def: $vgpr8
-; SDAG-NEXT: v_add3_u32 v5, v9, v5, v6
-; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v2, v4
-; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v3, v5, s[4:5]
+; SDAG-NEXT: v_add3_u32 v3, v9, v2, v3
+; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v5, v1
+; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v6, v3, s[4:5]
+; SDAG-NEXT: v_mov_b32_e32 v1, v4
; SDAG-NEXT: ; implicit-def: $vgpr6_vgpr7
; SDAG-NEXT: ; implicit-def: $vgpr4_vgpr5
; SDAG-NEXT: ; implicit-def: $vgpr9
@@ -737,17 +739,17 @@ define i128 @fptosi_f32_to_i128(float %x) {
; SDAG: ; %bb.0: ; %fp-to-i-entry
; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SDAG-NEXT: v_mov_b32_e32 v4, v0
+; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_bfe_u32 v5, v4, 23, 8
; SDAG-NEXT: s_movk_i32 s4, 0x7e
-; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_mov_b32_e32 v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v6, 0
; SDAG-NEXT: v_mov_b32_e32 v1, 0
; SDAG-NEXT: v_mov_b32_e32 v3, 0
; SDAG-NEXT: v_cmp_lt_u32_e32 vcc, s4, v5
; SDAG-NEXT: s_and_saveexec_b64 s[8:9], vcc
; SDAG-NEXT: s_cbranch_execz .LBB2_10
; SDAG-NEXT: ; %bb.1: ; %fp-to-i-if-end
+; SDAG-NEXT: v_mov_b32_e32 v6, 0
; SDAG-NEXT: v_add_co_u32_e32 v0, vcc, 0xffffff01, v5
; SDAG-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v6, vcc
; SDAG-NEXT: v_addc_co_u32_e32 v2, vcc, -1, v6, vcc
@@ -765,14 +767,14 @@ define i128 @fptosi_f32_to_i128(float %x) {
; SDAG-NEXT: s_cbranch_execz .LBB2_7
; SDAG-NEXT: ; %bb.2: ; %fp-to-i-if-end9
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; SDAG-NEXT: v_add_co_u32_e64 v9, s[4:5], -1, v0
+; SDAG-NEXT: v_add_co_u32_e64 v10, s[4:5], -1, v0
; SDAG-NEXT: s_mov_b64 s[4:5], 0x95
; SDAG-NEXT: v_and_b32_e32 v0, 0x7fffff, v4
; SDAG-NEXT: v_cmp_lt_u64_e64 s[4:5], s[4:5], v[5:6]
-; SDAG-NEXT: v_mov_b32_e32 v7, 0
-; SDAG-NEXT: v_cndmask_b32_e64 v8, -1, 0, vcc
-; SDAG-NEXT: v_cndmask_b32_e64 v10, -1, 1, vcc
-; SDAG-NEXT: v_or_b32_e32 v6, 0x800000, v0
+; SDAG-NEXT: v_cndmask_b32_e64 v9, -1, 0, vcc
+; SDAG-NEXT: v_cndmask_b32_e64 v11, -1, 1, vcc
+; SDAG-NEXT: v_or_b32_e32 v7, 0x800000, v0
+; SDAG-NEXT: v_mov_b32_e32 v8, v6
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; implicit-def: $vgpr2_vgpr3
; SDAG-NEXT: s_and_saveexec_b64 s[6:7], s[4:5]
@@ -782,56 +784,56 @@ define i128 @fptosi_f32_to_i128(float %x) {
; SDAG-NEXT: v_sub_u32_e32 v0, 0xd6, v5
; SDAG-NEXT: v_add_u32_e32 v2, 0xffffff2a, v5
; SDAG-NEXT: v_add_u32_e32 v4, 0xffffff6a, v5
-; SDAG-NEXT: v_lshrrev_b64 v[0:1], v0, v[6:7]
-; SDAG-NEXT: v_lshlrev_b64 v[2:3], v2, v[6:7]
+; SDAG-NEXT: v_lshrrev_b64 v[0:1], v0, v[7:8]
+; SDAG-NEXT: v_lshlrev_b64 v[2:3], v2, v[7:8]
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v4
; SDAG-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5]
; SDAG-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v4
; SDAG-NEXT: v_cndmask_b32_e64 v3, 0, v1, s[6:7]
; SDAG-NEXT: v_cndmask_b32_e64 v2, v2, v0, s[4:5]
-; SDAG-NEXT: v_lshlrev_b64 v[0:1], v4, v[6:7]
+; SDAG-NEXT: v_lshlrev_b64 v[0:1], v4, v[7:8]
; SDAG-NEXT: v_cndmask_b32_e64 v2, 0, v2, s[6:7]
-; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v0, s[4:5]
-; SDAG-NEXT: v_cndmask_b32_e64 v11, 0, v1, s[4:5]
-; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v12, v10, 0
-; SDAG-NEXT: v_mul_lo_u32 v13, v8, v2
-; SDAG-NEXT: v_mul_lo_u32 v14, v10, v3
-; SDAG-NEXT: v_mov_b32_e32 v6, v1
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v10, v[6:7]
-; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v10, v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v6, v4
-; SDAG-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v12, v8, v[6:7]
-; SDAG-NEXT: v_add3_u32 v3, v3, v14, v13
-; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v9, v12, v[2:3]
-; SDAG-NEXT: v_add_co_u32_e64 v3, s[4:5], v5, v7
-; SDAG-NEXT: v_addc_co_u32_e64 v4, s[4:5], 0, 0, s[4:5]
-; SDAG-NEXT: v_mul_lo_u32 v10, v9, v11
-; SDAG-NEXT: v_mul_lo_u32 v9, v9, v12
-; SDAG-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v11, v8, v[3:4]
-; SDAG-NEXT: ; implicit-def: $vgpr8
-; SDAG-NEXT: v_add3_u32 v5, v9, v2, v10
-; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v3, v1
-; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v4, v5, s[4:5]
-; SDAG-NEXT: v_mov_b32_e32 v1, v6
-; SDAG-NEXT: ; implicit-def: $vgpr5_vgpr6
-; SDAG-NEXT: ; implicit-def: $vgpr6_vgpr7
-; SDAG-NEXT: ; implicit-def: $vgpr10
+; SDAG-NEXT: v_cndmask_b32_e64 v13, 0, v0, s[4:5]
+; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v1, s[4:5]
+; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v13, v11, 0
+; SDAG-NEXT: v_mul_lo_u32 v4, v9, v2
+; SDAG-NEXT: v_mul_lo_u32 v14, v11, v3
+; SDAG-NEXT: v_mov_b32_e32 v5, v1
+; SDAG-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v12, v11, v[5:6]
+; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v11, v2, 0
+; SDAG-NEXT: v_mov_b32_e32 v5, v7
+; SDAG-NEXT: v_mul_lo_u32 v7, v10, v13
+; SDAG-NEXT: v_add3_u32 v3, v3, v14, v4
+; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v13, v9, v[5:6]
+; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v10, v13, v[2:3]
+; SDAG-NEXT: v_add_co_u32_e64 v5, s[4:5], v8, v5
+; SDAG-NEXT: v_addc_co_u32_e64 v6, s[4:5], 0, 0, s[4:5]
+; SDAG-NEXT: v_mul_lo_u32 v3, v10, v12
+; SDAG-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v9, v[5:6]
+; SDAG-NEXT: ; implicit-def: $vgpr11
; SDAG-NEXT: ; implicit-def: $vgpr9
+; SDAG-NEXT: ; implicit-def: $vgpr10
+; SDAG-NEXT: v_add3_u32 v3, v7, v2, v3
+; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v5, v1
+; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v6, v3, s[4:5]
+; SDAG-NEXT: v_mov_b32_e32 v1, v4
+; SDAG-NEXT: ; implicit-def: $vgpr5_vgpr6
+; SDAG-NEXT: ; implicit-def: $vgpr7_vgpr8
; SDAG-NEXT: .LBB2_4: ; %Flow
; SDAG-NEXT: s_andn2_saveexec_b64 s[6:7], s[12:13]
; SDAG-NEXT: s_cbranch_execz .LBB2_6
; SDAG-NEXT: ; %bb.5: ; %fp-to-i-if-then12
; SDAG-NEXT: v_sub_u32_e32 v2, 0x96, v5
-; SDAG-NEXT: v_lshrrev_b64 v[0:1], v2, v[6:7]
+; SDAG-NEXT: v_lshrrev_b64 v[0:1], v2, v[7:8]
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v2
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, v0, s[4:5]
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v2
-; SDAG-NEXT: v_cndmask_b32_e64 v3, v0, v6, s[4:5]
-; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v3, v10, 0
+; SDAG-NEXT: v_cndmask_b32_e64 v3, v0, v7, s[4:5]
+; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v3, v11, 0
; SDAG-NEXT: v_mov_b32_e32 v2, 0
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v8, v[1:2]
+; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v9, v[1:2]
; SDAG-NEXT: v_mov_b32_e32 v1, v5
-; SDAG-NEXT: v_mad_i64_i32 v[2:3], s[4:5], v9, v3, v[1:2]
+; SDAG-NEXT: v_mad_i64_i32 v[2:3], s[4:5], v10, v3, v[1:2]
; SDAG-NEXT: v_mov_b32_e32 v1, v4
; SDAG-NEXT: .LBB2_6: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[6:7]
@@ -1088,17 +1090,17 @@ define i128 @fptoui_f32_to_i128(float %x) {
; SDAG: ; %bb.0: ; %fp-to-i-entry
; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SDAG-NEXT: v_mov_b32_e32 v4, v0
+; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_bfe_u32 v5, v4, 23, 8
; SDAG-NEXT: s_movk_i32 s4, 0x7e
-; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_mov_b32_e32 v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v6, 0
; SDAG-NEXT: v_mov_b32_e32 v1, 0
; SDAG-NEXT: v_mov_b32_e32 v3, 0
; SDAG-NEXT: v_cmp_lt_u32_e32 vcc, s4, v5
; SDAG-NEXT: s_and_saveexec_b64 s[8:9], vcc
; SDAG-NEXT: s_cbranch_execz .LBB3_10
; SDAG-NEXT: ; %bb.1: ; %fp-to-i-if-end
+; SDAG-NEXT: v_mov_b32_e32 v6, 0
; SDAG-NEXT: v_add_co_u32_e32 v0, vcc, 0xffffff01, v5
; SDAG-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v6, vcc
; SDAG-NEXT: v_addc_co_u32_e32 v2, vcc, -1, v6, vcc
@@ -1116,14 +1118,14 @@ define i128 @fptoui_f32_to_i128(float %x) {
; SDAG-NEXT: s_cbranch_execz .LBB3_7
; SDAG-NEXT: ; %bb.2: ; %fp-to-i-if-end9
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; SDAG-NEXT: v_add_co_u32_e64 v9, s[4:5], -1, v0
+; SDAG-NEXT: v_add_co_u32_e64 v10, s[4:5], -1, v0
; SDAG-NEXT: s_mov_b64 s[4:5], 0x95
; SDAG-NEXT: v_and_b32_e32 v0, 0x7fffff, v4
; SDAG-NEXT: v_cmp_lt_u64_e64 s[4:5], s[4:5], v[5:6]
-; SDAG-NEXT: v_mov_b32_e32 v7, 0
-; SDAG-NEXT: v_cndmask_b32_e64 v8, -1, 0, vcc
-; SDAG-NEXT: v_cndmask_b32_e64 v10, -1, 1, vcc
-; SDAG-NEXT: v_or_b32_e32 v6, 0x800000, v0
+; SDAG-NEXT: v_cndmask_b32_e64 v9, -1, 0, vcc
+; SDAG-NEXT: v_cndmask_b32_e64 v11, -1, 1, vcc
+; SDAG-NEXT: v_or_b32_e32 v7, 0x800000, v0
+; SDAG-NEXT: v_mov_b32_e32 v8, v6
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; implicit-def: $vgpr2_vgpr3
; SDAG-NEXT: s_and_saveexec_b64 s[6:7], s[4:5]
@@ -1133,56 +1135,56 @@ define i128 @fptoui_f32_to_i128(float %x) {
; SDAG-NEXT: v_sub_u32_e32 v0, 0xd6, v5
; SDAG-NEXT: v_add_u32_e32 v2, 0xffffff2a, v5
; SDAG-NEXT: v_add_u32_e32 v4, 0xffffff6a, v5
-; SDAG-NEXT: v_lshrrev_b64 v[0:1], v0, v[6:7]
-; SDAG-NEXT: v_lshlrev_b64 v[2:3], v2, v[6:7]
+; SDAG-NEXT: v_lshrrev_b64 v[0:1], v0, v[7:8]
+; SDAG-NEXT: v_lshlrev_b64 v[2:3], v2, v[7:8]
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v4
; SDAG-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5]
; SDAG-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v4
; SDAG-NEXT: v_cndmask_b32_e64 v3, 0, v1, s[6:7]
; SDAG-NEXT: v_cndmask_b32_e64 v2, v2, v0, s[4:5]
-; SDAG-NEXT: v_lshlrev_b64 v[0:1], v4, v[6:7]
+; SDAG-NEXT: v_lshlrev_b64 v[0:1], v4, v[7:8]
; SDAG-NEXT: v_cndmask_b32_e64 v2, 0, v2, s[6:7]
-; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v0, s[4:5]
-; SDAG-NEXT: v_cndmask_b32_e64 v11, 0, v1, s[4:5]
-; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v12, v10, 0
-; SDAG-NEXT: v_mul_lo_u32 v13, v8, v2
-; SDAG-NEXT: v_mul_lo_u32 v14, v10, v3
-; SDAG-NEXT: v_mov_b32_e32 v6, v1
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v10, v[6:7]
-; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v10, v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v6, v4
-; SDAG-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v12, v8, v[6:7]
-; SDAG-NEXT: v_add3_u32 v3, v3, v14, v13
-; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v9, v12, v[2:3]
-; SDAG-NEXT: v_add_co_u32_e64 v3, s[4:5], v5, v7
-; SDAG-NEXT: v_addc_co_u32_e64 v4, s[4:5], 0, 0, s[4:5]
-; SDAG-NEXT: v_mul_lo_u32 v10, v9, v11
-; SDAG-NEXT: v_mul_lo_u32 v9, v9, v12
-; SDAG-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v11, v8, v[3:4]
-; SDAG-NEXT: ; implicit-def: $vgpr8
-; SDAG-NEXT: v_add3_u32 v5, v9, v2, v10
-; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v3, v1
-; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v4, v5, s[4:5]
-; SDAG-NEXT: v_mov_b32_e32 v1, v6
-; SDAG-NEXT: ; implicit-def: $vgpr5_vgpr6
-; SDAG-NEXT: ; implicit-def: $vgpr6_vgpr7
-; SDAG-NEXT: ; implicit-def: $vgpr10
+; SDAG-NEXT: v_cndmask_b32_e64 v13, 0, v0, s[4:5]
+; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v1, s[4:5]
+; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v13, v11, 0
+; SDAG-NEXT: v_mul_lo_u32 v4, v9, v2
+; SDAG-NEXT: v_mul_lo_u32 v14, v11, v3
+; SDAG-NEXT: v_mov_b32_e32 v5, v1
+; SDAG-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v12, v11, v[5:6]
+; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v11, v2, 0
+; SDAG-NEXT: v_mov_b32_e32 v5, v7
+; SDAG-NEXT: v_mul_lo_u32 v7, v10, v13
+; SDAG-NEXT: v_add3_u32 v3, v3, v14, v4
+; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v13, v9, v[5:6]
+; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v10, v13, v[2:3]
+; SDAG-NEXT: v_add_co_u32_e64 v5, s[4:5], v8, v5
+; SDAG-NEXT: v_addc_co_u32_e64 v6, s[4:5], 0, 0, s[4:5]
+; SDAG-NEXT: v_mul_lo_u32 v3, v10, v12
+; SDAG-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v9, v[5:6]
+; SDAG-NEXT: ; implicit-def: $vgpr11
; SDAG-NEXT: ; implicit-def: $vgpr9
+; SDAG-NEXT: ; implicit-def: $vgpr10
+; SDAG-NEXT: v_add3_u32 v3, v7, v2, v3
+; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v5, v1
+; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v6, v3, s[4:5]
+; SDAG-NEXT: v_mov_b32_e32 v1, v4
+; SDAG-NEXT: ; implicit-def: $vgpr5_vgpr6
+; SDAG-NEXT: ; implicit-def: $vgpr7_vgpr8
; SDAG-NEXT: .LBB3_4: ; %Flow
; SDAG-NEXT: s_andn2_saveexec_b64 s[6:7], s[12:13]
; SDAG-NEXT: s_cbranch_execz .LBB3_6
; SDAG-NEXT: ; %bb.5: ; %fp-to-i-if-then12
; SDAG-NEXT: v_sub_u32_e32 v2, 0x96, v5
-; SDAG-NEXT: v_lshrrev_b64 v[0:1], v2, v[6:7]
+; SDAG-NEXT: v_lshrrev_b64 v[0:1], v2, v[7:8]
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v2
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, v0, s[4:5]
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v2
-; SDAG-NEXT: v_cndmask_b32_e64 v3, v0, v6, s[4:5]
-; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v3, v10, 0
+; SDAG-NEXT: v_cndmask_b32_e64 v3, v0, v7, s[4:5]
+; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v3, v11, 0
; SDAG-NEXT: v_mov_b32_e32 v2, 0
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v8, v[1:2]
+; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v9, v[1:2]
; SDAG-NEXT: v_mov_b32_e32 v1, v5
-; SDAG-NEXT: v_mad_i64_i32 v[2:3], s[4:5], v9, v3, v[1:2]
+; SDAG-NEXT: v_mad_i64_i32 v[2:3], s[4:5], v10, v3, v[1:2]
; SDAG-NEXT: v_mov_b32_e32 v1, v4
; SDAG-NEXT: .LBB3_6: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[6:7]
@@ -1477,17 +1479,17 @@ define i128 @fptosi_bf16_to_i128(bfloat %x) {
; SDAG: ; %bb.0: ; %fp-to-i-entry
; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SDAG-NEXT: v_mov_b32_e32 v4, v0
+; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_bfe_u32 v5, v4, 7, 8
; SDAG-NEXT: s_movk_i32 s4, 0x7e
-; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_mov_b32_e32 v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v6, 0
; SDAG-NEXT: v_mov_b32_e32 v1, 0
; SDAG-NEXT: v_mov_b32_e32 v3, 0
; SDAG-NEXT: v_cmp_lt_u32_e32 vcc, s4, v5
; SDAG-NEXT: s_and_saveexec_b64 s[8:9], vcc
; SDAG-NEXT: s_cbranch_execz .LBB6_10
; SDAG-NEXT: ; %bb.1: ; %fp-to-i-if-end
+; SDAG-NEXT: v_mov_b32_e32 v6, 0
; SDAG-NEXT: v_add_co_u32_e32 v0, vcc, 0xffffff01, v5
; SDAG-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v6, vcc
; SDAG-NEXT: v_addc_co_u32_e32 v2, vcc, -1, v6, vcc
@@ -1508,10 +1510,10 @@ define i128 @fptosi_bf16_to_i128(bfloat %x) {
; SDAG-NEXT: v_and_b32_sdwa v0, v4, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; SDAG-NEXT: s_mov_b64 s[4:5], 0x85
; SDAG-NEXT: v_cmp_lt_u64_e64 s[4:5], s[4:5], v[5:6]
-; SDAG-NEXT: v_mov_b32_e32 v7, 0
-; SDAG-NEXT: v_cndmask_b32_e64 v9, -1, 0, vcc
-; SDAG-NEXT: v_cndmask_b32_e64 v8, -1, 1, vcc
-; SDAG-NEXT: v_or_b32_e32 v6, 0x80, v0
+; SDAG-NEXT: v_cndmask_b32_e64 v10, -1, 0, vcc
+; SDAG-NEXT: v_cndmask_b32_e64 v9, -1, 1, vcc
+; SDAG-NEXT: v_or_b32_e32 v7, 0x80, v0
+; SDAG-NEXT: v_mov_b32_e32 v8, v6
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; implicit-def: $vgpr2_vgpr3
; SDAG-NEXT: s_and_saveexec_b64 s[6:7], s[4:5]
@@ -1519,56 +1521,56 @@ define i128 @fptosi_bf16_to_i128(bfloat %x) {
; SDAG-NEXT: s_cbranch_execz .LBB6_4
; SDAG-NEXT: ; %bb.3: ; %fp-to-i-if-else
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; SDAG-NEXT: v_add_co_u32_e64 v10, s[4:5], -1, v0
+; SDAG-NEXT: v_add_co_u32_e64 v11, s[4:5], -1, v0
; SDAG-NEXT: v_sub_u32_e32 v0, 0xc6, v5
; SDAG-NEXT: v_add_u32_e32 v2, 0xffffff3a, v5
; SDAG-NEXT: v_add_u32_e32 v4, 0xffffff7a, v5
-; SDAG-NEXT: v_lshrrev_b64 v[0:1], v0, v[6:7]
-; SDAG-NEXT: v_lshlrev_b64 v[2:3], v2, v[6:7]
+; SDAG-NEXT: v_lshrrev_b64 v[0:1], v0, v[7:8]
+; SDAG-NEXT: v_lshlrev_b64 v[2:3], v2, v[7:8]
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v4
; SDAG-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5]
; SDAG-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v4
; SDAG-NEXT: v_cndmask_b32_e64 v3, 0, v1, s[6:7]
; SDAG-NEXT: v_cndmask_b32_e64 v2, v2, v0, s[4:5]
-; SDAG-NEXT: v_lshlrev_b64 v[0:1], v4, v[6:7]
+; SDAG-NEXT: v_lshlrev_b64 v[0:1], v4, v[7:8]
; SDAG-NEXT: v_cndmask_b32_e64 v2, 0, v2, s[6:7]
-; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v0, s[4:5]
-; SDAG-NEXT: v_cndmask_b32_e64 v11, 0, v1, s[4:5]
-; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v12, v8, 0
-; SDAG-NEXT: v_mul_lo_u32 v13, v9, v2
-; SDAG-NEXT: v_mul_lo_u32 v14, v8, v3
-; SDAG-NEXT: v_mov_b32_e32 v6, v1
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v8, v[6:7]
-; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v8, v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v6, v4
-; SDAG-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v12, v9, v[6:7]
-; SDAG-NEXT: v_add3_u32 v3, v3, v14, v13
-; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v10, v12, v[2:3]
-; SDAG-NEXT: v_add_co_u32_e64 v3, s[4:5], v5, v7
-; SDAG-NEXT: v_addc_co_u32_e64 v4, s[4:5], 0, 0, s[4:5]
-; SDAG-NEXT: v_mul_lo_u32 v8, v10, v11
-; SDAG-NEXT: v_mul_lo_u32 v10, v10, v12
-; SDAG-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v11, v9, v[3:4]
-; SDAG-NEXT: v_add3_u32 v5, v10, v2, v8
-; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v3, v1
-; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v4, v5, s[4:5]
-; SDAG-NEXT: v_mov_b32_e32 v1, v6
+; SDAG-NEXT: v_cndmask_b32_e64 v13, 0, v0, s[4:5]
+; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v1, s[4:5]
+; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v13, v9, 0
+; SDAG-NEXT: v_mul_lo_u32 v4, v10, v2
+; SDAG-NEXT: v_mul_lo_u32 v14, v9, v3
+; SDAG-NEXT: v_mov_b32_e32 v5, v1
+; SDAG-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v12, v9, v[5:6]
+; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v9, v2, 0
+; SDAG-NEXT: v_mov_b32_e32 v5, v7
+; SDAG-NEXT: v_mul_lo_u32 v7, v11, v13
+; SDAG-NEXT: v_add3_u32 v3, v3, v14, v4
+; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v13, v10, v[5:6]
+; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v11, v13, v[2:3]
+; SDAG-NEXT: v_add_co_u32_e64 v5, s[4:5], v8, v5
+; SDAG-NEXT: v_addc_co_u32_e64 v6, s[4:5], 0, 0, s[4:5]
+; SDAG-NEXT: v_mul_lo_u32 v3, v11, v12
+; SDAG-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v10, v[5:6]
+; SDAG-NEXT: ; implicit-def: $vgpr9
+; SDAG-NEXT: v_add3_u32 v3, v7, v2, v3
+; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v5, v1
+; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v6, v3, s[4:5]
+; SDAG-NEXT: v_mov_b32_e32 v1, v4
; SDAG-NEXT: ; implicit-def: $vgpr5_vgpr6
-; SDAG-NEXT: ; implicit-def: $vgpr6_vgpr7
-; SDAG-NEXT: ; implicit-def: $vgpr8
+; SDAG-NEXT: ; implicit-def: $vgpr7_vgpr8
; SDAG-NEXT: .LBB6_4: ; %Flow
; SDAG-NEXT: s_andn2_saveexec_b64 s[6:7], s[12:13]
; SDAG-NEXT: s_cbranch_execz .LBB6_6
; SDAG-NEXT: ; %bb.5: ; %fp-to-i-if-then12
; SDAG-NEXT: v_sub_u32_e32 v2, 0x86, v5
-; SDAG-NEXT: v_lshrrev_b64 v[0:1], v2, v[6:7]
+; SDAG-NEXT: v_lshrrev_b64 v[0:1], v2, v[7:8]
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v2
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, v0, s[4:5]
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v2
-; SDAG-NEXT: v_cndmask_b32_e64 v0, v0, v6, s[4:5]
-; SDAG-NEXT: v_mul_hi_i32_i24_e32 v1, v0, v8
+; SDAG-NEXT: v_cndmask_b32_e64 v0, v0, v7, s[4:5]
+; SDAG-NEXT: v_mul_hi_i32_i24_e32 v1, v0, v9
; SDAG-NEXT: v_ashrrev_i32_e32 v2, 31, v1
-; SDAG-NEXT: v_mul_i32_i24_e32 v0, v0, v8
+; SDAG-NEXT: v_mul_i32_i24_e32 v0, v0, v9
; SDAG-NEXT: v_mov_b32_e32 v3, v2
; SDAG-NEXT: .LBB6_6: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[6:7]
@@ -1824,17 +1826,17 @@ define i128 @fptoui_bf16_to_i128(bfloat %x) {
; SDAG: ; %bb.0: ; %fp-to-i-entry
; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SDAG-NEXT: v_mov_b32_e32 v4, v0
+; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_bfe_u32 v5, v4, 7, 8
; SDAG-NEXT: s_movk_i32 s4, 0x7e
-; SDAG-NEXT: v_mov_b32_e32 v0, 0
; SDAG-NEXT: v_mov_b32_e32 v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v6, 0
; SDAG-NEXT: v_mov_b32_e32 v1, 0
; SDAG-NEXT: v_mov_b32_e32 v3, 0
; SDAG-NEXT: v_cmp_lt_u32_e32 vcc, s4, v5
; SDAG-NEXT: s_and_saveexec_b64 s[8:9], vcc
; SDAG-NEXT: s_cbranch_execz .LBB7_10
; SDAG-NEXT: ; %bb.1: ; %fp-to-i-if-end
+; SDAG-NEXT: v_mov_b32_e32 v6, 0
; SDAG-NEXT: v_add_co_u32_e32 v0, vcc, 0xffffff01, v5
; SDAG-NEXT: v_addc_co_u32_e32 v1, vcc, -1, v6, vcc
; SDAG-NEXT: v_addc_co_u32_e32 v2, vcc, -1, v6, vcc
@@ -1855,10 +1857,10 @@ define i128 @fptoui_bf16_to_i128(bfloat %x) {
; SDAG-NEXT: v_and_b32_sdwa v0, v4, s4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; SDAG-NEXT: s_mov_b64 s[4:5], 0x85
; SDAG-NEXT: v_cmp_lt_u64_e64 s[4:5], s[4:5], v[5:6]
-; SDAG-NEXT: v_mov_b32_e32 v7, 0
-; SDAG-NEXT: v_cndmask_b32_e64 v9, -1, 0, vcc
-; SDAG-NEXT: v_cndmask_b32_e64 v8, -1, 1, vcc
-; SDAG-NEXT: v_or_b32_e32 v6, 0x80, v0
+; SDAG-NEXT: v_cndmask_b32_e64 v10, -1, 0, vcc
+; SDAG-NEXT: v_cndmask_b32_e64 v9, -1, 1, vcc
+; SDAG-NEXT: v_or_b32_e32 v7, 0x80, v0
+; SDAG-NEXT: v_mov_b32_e32 v8, v6
; SDAG-NEXT: ; implicit-def: $vgpr0_vgpr1
; SDAG-NEXT: ; implicit-def: $vgpr2_vgpr3
; SDAG-NEXT: s_and_saveexec_b64 s[6:7], s[4:5]
@@ -1866,56 +1868,56 @@ define i128 @fptoui_bf16_to_i128(bfloat %x) {
; SDAG-NEXT: s_cbranch_execz .LBB7_4
; SDAG-NEXT: ; %bb.3: ; %fp-to-i-if-else
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; SDAG-NEXT: v_add_co_u32_e64 v10, s[4:5], -1, v0
+; SDAG-NEXT: v_add_co_u32_e64 v11, s[4:5], -1, v0
; SDAG-NEXT: v_sub_u32_e32 v0, 0xc6, v5
; SDAG-NEXT: v_add_u32_e32 v2, 0xffffff3a, v5
; SDAG-NEXT: v_add_u32_e32 v4, 0xffffff7a, v5
-; SDAG-NEXT: v_lshrrev_b64 v[0:1], v0, v[6:7]
-; SDAG-NEXT: v_lshlrev_b64 v[2:3], v2, v[6:7]
+; SDAG-NEXT: v_lshrrev_b64 v[0:1], v0, v[7:8]
+; SDAG-NEXT: v_lshlrev_b64 v[2:3], v2, v[7:8]
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v4
; SDAG-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5]
; SDAG-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v4
; SDAG-NEXT: v_cndmask_b32_e64 v3, 0, v1, s[6:7]
; SDAG-NEXT: v_cndmask_b32_e64 v2, v2, v0, s[4:5]
-; SDAG-NEXT: v_lshlrev_b64 v[0:1], v4, v[6:7]
+; SDAG-NEXT: v_lshlrev_b64 v[0:1], v4, v[7:8]
; SDAG-NEXT: v_cndmask_b32_e64 v2, 0, v2, s[6:7]
-; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v0, s[4:5]
-; SDAG-NEXT: v_cndmask_b32_e64 v11, 0, v1, s[4:5]
-; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v12, v8, 0
-; SDAG-NEXT: v_mul_lo_u32 v13, v9, v2
-; SDAG-NEXT: v_mul_lo_u32 v14, v8, v3
-; SDAG-NEXT: v_mov_b32_e32 v6, v1
-; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v8, v[6:7]
-; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v8, v2, 0
-; SDAG-NEXT: v_mov_b32_e32 v6, v4
-; SDAG-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v12, v9, v[6:7]
-; SDAG-NEXT: v_add3_u32 v3, v3, v14, v13
-; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v10, v12, v[2:3]
-; SDAG-NEXT: v_add_co_u32_e64 v3, s[4:5], v5, v7
-; SDAG-NEXT: v_addc_co_u32_e64 v4, s[4:5], 0, 0, s[4:5]
-; SDAG-NEXT: v_mul_lo_u32 v8, v10, v11
-; SDAG-NEXT: v_mul_lo_u32 v10, v10, v12
-; SDAG-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v11, v9, v[3:4]
-; SDAG-NEXT: v_add3_u32 v5, v10, v2, v8
-; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v3, v1
-; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v4, v5, s[4:5]
-; SDAG-NEXT: v_mov_b32_e32 v1, v6
+; SDAG-NEXT: v_cndmask_b32_e64 v13, 0, v0, s[4:5]
+; SDAG-NEXT: v_cndmask_b32_e64 v12, 0, v1, s[4:5]
+; SDAG-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v13, v9, 0
+; SDAG-NEXT: v_mul_lo_u32 v4, v10, v2
+; SDAG-NEXT: v_mul_lo_u32 v14, v9, v3
+; SDAG-NEXT: v_mov_b32_e32 v5, v1
+; SDAG-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v12, v9, v[5:6]
+; SDAG-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v9, v2, 0
+; SDAG-NEXT: v_mov_b32_e32 v5, v7
+; SDAG-NEXT: v_mul_lo_u32 v7, v11, v13
+; SDAG-NEXT: v_add3_u32 v3, v3, v14, v4
+; SDAG-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v13, v10, v[5:6]
+; SDAG-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v11, v13, v[2:3]
+; SDAG-NEXT: v_add_co_u32_e64 v5, s[4:5], v8, v5
+; SDAG-NEXT: v_addc_co_u32_e64 v6, s[4:5], 0, 0, s[4:5]
+; SDAG-NEXT: v_mul_lo_u32 v3, v11, v12
+; SDAG-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v10, v[5:6]
+; SDAG-NEXT: ; implicit-def: $vgpr9
+; SDAG-NEXT: v_add3_u32 v3, v7, v2, v3
+; SDAG-NEXT: v_add_co_u32_e64 v2, s[4:5], v5, v1
+; SDAG-NEXT: v_addc_co_u32_e64 v3, s[4:5], v6, v3, s[4:5]
+; SDAG-NEXT: v_mov_b32_e32 v1, v4
; SDAG-NEXT: ; implicit-def: $vgpr5_vgpr6
-; SDAG-NEXT: ; implicit-def: $vgpr6_vgpr7
-; SDAG-NEXT: ; implicit-def: $vgpr8
+; SDAG-NEXT: ; implicit-def: $vgpr7_vgpr8
; SDAG-NEXT: .LBB7_4: ; %Flow
; SDAG-NEXT: s_andn2_saveexec_b64 s[6:7], s[12:13]
; SDAG-NEXT: s_cbranch_execz .LBB7_6
; SDAG-NEXT: ; %bb.5: ; %fp-to-i-if-then12
; SDAG-NEXT: v_sub_u32_e32 v2, 0x86, v5
-; SDAG-NEXT: v_lshrrev_b64 v[0:1], v2, v[6:7]
+; SDAG-NEXT: v_lshrrev_b64 v[0:1], v2, v[7:8]
; SDAG-NEXT: v_cmp_gt_u32_e64 s[4:5], 64, v2
; SDAG-NEXT: v_cndmask_b32_e64 v0, 0, v0, s[4:5]
; SDAG-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v2
-; SDAG-NEXT: v_cndmask_b32_e64 v0, v0, v6, s[4:5]
-; SDAG-NEXT: v_mul_hi_i32_i24_e32 v1, v0, v8
+; SDAG-NEXT: v_cndmask_b32_e64 v0, v0, v7, s[4:5]
+; SDAG-NEXT: v_mul_hi_i32_i24_e32 v1, v0, v9
; SDAG-NEXT: v_ashrrev_i32_e32 v2, 31, v1
-; SDAG-NEXT: v_mul_i32_i24_e32 v0, v0, v8
+; SDAG-NEXT: v_mul_i32_i24_e32 v0, v0, v9
; SDAG-NEXT: v_mov_b32_e32 v3, v2
; SDAG-NEXT: .LBB7_6: ; %Flow1
; SDAG-NEXT: s_or_b64 exec, exec, s[6:7]
generated by cgit v1.2.3 (git 2.25.1) at 2026-05-31 14:54:19 +0000