Files
clang-p2996/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
Nicolai Hähnle 264d1136f9 AMDGPU/GISel: Introduce custom legalization of G_MUL
The generic legalizer framework is still used to reduce the problem
to scalar multiplication with the bit size a multiple of 32.

Generating optimal code sequences for big integer multiplication is
somewhat tricky and has a number of target-specific intricacies:

- The target has V_MAD_U64_U32 instructions that multiply two 32-bit
  factors and add a 64-bit accumulator. Most partial products should
  use this instruction.
- The accumulator is mapped to consecutive 32-bit GPRs, and partial-
  product multiply-adds can feed the accumulator into each other
  directly. (The register allocator's support for that is somewhat
  limited, but that only matters for 128-bit integers and larger.)
- OTOH, on some hardware, V_MAD_U64_U32 requires the accumulator
  to be stored in an even-aligned pair of GPRs. To avoid excessive
  register copies, it makes sense to compute odd partial products
  separately from even partial products (where a partial product
  src0[j0] * src1[j1] is "odd" if j0 + j1 is odd) and add both
  halves together as a final step.
- We can combine G_MUL+G_ADD into a single cascade of multiply-adds.
- The target can keep many carry-bits in flight simultaneously, so
  combining carries using G_UADDE is preferable over G_ZEXT + G_ADD.
- Not addressed by this patch: When the factors are sign-extended,
  the V_MAD_I64_I32 instruction (signed version!) can be used.

It is difficult to address these points generically:

1) Finding matching pairs of G_MUL and G_UMULH to find a wide
multiply is expensive. We could add a G_UMUL_LOHI generic instruction
and conditionally use that in the generic legalizer, but by itself
this wouldn't allow us to use the accumulation capability of
V_MAD_U64_U32. One could attempt to find matching G_ADD + G_UADDE
post-legalization, but this is also expensive.

2) Similarly, making sense of the legalization outcome of a wide
pre-legalization G_MUL+G_ADD pair is extremely expensive.

3) How could the generic legalizer possibly deal with the
particular idiosyncracy of "odd" vs. "even" partial products.

All this points in the direction of directly emitting an ideal code
sequence during legalization, but the generic legalizer should not
be burdened with such overly target-specific concerns. Hence, a
custom legalization.

Note that the implemented approach is different from that used by
SelectionDAG because narrowing of scalars works differently in
general. SelectionDAG iteratively cuts wide scalars into low and
high halves until a legal size is reached. By contrast, GlobalISel
does the narrowing in a single shot, which should be better for
compile-time and for the quality of the generated code.

This patch leaves three gaps open:

1. When the factors are uniform, we should execute the multiplication on
   the SALU. Register bank mapping already ensures this.

   However, the resulting code sequence is not optimal because it doesn't
   fully use the carry-in capabilities of S_ADDC_U32. (V_MAD_U64_U32
   doesn't have a carry-in.) It is very difficult to fix this after the
   fact, so we should really use a different legalization sequence in
   this case. Unfortunately, we don't have a divergence analysis and so
   cannot make that choice.

   (This only matters for 128-bit integers and larger.)

2. Avoid unnecessary multiplies when sources are known to be zero- or
   sign-extended. The challenge is that the legalizer does not currently
   have access to GISelKnownBits.

3. When the G_MUL is followed by a G_ADD, we should consider combining
   the two instructions into a single multiply-add sequence, to utilize
   the accumulator of V_MAD_U64_U32 fully. (Unless the multiply has
   multiple uses and the implied duplication of the multiply is an
   overall negative). However, this is also not true when the factors
   are uniform: in that case, it is generally better to *not* combine
   the two operations, so that the multiply can be done on the SALU.

   Again, we don't have a divergence analysis available and so cannot
   make an informed choice.

Differential Revision: https://reviews.llvm.org/D124844
2022-06-09 13:38:56 +02:00

3470 lines
157 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel -amdgpu-codegenprepare-disable-idiv-expansion=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=hawaii -denormal-fp-math-f32=preserve-sign < %s | FileCheck -check-prefixes=CHECK,GISEL %s
; RUN: llc -global-isel -amdgpu-codegenprepare-disable-idiv-expansion=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=hawaii -denormal-fp-math-f32=preserve-sign < %s | FileCheck -check-prefixes=CHECK,CGP %s
; The same 32-bit expansion is implemented in the legalizer and in AMDGPUCodeGenPrepare.
define i64 @v_sdiv_i64(i64 %num, i64 %den) {
; CHECK-LABEL: v_sdiv_i64:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_mov_b32_e32 v5, v1
; CHECK-NEXT: v_mov_b32_e32 v4, v0
; CHECK-NEXT: v_or_b32_e32 v1, v5, v3
; CHECK-NEXT: v_mov_b32_e32 v0, 0
; CHECK-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1
; CHECK-NEXT: s_and_saveexec_b64 s[4:5], vcc
; CHECK-NEXT: s_xor_b64 s[6:7], exec, s[4:5]
; CHECK-NEXT: s_cbranch_execz .LBB0_2
; CHECK-NEXT: ; %bb.1:
; CHECK-NEXT: v_ashrrev_i32_e32 v0, 31, v3
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v2, v0
; CHECK-NEXT: v_addc_u32_e32 v2, vcc, v3, v0, vcc
; CHECK-NEXT: v_xor_b32_e32 v6, v1, v0
; CHECK-NEXT: v_xor_b32_e32 v7, v2, v0
; CHECK-NEXT: v_cvt_f32_u32_e32 v1, v6
; CHECK-NEXT: v_cvt_f32_u32_e32 v2, v7
; CHECK-NEXT: v_sub_i32_e32 v9, vcc, 0, v6
; CHECK-NEXT: v_subb_u32_e32 v10, vcc, 0, v7, vcc
; CHECK-NEXT: v_mac_f32_e32 v1, 0x4f800000, v2
; CHECK-NEXT: v_rcp_iflag_f32_e32 v1, v1
; CHECK-NEXT: v_mul_f32_e32 v1, 0x5f7ffffc, v1
; CHECK-NEXT: v_mul_f32_e32 v2, 0x2f800000, v1
; CHECK-NEXT: v_trunc_f32_e32 v3, v2
; CHECK-NEXT: v_mac_f32_e32 v1, 0xcf800000, v3
; CHECK-NEXT: v_cvt_u32_f32_e32 v8, v1
; CHECK-NEXT: v_cvt_u32_f32_e32 v11, v3
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v9, v8, 0
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v9, v11, v[2:3]
; CHECK-NEXT: v_mul_hi_u32 v12, v8, v1
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v10, v8, v[2:3]
; CHECK-NEXT: v_mul_lo_u32 v3, v11, v1
; CHECK-NEXT: v_mul_hi_u32 v1, v11, v1
; CHECK-NEXT: v_mul_lo_u32 v13, v8, v2
; CHECK-NEXT: v_mul_lo_u32 v14, v11, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v13
; CHECK-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v12
; CHECK-NEXT: v_mul_hi_u32 v12, v8, v2
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v13, v3
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v14, v1
; CHECK-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v12
; CHECK-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v12, vcc, v13, v12
; CHECK-NEXT: v_mul_hi_u32 v2, v11, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v3
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v12, v3
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CHECK-NEXT: v_add_i32_e32 v8, vcc, v8, v1
; CHECK-NEXT: v_addc_u32_e32 v11, vcc, v11, v2, vcc
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v9, v8, 0
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v9, v11, v[2:3]
; CHECK-NEXT: v_ashrrev_i32_e32 v9, 31, v5
; CHECK-NEXT: v_mul_hi_u32 v12, v8, v1
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v10, v8, v[2:3]
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v4, v9
; CHECK-NEXT: v_addc_u32_e32 v4, vcc, v5, v9, vcc
; CHECK-NEXT: v_xor_b32_e32 v5, v3, v9
; CHECK-NEXT: v_mul_lo_u32 v3, v11, v1
; CHECK-NEXT: v_mul_lo_u32 v10, v8, v2
; CHECK-NEXT: v_mul_hi_u32 v1, v11, v1
; CHECK-NEXT: v_xor_b32_e32 v4, v4, v9
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v10
; CHECK-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v12
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v12, v11, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v10, v3
; CHECK-NEXT: v_mul_hi_u32 v10, v8, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v12, v1
; CHECK-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v10
; CHECK-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v10, vcc, v12, v10
; CHECK-NEXT: v_mul_hi_u32 v2, v11, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v3
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v10, v3
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v8, v1
; CHECK-NEXT: v_addc_u32_e32 v2, vcc, v11, v2, vcc
; CHECK-NEXT: v_mul_lo_u32 v3, v4, v1
; CHECK-NEXT: v_mul_lo_u32 v8, v5, v2
; CHECK-NEXT: v_mul_hi_u32 v10, v5, v1
; CHECK-NEXT: v_mul_hi_u32 v1, v4, v1
; CHECK-NEXT: v_mul_hi_u32 v11, v4, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v10
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v10, v4, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v8, v3
; CHECK-NEXT: v_mul_hi_u32 v8, v5, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; CHECK-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v8, vcc, v10, v8
; CHECK-NEXT: v_add_i32_e32 v10, vcc, v1, v3
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v6, v10, 0
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v8, v3
; CHECK-NEXT: v_add_i32_e32 v8, vcc, v11, v3
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v6, v8, v[2:3]
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, v5, v1
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v7, v10, v[2:3]
; CHECK-NEXT: v_subb_u32_e64 v3, s[4:5], v4, v2, vcc
; CHECK-NEXT: v_sub_i32_e64 v2, s[4:5], v4, v2
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v3, v7
; CHECK-NEXT: v_subb_u32_e32 v2, vcc, v2, v7, vcc
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, -1, s[4:5]
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v6
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, v1, v6
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, -1, s[4:5]
; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], v3, v7
; CHECK-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
; CHECK-NEXT: v_cndmask_b32_e64 v3, v4, v5, s[4:5]
; CHECK-NEXT: v_add_i32_e32 v4, vcc, 1, v10
; CHECK-NEXT: v_addc_u32_e32 v5, vcc, 0, v8, vcc
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v2, v7
; CHECK-NEXT: v_cndmask_b32_e64 v11, 0, -1, vcc
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v1, v6
; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, -1, vcc
; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, v2, v7
; CHECK-NEXT: v_cndmask_b32_e32 v1, v11, v1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, 1, v4
; CHECK-NEXT: v_addc_u32_e32 v6, vcc, 0, v5, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; CHECK-NEXT: v_cndmask_b32_e32 v1, v4, v2, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v2, v5, v6, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; CHECK-NEXT: v_cndmask_b32_e32 v1, v10, v1, vcc
; CHECK-NEXT: v_xor_b32_e32 v3, v9, v0
; CHECK-NEXT: v_cndmask_b32_e32 v2, v8, v2, vcc
; CHECK-NEXT: v_xor_b32_e32 v0, v1, v3
; CHECK-NEXT: v_xor_b32_e32 v1, v2, v3
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
; CHECK-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
; CHECK-NEXT: ; implicit-def: $vgpr2
; CHECK-NEXT: ; implicit-def: $vgpr4
; CHECK-NEXT: .LBB0_2: ; %Flow
; CHECK-NEXT: s_or_saveexec_b64 s[6:7], s[6:7]
; CHECK-NEXT: s_xor_b64 exec, exec, s[6:7]
; CHECK-NEXT: s_cbranch_execz .LBB0_4
; CHECK-NEXT: ; %bb.3:
; CHECK-NEXT: v_cvt_f32_u32_e32 v0, v2
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, 0, v2
; CHECK-NEXT: v_rcp_iflag_f32_e32 v0, v0
; CHECK-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0
; CHECK-NEXT: v_cvt_u32_f32_e32 v0, v0
; CHECK-NEXT: v_mul_lo_u32 v1, v1, v0
; CHECK-NEXT: v_mul_hi_u32 v1, v0, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CHECK-NEXT: v_mul_hi_u32 v0, v4, v0
; CHECK-NEXT: v_mul_lo_u32 v1, v0, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, 1, v0
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, v4, v1
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v1, v2
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; CHECK-NEXT: v_sub_i32_e64 v3, s[4:5], v1, v2
; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, 1, v0
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v1, v2
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; CHECK-NEXT: v_mov_b32_e32 v1, 0
; CHECK-NEXT: .LBB0_4:
; CHECK-NEXT: s_or_b64 exec, exec, s[6:7]
; CHECK-NEXT: s_setpc_b64 s[30:31]
%result = sdiv i64 %num, %den
ret i64 %result
}
; FIXME: This is a workaround for not handling uniform VGPR case.
declare i32 @llvm.amdgcn.readfirstlane(i32)
define amdgpu_ps i64 @s_sdiv_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-LABEL: s_sdiv_i64:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_or_b64 s[6:7], s[2:3], s[4:5]
; CHECK-NEXT: s_mov_b32 s0, 0
; CHECK-NEXT: s_mov_b32 s1, -1
; CHECK-NEXT: s_and_b64 s[6:7], s[6:7], s[0:1]
; CHECK-NEXT: v_cmp_ne_u64_e64 vcc, s[6:7], 0
; CHECK-NEXT: s_cbranch_vccz .LBB1_2
; CHECK-NEXT: ; %bb.1:
; CHECK-NEXT: s_ashr_i32 s6, s3, 31
; CHECK-NEXT: s_ashr_i32 s8, s5, 31
; CHECK-NEXT: s_add_u32 s0, s2, s6
; CHECK-NEXT: s_addc_u32 s1, s3, s6
; CHECK-NEXT: s_add_u32 s10, s4, s8
; CHECK-NEXT: s_mov_b32 s9, s8
; CHECK-NEXT: s_addc_u32 s11, s5, s8
; CHECK-NEXT: s_xor_b64 s[10:11], s[10:11], s[8:9]
; CHECK-NEXT: v_cvt_f32_u32_e32 v0, s10
; CHECK-NEXT: v_cvt_f32_u32_e32 v1, s11
; CHECK-NEXT: s_mov_b32 s7, s6
; CHECK-NEXT: s_xor_b64 s[12:13], s[0:1], s[6:7]
; CHECK-NEXT: s_sub_u32 s3, 0, s10
; CHECK-NEXT: v_mac_f32_e32 v0, 0x4f800000, v1
; CHECK-NEXT: v_rcp_iflag_f32_e32 v0, v0
; CHECK-NEXT: s_subb_u32 s5, 0, s11
; CHECK-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v0
; CHECK-NEXT: v_mul_f32_e32 v1, 0x2f800000, v0
; CHECK-NEXT: v_trunc_f32_e32 v2, v1
; CHECK-NEXT: v_mac_f32_e32 v0, 0xcf800000, v2
; CHECK-NEXT: v_cvt_u32_f32_e32 v3, v0
; CHECK-NEXT: v_cvt_u32_f32_e32 v4, v2
; CHECK-NEXT: v_mad_u64_u32 v[0:1], s[0:1], s3, v3, 0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[0:1], s3, v4, v[1:2]
; CHECK-NEXT: v_mul_hi_u32 v5, v3, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[0:1], s5, v3, v[1:2]
; CHECK-NEXT: v_mul_lo_u32 v2, v4, v0
; CHECK-NEXT: v_mul_hi_u32 v0, v4, v0
; CHECK-NEXT: v_mul_lo_u32 v6, v3, v1
; CHECK-NEXT: v_mul_lo_u32 v7, v4, v1
; CHECK-NEXT: v_mul_hi_u32 v8, v3, v1
; CHECK-NEXT: v_mul_hi_u32 v1, v4, v1
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v6
; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v5
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v6, v2
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v7, v0
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v8
; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v0
; CHECK-NEXT: v_addc_u32_e32 v4, vcc, v4, v1, vcc
; CHECK-NEXT: v_mad_u64_u32 v[0:1], s[0:1], s3, v3, 0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[0:1], s3, v4, v[1:2]
; CHECK-NEXT: v_mul_hi_u32 v6, v3, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[0:1], s5, v3, v[1:2]
; CHECK-NEXT: v_mul_lo_u32 v2, v4, v0
; CHECK-NEXT: v_mul_hi_u32 v0, v4, v0
; CHECK-NEXT: v_mul_lo_u32 v5, v3, v1
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v5
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v6
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v6, v4, v1
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; CHECK-NEXT: v_mul_hi_u32 v5, v3, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v6, v0
; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v5
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; CHECK-NEXT: v_mul_hi_u32 v1, v4, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v2
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; CHECK-NEXT: v_addc_u32_e32 v1, vcc, v4, v1, vcc
; CHECK-NEXT: v_mul_lo_u32 v2, s13, v0
; CHECK-NEXT: v_mul_lo_u32 v3, s12, v1
; CHECK-NEXT: v_mul_hi_u32 v4, s12, v0
; CHECK-NEXT: v_mul_hi_u32 v0, s13, v0
; CHECK-NEXT: v_mul_hi_u32 v5, s13, v1
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v4
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v4, s13, v1
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CHECK-NEXT: v_mul_hi_u32 v3, s12, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v4, v0
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v3
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v4, v3
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v0, v2
; CHECK-NEXT: v_mad_u64_u32 v[0:1], s[0:1], s10, v4, 0
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[0:1], s10, v2, v[1:2]
; CHECK-NEXT: v_mov_b32_e32 v5, s13
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, s12, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[0:1], s11, v4, v[1:2]
; CHECK-NEXT: v_mov_b32_e32 v3, s11
; CHECK-NEXT: v_subb_u32_e64 v2, s[0:1], v5, v1, vcc
; CHECK-NEXT: v_sub_i32_e64 v1, s[0:1], s13, v1
; CHECK-NEXT: v_cmp_le_u32_e64 s[0:1], s11, v2
; CHECK-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, -1, s[0:1]
; CHECK-NEXT: v_cmp_le_u32_e64 s[0:1], s10, v0
; CHECK-NEXT: v_subrev_i32_e32 v0, vcc, s10, v0
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, 1, v4
; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[0:1]
; CHECK-NEXT: v_cmp_eq_u32_e64 s[0:1], s11, v2
; CHECK-NEXT: v_cmp_le_u32_e32 vcc, s11, v1
; CHECK-NEXT: v_cndmask_b32_e64 v2, v5, v6, s[0:1]
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, -1, vcc
; CHECK-NEXT: v_cmp_le_u32_e32 vcc, s10, v0
; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, s11, v1
; CHECK-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, 1, v3
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0
; CHECK-NEXT: v_cndmask_b32_e32 v0, v3, v1, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; CHECK-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc
; CHECK-NEXT: s_xor_b64 s[0:1], s[6:7], s[8:9]
; CHECK-NEXT: v_xor_b32_e32 v0, s0, v0
; CHECK-NEXT: v_subrev_i32_e32 v0, vcc, s0, v0
; CHECK-NEXT: s_mov_b32 s1, 0
; CHECK-NEXT: s_branch .LBB1_3
; CHECK-NEXT: .LBB1_2:
; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1
; CHECK-NEXT: .LBB1_3: ; %Flow
; CHECK-NEXT: s_xor_b32 s0, s1, -1
; CHECK-NEXT: s_and_b32 s0, s0, 1
; CHECK-NEXT: s_cmp_lg_u32 s0, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB1_5
; CHECK-NEXT: ; %bb.4:
; CHECK-NEXT: v_cvt_f32_u32_e32 v0, s4
; CHECK-NEXT: s_sub_i32 s0, 0, s4
; CHECK-NEXT: v_rcp_iflag_f32_e32 v0, v0
; CHECK-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0
; CHECK-NEXT: v_cvt_u32_f32_e32 v0, v0
; CHECK-NEXT: v_mul_lo_u32 v1, s0, v0
; CHECK-NEXT: v_mul_hi_u32 v1, v0, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CHECK-NEXT: v_mul_hi_u32 v0, s2, v0
; CHECK-NEXT: v_mul_lo_u32 v1, v0, s4
; CHECK-NEXT: v_add_i32_e32 v2, vcc, 1, v0
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, s2, v1
; CHECK-NEXT: v_cmp_le_u32_e32 vcc, s4, v1
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; CHECK-NEXT: v_subrev_i32_e64 v2, s[0:1], s4, v1
; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, 1, v0
; CHECK-NEXT: v_cmp_le_u32_e32 vcc, s4, v1
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; CHECK-NEXT: .LBB1_5:
; CHECK-NEXT: v_readfirstlane_b32 s0, v0
; CHECK-NEXT: s_mov_b32 s1, s0
; CHECK-NEXT: ; return to shader part epilog
%result = sdiv i64 %num, %den
%cast = bitcast i64 %result to <2 x i32>
%elt.0 = extractelement <2 x i32> %cast, i32 0
%elt.1 = extractelement <2 x i32> %cast, i32 1
%res.0 = call i32 @llvm.amdgcn.readfirstlane(i32 %elt.0)
%res.1 = call i32 @llvm.amdgcn.readfirstlane(i32 %elt.1)
%ins.0 = insertelement <2 x i32> undef, i32 %res.0, i32 0
%ins.1 = insertelement <2 x i32> %ins.0, i32 %res.0, i32 1
%cast.back = bitcast <2 x i32> %ins.1 to i64
ret i64 %cast.back
}
define <2 x i64> @v_sdiv_v2i64(<2 x i64> %num, <2 x i64> %den) {
; GISEL-LABEL: v_sdiv_v2i64:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_ashrrev_i32_e32 v8, 31, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v5, v8, vcc
; GISEL-NEXT: v_xor_b32_e32 v5, v4, v8
; GISEL-NEXT: v_xor_b32_e32 v4, v9, v8
; GISEL-NEXT: v_cvt_f32_u32_e32 v9, v5
; GISEL-NEXT: v_cvt_f32_u32_e32 v10, v4
; GISEL-NEXT: v_sub_i32_e32 v13, vcc, 0, v5
; GISEL-NEXT: v_subb_u32_e32 v14, vcc, 0, v4, vcc
; GISEL-NEXT: v_mac_f32_e32 v9, 0x4f800000, v10
; GISEL-NEXT: v_rcp_iflag_f32_e32 v9, v9
; GISEL-NEXT: v_mul_f32_e32 v9, 0x5f7ffffc, v9
; GISEL-NEXT: v_mul_f32_e32 v10, 0x2f800000, v9
; GISEL-NEXT: v_trunc_f32_e32 v11, v10
; GISEL-NEXT: v_mac_f32_e32 v9, 0xcf800000, v11
; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v9
; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v11
; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[4:5], v13, v12, 0
; GISEL-NEXT: v_mad_u64_u32 v[10:11], s[4:5], v13, v15, v[10:11]
; GISEL-NEXT: v_mul_hi_u32 v17, v12, v9
; GISEL-NEXT: v_mad_u64_u32 v[10:11], s[4:5], v14, v12, v[10:11]
; GISEL-NEXT: v_mul_lo_u32 v11, v15, v9
; GISEL-NEXT: v_mul_hi_u32 v9, v15, v9
; GISEL-NEXT: v_mul_lo_u32 v16, v12, v10
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v11, v16
; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v11, v17
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v17, v15, v10
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v16, v11
; GISEL-NEXT: v_mul_hi_u32 v16, v12, v10
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v17, v9
; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v16
; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v16, vcc, v17, v16
; GISEL-NEXT: v_mul_hi_u32 v10, v15, v10
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v11
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v16, v11
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
; GISEL-NEXT: v_add_i32_e32 v16, vcc, v12, v9
; GISEL-NEXT: v_addc_u32_e32 v15, vcc, v15, v10, vcc
; GISEL-NEXT: v_mad_u64_u32 v[10:11], s[4:5], v13, v16, 0
; GISEL-NEXT: v_mov_b32_e32 v9, v11
; GISEL-NEXT: v_mad_u64_u32 v[11:12], s[4:5], v13, v15, v[9:10]
; GISEL-NEXT: v_ashrrev_i32_e32 v9, 31, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v9
; GISEL-NEXT: v_mad_u64_u32 v[11:12], s[4:5], v14, v16, v[11:12]
; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v1, v9, vcc
; GISEL-NEXT: v_xor_b32_e32 v13, v0, v9
; GISEL-NEXT: v_mul_lo_u32 v0, v15, v10
; GISEL-NEXT: v_mul_lo_u32 v12, v16, v11
; GISEL-NEXT: v_xor_b32_e32 v14, v1, v9
; GISEL-NEXT: v_mul_hi_u32 v1, v16, v10
; GISEL-NEXT: v_mul_hi_u32 v10, v15, v10
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v12
; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v1, v15, v11
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v12, v0
; GISEL-NEXT: v_mul_hi_u32 v12, v16, v11
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v12
; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v12
; GISEL-NEXT: v_mul_hi_u32 v11, v15, v11
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v11, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v16, v0
; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v15, v1, vcc
; GISEL-NEXT: v_mul_lo_u32 v10, v14, v0
; GISEL-NEXT: v_mul_lo_u32 v11, v13, v1
; GISEL-NEXT: v_mul_hi_u32 v12, v13, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v14, v0
; GISEL-NEXT: v_xor_b32_e32 v8, v9, v8
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v12
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v12, v14, v1
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10
; GISEL-NEXT: v_mul_hi_u32 v11, v13, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v12, v0
; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v11
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v12, v11
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v0, v10
; GISEL-NEXT: v_mul_hi_u32 v12, v14, v1
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v5, v15, 0
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10
; GISEL-NEXT: v_add_i32_e32 v16, vcc, v12, v10
; GISEL-NEXT: v_mad_u64_u32 v[10:11], s[4:5], v5, v16, v[1:2]
; GISEL-NEXT: v_mad_u64_u32 v[11:12], s[4:5], v4, v15, v[10:11]
; GISEL-NEXT: v_ashrrev_i32_e32 v10, 31, v7
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v6, v10
; GISEL-NEXT: v_addc_u32_e32 v6, vcc, v7, v10, vcc
; GISEL-NEXT: v_xor_b32_e32 v7, v1, v10
; GISEL-NEXT: v_xor_b32_e32 v12, v6, v10
; GISEL-NEXT: v_cvt_f32_u32_e32 v1, v7
; GISEL-NEXT: v_cvt_f32_u32_e32 v6, v12
; GISEL-NEXT: v_sub_i32_e32 v13, vcc, v13, v0
; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], v14, v11
; GISEL-NEXT: v_mac_f32_e32 v1, 0x4f800000, v6
; GISEL-NEXT: v_rcp_iflag_f32_e32 v1, v1
; GISEL-NEXT: v_subb_u32_e64 v17, s[4:5], v14, v11, vcc
; GISEL-NEXT: v_subb_u32_e32 v6, vcc, v0, v4, vcc
; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v1
; GISEL-NEXT: v_mul_f32_e32 v1, 0x2f800000, v0
; GISEL-NEXT: v_trunc_f32_e32 v11, v1
; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v11
; GISEL-NEXT: v_cvt_u32_f32_e32 v14, v0
; GISEL-NEXT: v_sub_i32_e32 v18, vcc, 0, v7
; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v11
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[6:7], v18, v14, 0
; GISEL-NEXT: v_subb_u32_e32 v19, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v13, v5
; GISEL-NEXT: v_sub_i32_e32 v13, vcc, v13, v5
; GISEL-NEXT: v_subbrev_u32_e32 v20, vcc, 0, v6, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v13, v5
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[6:7], v18, v11, v[1:2]
; GISEL-NEXT: v_mul_lo_u32 v1, v11, v0
; GISEL-NEXT: v_cmp_ge_u32_e64 s[8:9], v20, v4
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[6:7], v19, v14, v[5:6]
; GISEL-NEXT: v_cndmask_b32_e64 v21, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], v17, v4
; GISEL-NEXT: v_mul_lo_u32 v6, v14, v5
; GISEL-NEXT: v_add_i32_e64 v1, s[6:7], v1, v6
; GISEL-NEXT: v_mul_hi_u32 v6, v14, v0
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[6:7]
; GISEL-NEXT: v_mul_hi_u32 v0, v11, v0
; GISEL-NEXT: v_add_i32_e64 v1, s[6:7], v1, v6
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, -1, s[8:9]
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, v20, v4
; GISEL-NEXT: v_cmp_ge_u32_e64 s[8:9], v17, v4
; GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v6, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v15
; GISEL-NEXT: v_cndmask_b32_e64 v20, 0, -1, s[8:9]
; GISEL-NEXT: v_addc_u32_e32 v17, vcc, 0, v16, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v4, v20, v21, s[4:5]
; GISEL-NEXT: v_add_i32_e32 v20, vcc, 1, v6
; GISEL-NEXT: v_addc_u32_e32 v21, vcc, 0, v17, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, s[6:7]
; GISEL-NEXT: v_cndmask_b32_e32 v6, v6, v20, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v17, v17, v21, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v13, v1
; GISEL-NEXT: v_mul_lo_u32 v13, v11, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v13, v0
; GISEL-NEXT: v_mul_hi_u32 v13, v14, v5
; GISEL-NEXT: v_cndmask_b32_e64 v20, 0, 1, vcc
; GISEL-NEXT: v_mul_hi_u32 v5, v11, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v13
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v20, v13
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v13, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v5, v1
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v14, v0
; GISEL-NEXT: v_addc_u32_e32 v11, vcc, v11, v1, vcc
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v18, v13, 0
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; GISEL-NEXT: v_cndmask_b32_e32 v6, v15, v6, vcc
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v18, v11, v[1:2]
; GISEL-NEXT: v_xor_b32_e32 v1, v6, v8
; GISEL-NEXT: v_ashrrev_i32_e32 v6, 31, v3
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v19, v13, v[4:5]
; GISEL-NEXT: v_cndmask_b32_e32 v14, v16, v17, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v6
; GISEL-NEXT: v_addc_u32_e32 v3, vcc, v3, v6, vcc
; GISEL-NEXT: v_xor_b32_e32 v5, v2, v6
; GISEL-NEXT: v_mul_lo_u32 v2, v11, v0
; GISEL-NEXT: v_mul_lo_u32 v9, v13, v4
; GISEL-NEXT: v_xor_b32_e32 v15, v3, v6
; GISEL-NEXT: v_mul_hi_u32 v3, v13, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v11, v0
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v9
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v3, v11, v4
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v9, v2
; GISEL-NEXT: v_mul_hi_u32 v9, v13, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v9
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9
; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v4, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v13, v0
; GISEL-NEXT: v_addc_u32_e32 v2, vcc, v11, v2, vcc
; GISEL-NEXT: v_mul_lo_u32 v3, v15, v0
; GISEL-NEXT: v_mul_lo_u32 v4, v5, v2
; GISEL-NEXT: v_mul_hi_u32 v11, v5, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v15, v0
; GISEL-NEXT: v_mul_hi_u32 v13, v15, v2
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v4
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v11
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v11, v15, v2
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3
; GISEL-NEXT: v_mul_hi_u32 v4, v5, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v11, v0
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v11, v4
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v0, v3
; GISEL-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v7, v11, 0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v3
; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v7, v13, v[0:1]
; GISEL-NEXT: v_xor_b32_e32 v9, v14, v8
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v1, v8
; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v12, v11, v[3:4]
; GISEL-NEXT: v_subb_u32_e32 v1, vcc, v9, v8, vcc
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v5, v2
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], v15, v3, vcc
; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v15, v3
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v4, v12
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v12, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v7
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v7
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], v4, v12
; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v4, v5, v8, s[4:5]
; GISEL-NEXT: v_add_i32_e32 v5, vcc, 1, v11
; GISEL-NEXT: v_addc_u32_e32 v8, vcc, 0, v13, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v3, v12
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, -1, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v2, v7
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, v3, v12
; GISEL-NEXT: v_cndmask_b32_e32 v2, v9, v2, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v5
; GISEL-NEXT: v_addc_u32_e32 v7, vcc, 0, v8, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; GISEL-NEXT: v_cndmask_b32_e32 v2, v5, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v8, v7, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; GISEL-NEXT: v_cndmask_b32_e32 v2, v11, v2, vcc
; GISEL-NEXT: v_xor_b32_e32 v4, v6, v10
; GISEL-NEXT: v_cndmask_b32_e32 v3, v13, v3, vcc
; GISEL-NEXT: v_xor_b32_e32 v2, v2, v4
; GISEL-NEXT: v_xor_b32_e32 v3, v3, v4
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v4
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v4, vcc
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_v2i64:
; CGP: ; %bb.0:
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CGP-NEXT: v_mov_b32_e32 v11, v1
; CGP-NEXT: v_mov_b32_e32 v10, v0
; CGP-NEXT: v_or_b32_e32 v1, v11, v5
; CGP-NEXT: v_mov_b32_e32 v0, 0
; CGP-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; CGP-NEXT: v_mov_b32_e32 v8, v2
; CGP-NEXT: v_mov_b32_e32 v9, v3
; CGP-NEXT: ; implicit-def: $vgpr0_vgpr1
; CGP-NEXT: s_and_saveexec_b64 s[4:5], vcc
; CGP-NEXT: s_xor_b64 s[6:7], exec, s[4:5]
; CGP-NEXT: s_cbranch_execz .LBB2_2
; CGP-NEXT: ; %bb.1:
; CGP-NEXT: v_ashrrev_i32_e32 v0, 31, v5
; CGP-NEXT: v_add_i32_e32 v1, vcc, v4, v0
; CGP-NEXT: v_addc_u32_e32 v2, vcc, v5, v0, vcc
; CGP-NEXT: v_xor_b32_e32 v4, v1, v0
; CGP-NEXT: v_xor_b32_e32 v5, v2, v0
; CGP-NEXT: v_cvt_f32_u32_e32 v1, v4
; CGP-NEXT: v_cvt_f32_u32_e32 v2, v5
; CGP-NEXT: v_sub_i32_e32 v13, vcc, 0, v4
; CGP-NEXT: v_subb_u32_e32 v14, vcc, 0, v5, vcc
; CGP-NEXT: v_mac_f32_e32 v1, 0x4f800000, v2
; CGP-NEXT: v_rcp_iflag_f32_e32 v1, v1
; CGP-NEXT: v_mul_f32_e32 v1, 0x5f7ffffc, v1
; CGP-NEXT: v_mul_f32_e32 v2, 0x2f800000, v1
; CGP-NEXT: v_trunc_f32_e32 v3, v2
; CGP-NEXT: v_mac_f32_e32 v1, 0xcf800000, v3
; CGP-NEXT: v_cvt_u32_f32_e32 v12, v1
; CGP-NEXT: v_cvt_u32_f32_e32 v15, v3
; CGP-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v13, v12, 0
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v13, v15, v[2:3]
; CGP-NEXT: v_mul_hi_u32 v16, v12, v1
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v14, v12, v[2:3]
; CGP-NEXT: v_mul_lo_u32 v3, v15, v1
; CGP-NEXT: v_mul_hi_u32 v1, v15, v1
; CGP-NEXT: v_mul_lo_u32 v17, v12, v2
; CGP-NEXT: v_mul_lo_u32 v18, v15, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v17
; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v16
; CGP-NEXT: v_mul_hi_u32 v16, v12, v2
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v17, v3
; CGP-NEXT: v_add_i32_e32 v1, vcc, v18, v1
; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v16
; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v16, vcc, v17, v16
; CGP-NEXT: v_mul_hi_u32 v2, v15, v2
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v3
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v16, v3
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_add_i32_e32 v12, vcc, v12, v1
; CGP-NEXT: v_addc_u32_e32 v15, vcc, v15, v2, vcc
; CGP-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v13, v12, 0
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v13, v15, v[2:3]
; CGP-NEXT: v_ashrrev_i32_e32 v13, 31, v11
; CGP-NEXT: v_mul_hi_u32 v16, v12, v1
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v14, v12, v[2:3]
; CGP-NEXT: v_add_i32_e32 v3, vcc, v10, v13
; CGP-NEXT: v_addc_u32_e32 v10, vcc, v11, v13, vcc
; CGP-NEXT: v_xor_b32_e32 v11, v3, v13
; CGP-NEXT: v_mul_lo_u32 v3, v15, v1
; CGP-NEXT: v_mul_lo_u32 v14, v12, v2
; CGP-NEXT: v_mul_hi_u32 v1, v15, v1
; CGP-NEXT: v_xor_b32_e32 v10, v10, v13
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v14
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v16
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v16, v15, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v14, v3
; CGP-NEXT: v_mul_hi_u32 v14, v12, v2
; CGP-NEXT: v_add_i32_e32 v1, vcc, v16, v1
; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v14
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v14, vcc, v16, v14
; CGP-NEXT: v_mul_hi_u32 v2, v15, v2
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v3
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v14, v3
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_add_i32_e32 v1, vcc, v12, v1
; CGP-NEXT: v_addc_u32_e32 v2, vcc, v15, v2, vcc
; CGP-NEXT: v_mul_lo_u32 v3, v10, v1
; CGP-NEXT: v_mul_lo_u32 v12, v11, v2
; CGP-NEXT: v_mul_hi_u32 v14, v11, v1
; CGP-NEXT: v_mul_hi_u32 v1, v10, v1
; CGP-NEXT: v_mul_hi_u32 v15, v10, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v12
; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v14
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v14, v10, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v12, v3
; CGP-NEXT: v_mul_hi_u32 v12, v11, v2
; CGP-NEXT: v_add_i32_e32 v1, vcc, v14, v1
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v12
; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v12, vcc, v14, v12
; CGP-NEXT: v_add_i32_e32 v14, vcc, v1, v3
; CGP-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v4, v14, 0
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v12, v3
; CGP-NEXT: v_add_i32_e32 v12, vcc, v15, v3
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v4, v12, v[2:3]
; CGP-NEXT: v_sub_i32_e32 v1, vcc, v11, v1
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v5, v14, v[2:3]
; CGP-NEXT: v_subb_u32_e64 v3, s[4:5], v10, v2, vcc
; CGP-NEXT: v_sub_i32_e64 v2, s[4:5], v10, v2
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v3, v5
; CGP-NEXT: v_subb_u32_e32 v2, vcc, v2, v5, vcc
; CGP-NEXT: v_cndmask_b32_e64 v10, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v4
; CGP-NEXT: v_sub_i32_e32 v1, vcc, v1, v4
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], v3, v5
; CGP-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
; CGP-NEXT: v_cndmask_b32_e64 v3, v10, v11, s[4:5]
; CGP-NEXT: v_add_i32_e32 v10, vcc, 1, v14
; CGP-NEXT: v_addc_u32_e32 v11, vcc, 0, v12, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v15, 0, -1, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v4
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e32 v1, v15, v1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, 1, v10
; CGP-NEXT: v_addc_u32_e32 v4, vcc, 0, v11, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; CGP-NEXT: v_cndmask_b32_e32 v1, v10, v2, vcc
; CGP-NEXT: v_cndmask_b32_e32 v2, v11, v4, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; CGP-NEXT: v_cndmask_b32_e32 v1, v14, v1, vcc
; CGP-NEXT: v_xor_b32_e32 v3, v13, v0
; CGP-NEXT: v_cndmask_b32_e32 v2, v12, v2, vcc
; CGP-NEXT: v_xor_b32_e32 v0, v1, v3
; CGP-NEXT: v_xor_b32_e32 v1, v2, v3
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
; CGP-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
; CGP-NEXT: ; implicit-def: $vgpr4
; CGP-NEXT: ; implicit-def: $vgpr10
; CGP-NEXT: .LBB2_2: ; %Flow1
; CGP-NEXT: s_or_saveexec_b64 s[6:7], s[6:7]
; CGP-NEXT: s_xor_b64 exec, exec, s[6:7]
; CGP-NEXT: s_cbranch_execz .LBB2_4
; CGP-NEXT: ; %bb.3:
; CGP-NEXT: v_cvt_f32_u32_e32 v0, v4
; CGP-NEXT: v_sub_i32_e32 v1, vcc, 0, v4
; CGP-NEXT: v_rcp_iflag_f32_e32 v0, v0
; CGP-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0
; CGP-NEXT: v_cvt_u32_f32_e32 v0, v0
; CGP-NEXT: v_mul_lo_u32 v1, v1, v0
; CGP-NEXT: v_mul_hi_u32 v1, v0, v1
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CGP-NEXT: v_mul_hi_u32 v0, v10, v0
; CGP-NEXT: v_mul_lo_u32 v1, v0, v4
; CGP-NEXT: v_add_i32_e32 v2, vcc, 1, v0
; CGP-NEXT: v_sub_i32_e32 v1, vcc, v10, v1
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v4
; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; CGP-NEXT: v_sub_i32_e64 v2, s[4:5], v1, v4
; CGP-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, 1, v0
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v4
; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; CGP-NEXT: v_mov_b32_e32 v1, 0
; CGP-NEXT: .LBB2_4:
; CGP-NEXT: s_or_b64 exec, exec, s[6:7]
; CGP-NEXT: v_or_b32_e32 v3, v9, v7
; CGP-NEXT: v_mov_b32_e32 v2, 0
; CGP-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[2:3]
; CGP-NEXT: ; implicit-def: $vgpr2_vgpr3
; CGP-NEXT: s_and_saveexec_b64 s[4:5], vcc
; CGP-NEXT: s_xor_b64 s[6:7], exec, s[4:5]
; CGP-NEXT: s_cbranch_execz .LBB2_6
; CGP-NEXT: ; %bb.5:
; CGP-NEXT: v_ashrrev_i32_e32 v2, 31, v7
; CGP-NEXT: v_add_i32_e32 v3, vcc, v6, v2
; CGP-NEXT: v_addc_u32_e32 v4, vcc, v7, v2, vcc
; CGP-NEXT: v_xor_b32_e32 v6, v3, v2
; CGP-NEXT: v_xor_b32_e32 v7, v4, v2
; CGP-NEXT: v_cvt_f32_u32_e32 v3, v6
; CGP-NEXT: v_cvt_f32_u32_e32 v4, v7
; CGP-NEXT: v_sub_i32_e32 v11, vcc, 0, v6
; CGP-NEXT: v_subb_u32_e32 v12, vcc, 0, v7, vcc
; CGP-NEXT: v_mac_f32_e32 v3, 0x4f800000, v4
; CGP-NEXT: v_rcp_iflag_f32_e32 v3, v3
; CGP-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3
; CGP-NEXT: v_mul_f32_e32 v4, 0x2f800000, v3
; CGP-NEXT: v_trunc_f32_e32 v5, v4
; CGP-NEXT: v_mac_f32_e32 v3, 0xcf800000, v5
; CGP-NEXT: v_cvt_u32_f32_e32 v10, v3
; CGP-NEXT: v_cvt_u32_f32_e32 v13, v5
; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v11, v10, 0
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v13, v[4:5]
; CGP-NEXT: v_mul_hi_u32 v14, v10, v3
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v12, v10, v[4:5]
; CGP-NEXT: v_mul_lo_u32 v5, v13, v3
; CGP-NEXT: v_mul_hi_u32 v3, v13, v3
; CGP-NEXT: v_mul_lo_u32 v15, v10, v4
; CGP-NEXT: v_mul_lo_u32 v16, v13, v4
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v15
; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v14
; CGP-NEXT: v_mul_hi_u32 v14, v10, v4
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v15, v5
; CGP-NEXT: v_add_i32_e32 v3, vcc, v16, v3
; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v14
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v14, vcc, v15, v14
; CGP-NEXT: v_mul_hi_u32 v4, v13, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v5
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v14, v5
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v5
; CGP-NEXT: v_add_i32_e32 v10, vcc, v10, v3
; CGP-NEXT: v_addc_u32_e32 v13, vcc, v13, v4, vcc
; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v11, v10, 0
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v13, v[4:5]
; CGP-NEXT: v_ashrrev_i32_e32 v11, 31, v9
; CGP-NEXT: v_mul_hi_u32 v14, v10, v3
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v12, v10, v[4:5]
; CGP-NEXT: v_add_i32_e32 v5, vcc, v8, v11
; CGP-NEXT: v_addc_u32_e32 v8, vcc, v9, v11, vcc
; CGP-NEXT: v_xor_b32_e32 v9, v5, v11
; CGP-NEXT: v_mul_lo_u32 v5, v13, v3
; CGP-NEXT: v_mul_lo_u32 v12, v10, v4
; CGP-NEXT: v_mul_hi_u32 v3, v13, v3
; CGP-NEXT: v_xor_b32_e32 v8, v8, v11
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v12
; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v14
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v14, v13, v4
; CGP-NEXT: v_add_i32_e32 v5, vcc, v12, v5
; CGP-NEXT: v_mul_hi_u32 v12, v10, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v14, v3
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v12
; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v12, vcc, v14, v12
; CGP-NEXT: v_mul_hi_u32 v4, v13, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v5
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v12, v5
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v5
; CGP-NEXT: v_add_i32_e32 v3, vcc, v10, v3
; CGP-NEXT: v_addc_u32_e32 v4, vcc, v13, v4, vcc
; CGP-NEXT: v_mul_lo_u32 v5, v8, v3
; CGP-NEXT: v_mul_lo_u32 v10, v9, v4
; CGP-NEXT: v_mul_hi_u32 v12, v9, v3
; CGP-NEXT: v_mul_hi_u32 v3, v8, v3
; CGP-NEXT: v_mul_hi_u32 v13, v8, v4
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v10
; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v12
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v12, v8, v4
; CGP-NEXT: v_add_i32_e32 v5, vcc, v10, v5
; CGP-NEXT: v_mul_hi_u32 v10, v9, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v12, v3
; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v10
; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v10, vcc, v12, v10
; CGP-NEXT: v_add_i32_e32 v12, vcc, v3, v5
; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v6, v12, 0
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v10, v5
; CGP-NEXT: v_add_i32_e32 v10, vcc, v13, v5
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v6, v10, v[4:5]
; CGP-NEXT: v_sub_i32_e32 v3, vcc, v9, v3
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v7, v12, v[4:5]
; CGP-NEXT: v_subb_u32_e64 v5, s[4:5], v8, v4, vcc
; CGP-NEXT: v_sub_i32_e64 v4, s[4:5], v8, v4
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v5, v7
; CGP-NEXT: v_subb_u32_e32 v4, vcc, v4, v7, vcc
; CGP-NEXT: v_cndmask_b32_e64 v8, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v3, v6
; CGP-NEXT: v_sub_i32_e32 v3, vcc, v3, v6
; CGP-NEXT: v_cndmask_b32_e64 v9, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], v5, v7
; CGP-NEXT: v_subbrev_u32_e32 v4, vcc, 0, v4, vcc
; CGP-NEXT: v_cndmask_b32_e64 v5, v8, v9, s[4:5]
; CGP-NEXT: v_add_i32_e32 v8, vcc, 1, v12
; CGP-NEXT: v_addc_u32_e32 v9, vcc, 0, v10, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v4, v7
; CGP-NEXT: v_cndmask_b32_e64 v13, 0, -1, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v3, v6
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, v4, v7
; CGP-NEXT: v_cndmask_b32_e32 v3, v13, v3, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, 1, v8
; CGP-NEXT: v_addc_u32_e32 v6, vcc, 0, v9, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; CGP-NEXT: v_cndmask_b32_e32 v3, v8, v4, vcc
; CGP-NEXT: v_cndmask_b32_e32 v4, v9, v6, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
; CGP-NEXT: v_cndmask_b32_e32 v3, v12, v3, vcc
; CGP-NEXT: v_xor_b32_e32 v5, v11, v2
; CGP-NEXT: v_cndmask_b32_e32 v4, v10, v4, vcc
; CGP-NEXT: v_xor_b32_e32 v2, v3, v5
; CGP-NEXT: v_xor_b32_e32 v3, v4, v5
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v5, vcc
; CGP-NEXT: ; implicit-def: $vgpr6
; CGP-NEXT: ; implicit-def: $vgpr8
; CGP-NEXT: .LBB2_6: ; %Flow
; CGP-NEXT: s_or_saveexec_b64 s[6:7], s[6:7]
; CGP-NEXT: s_xor_b64 exec, exec, s[6:7]
; CGP-NEXT: s_cbranch_execz .LBB2_8
; CGP-NEXT: ; %bb.7:
; CGP-NEXT: v_cvt_f32_u32_e32 v2, v6
; CGP-NEXT: v_sub_i32_e32 v3, vcc, 0, v6
; CGP-NEXT: v_rcp_iflag_f32_e32 v2, v2
; CGP-NEXT: v_mul_f32_e32 v2, 0x4f7ffffe, v2
; CGP-NEXT: v_cvt_u32_f32_e32 v2, v2
; CGP-NEXT: v_mul_lo_u32 v3, v3, v2
; CGP-NEXT: v_mul_hi_u32 v3, v2, v3
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_mul_hi_u32 v2, v8, v2
; CGP-NEXT: v_mul_lo_u32 v3, v2, v6
; CGP-NEXT: v_add_i32_e32 v4, vcc, 1, v2
; CGP-NEXT: v_sub_i32_e32 v3, vcc, v8, v3
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v3, v6
; CGP-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
; CGP-NEXT: v_sub_i32_e64 v4, s[4:5], v3, v6
; CGP-NEXT: v_cndmask_b32_e32 v3, v3, v4, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, 1, v2
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v3, v6
; CGP-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
; CGP-NEXT: v_mov_b32_e32 v3, 0
; CGP-NEXT: .LBB2_8:
; CGP-NEXT: s_or_b64 exec, exec, s[6:7]
; CGP-NEXT: s_setpc_b64 s[30:31]
%result = sdiv <2 x i64> %num, %den
ret <2 x i64> %result
}
define i64 @v_sdiv_i64_pow2k_denom(i64 %num) {
; CHECK-LABEL: v_sdiv_i64_pow2k_denom:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_cvt_f32_u32_e32 v2, 0x1000
; CHECK-NEXT: v_cvt_f32_ubyte0_e32 v3, 0
; CHECK-NEXT: s_movk_i32 s6, 0xf000
; CHECK-NEXT: v_mac_f32_e32 v2, 0x4f800000, v3
; CHECK-NEXT: v_rcp_iflag_f32_e32 v2, v2
; CHECK-NEXT: v_mul_f32_e32 v2, 0x5f7ffffc, v2
; CHECK-NEXT: v_mul_f32_e32 v3, 0x2f800000, v2
; CHECK-NEXT: v_trunc_f32_e32 v4, v3
; CHECK-NEXT: v_mac_f32_e32 v2, 0xcf800000, v4
; CHECK-NEXT: v_cvt_u32_f32_e32 v5, v2
; CHECK-NEXT: v_mov_b32_e32 v2, 0xfffff000
; CHECK-NEXT: v_cvt_u32_f32_e32 v6, v4
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v2, v5, 0
; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], s6, v6, v[3:4]
; CHECK-NEXT: v_mul_hi_u32 v7, v5, v2
; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], -1, v5, v[3:4]
; CHECK-NEXT: v_mul_lo_u32 v4, v6, v2
; CHECK-NEXT: v_mul_hi_u32 v2, v6, v2
; CHECK-NEXT: v_mul_lo_u32 v8, v5, v3
; CHECK-NEXT: v_mul_lo_u32 v9, v6, v3
; CHECK-NEXT: v_mul_hi_u32 v10, v5, v3
; CHECK-NEXT: v_mul_hi_u32 v3, v6, v3
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v4, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v4, v7
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v8, v4
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v9, v2
; CHECK-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v10
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v8
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v4
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v7, v4
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v4
; CHECK-NEXT: v_add_i32_e32 v5, vcc, v5, v2
; CHECK-NEXT: v_addc_u32_e32 v6, vcc, v6, v3, vcc
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s6, v5, 0
; CHECK-NEXT: v_ashrrev_i32_e32 v7, 31, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], s6, v6, v[3:4]
; CHECK-NEXT: v_addc_u32_e32 v1, vcc, v1, v7, vcc
; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], -1, v5, v[3:4]
; CHECK-NEXT: v_xor_b32_e32 v4, v0, v7
; CHECK-NEXT: v_mul_lo_u32 v0, v6, v2
; CHECK-NEXT: v_mul_lo_u32 v8, v5, v3
; CHECK-NEXT: v_xor_b32_e32 v9, v1, v7
; CHECK-NEXT: v_mul_hi_u32 v1, v5, v2
; CHECK-NEXT: v_mul_hi_u32 v2, v6, v2
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v1, v6, v3
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v8, v0
; CHECK-NEXT: v_mul_hi_u32 v8, v5, v3
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v2
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v8
; CHECK-NEXT: v_mul_hi_u32 v3, v6, v3
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v2, v1
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v3, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; CHECK-NEXT: v_addc_u32_e32 v1, vcc, v6, v1, vcc
; CHECK-NEXT: v_mul_lo_u32 v2, v9, v0
; CHECK-NEXT: v_mul_lo_u32 v3, v4, v1
; CHECK-NEXT: v_mul_hi_u32 v5, v4, v0
; CHECK-NEXT: v_mul_hi_u32 v0, v9, v0
; CHECK-NEXT: s_movk_i32 s6, 0x1000
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v5
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v5, v9, v1
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CHECK-NEXT: v_mul_hi_u32 v3, v4, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v3
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; CHECK-NEXT: v_add_i32_e32 v5, vcc, v0, v2
; CHECK-NEXT: v_mul_hi_u32 v6, v9, v1
; CHECK-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v5, 0
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v6, v2
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], s6, v3, v[1:2]
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v4, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], 0, v5, v[1:2]
; CHECK-NEXT: v_mov_b32_e32 v6, 0x1000
; CHECK-NEXT: s_bfe_i32 s6, -1, 0x10000
; CHECK-NEXT: v_subb_u32_e64 v2, s[4:5], v9, v1, vcc
; CHECK-NEXT: v_sub_i32_e64 v1, s[4:5], v9, v1
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v6
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v6
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, -1, s[4:5]
; CHECK-NEXT: v_mov_b32_e32 v8, s6
; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v2
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
; CHECK-NEXT: v_cndmask_b32_e64 v2, v8, v4, s[4:5]
; CHECK-NEXT: v_add_i32_e32 v4, vcc, 1, v5
; CHECK-NEXT: v_addc_u32_e32 v8, vcc, 0, v3, vcc
; CHECK-NEXT: s_bfe_i32 s4, -1, 0x10000
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v0, v6
; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
; CHECK-NEXT: v_mov_b32_e32 v6, s4
; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1
; CHECK-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, 1, v4
; CHECK-NEXT: v_addc_u32_e32 v6, vcc, 0, v8, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0
; CHECK-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v1, v8, v6, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; CHECK-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; CHECK-NEXT: v_xor_b32_e32 v0, v0, v7
; CHECK-NEXT: v_xor_b32_e32 v1, v1, v7
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v7
; CHECK-NEXT: v_subb_u32_e32 v1, vcc, v1, v7, vcc
; CHECK-NEXT: s_setpc_b64 s[30:31]
%result = sdiv i64 %num, 4096
ret i64 %result
}
define <2 x i64> @v_sdiv_v2i64_pow2k_denom(<2 x i64> %num) {
; GISEL-LABEL: v_sdiv_v2i64_pow2k_denom:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: s_mov_b32 s8, 0
; GISEL-NEXT: s_add_u32 s4, 0x1000, 0
; GISEL-NEXT: s_mov_b32 s9, s8
; GISEL-NEXT: s_addc_u32 s5, 0, 0
; GISEL-NEXT: s_xor_b64 s[6:7], s[4:5], s[8:9]
; GISEL-NEXT: v_cvt_f32_u32_e32 v4, s6
; GISEL-NEXT: v_cvt_f32_u32_e32 v5, s7
; GISEL-NEXT: s_sub_u32 s10, 0, s6
; GISEL-NEXT: s_subb_u32 s11, 0, s7
; GISEL-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; GISEL-NEXT: v_rcp_iflag_f32_e32 v4, v4
; GISEL-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; GISEL-NEXT: v_trunc_f32_e32 v6, v5
; GISEL-NEXT: v_mac_f32_e32 v4, 0xcf800000, v6
; GISEL-NEXT: v_cvt_u32_f32_e32 v7, v4
; GISEL-NEXT: v_cvt_u32_f32_e32 v8, v6
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], s10, v7, 0
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s10, v8, v[5:6]
; GISEL-NEXT: v_mul_hi_u32 v9, v7, v4
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s11, v7, v[5:6]
; GISEL-NEXT: v_mul_lo_u32 v6, v8, v4
; GISEL-NEXT: v_mul_hi_u32 v4, v8, v4
; GISEL-NEXT: v_mul_lo_u32 v10, v7, v5
; GISEL-NEXT: v_mul_lo_u32 v11, v8, v5
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v9
; GISEL-NEXT: v_mul_hi_u32 v9, v7, v5
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v10, v6
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v11, v4
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v9
; GISEL-NEXT: v_mul_hi_u32 v5, v8, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v9, v6
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v7, v4
; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v8, v5, vcc
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s10, v9, 0
; GISEL-NEXT: v_mov_b32_e32 v4, v6
; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s10, v8, v[4:5]
; GISEL-NEXT: v_ashrrev_i32_e32 v4, 31, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4
; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s11, v9, v[6:7]
; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v1, v4, vcc
; GISEL-NEXT: v_xor_b32_e32 v7, v0, v4
; GISEL-NEXT: v_mul_lo_u32 v0, v8, v5
; GISEL-NEXT: v_mul_lo_u32 v10, v9, v6
; GISEL-NEXT: v_xor_b32_e32 v11, v1, v4
; GISEL-NEXT: v_mul_hi_u32 v1, v9, v5
; GISEL-NEXT: v_mul_hi_u32 v5, v8, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v1, v8, v6
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v10, v9, v6
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
; GISEL-NEXT: v_mul_hi_u32 v6, v8, v6
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v5, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v6, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v8, v1, vcc
; GISEL-NEXT: v_mul_lo_u32 v5, v11, v0
; GISEL-NEXT: v_mul_lo_u32 v6, v7, v1
; GISEL-NEXT: v_mul_hi_u32 v8, v7, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v11, v0
; GISEL-NEXT: v_mul_hi_u32 v9, v11, v1
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v8, v11, v1
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; GISEL-NEXT: v_mul_hi_u32 v6, v7, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v8, v6
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v0, v5
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v8, 0
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v5
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v9, v[1:2]
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v7, v0
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s7, v8, v[5:6]
; GISEL-NEXT: v_mov_b32_e32 v1, s7
; GISEL-NEXT: v_subb_u32_e64 v6, s[4:5], v11, v5, vcc
; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], v11, v5
; GISEL-NEXT: v_cmp_le_u32_e64 s[4:5], s7, v6
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_le_u32_e64 s[4:5], s6, v0
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], s7, v6
; GISEL-NEXT: v_cndmask_b32_e64 v7, v7, v10, s[4:5]
; GISEL-NEXT: s_add_u32 s4, 0x1000, 0
; GISEL-NEXT: s_addc_u32 s5, 0, 0
; GISEL-NEXT: s_xor_b64 s[8:9], s[4:5], s[8:9]
; GISEL-NEXT: v_subb_u32_e32 v1, vcc, v5, v1, vcc
; GISEL-NEXT: v_cvt_f32_u32_e32 v5, s8
; GISEL-NEXT: v_cvt_f32_u32_e32 v6, s9
; GISEL-NEXT: v_subrev_i32_e32 v0, vcc, s6, v0
; GISEL-NEXT: v_subbrev_u32_e32 v10, vcc, 0, v1, vcc
; GISEL-NEXT: v_mac_f32_e32 v5, 0x4f800000, v6
; GISEL-NEXT: v_rcp_iflag_f32_e32 v1, v5
; GISEL-NEXT: v_add_i32_e32 v11, vcc, 1, v8
; GISEL-NEXT: v_addc_u32_e32 v12, vcc, 0, v9, vcc
; GISEL-NEXT: v_mul_f32_e32 v1, 0x5f7ffffc, v1
; GISEL-NEXT: v_mul_f32_e32 v6, 0x2f800000, v1
; GISEL-NEXT: v_trunc_f32_e32 v6, v6
; GISEL-NEXT: v_mac_f32_e32 v1, 0xcf800000, v6
; GISEL-NEXT: v_cvt_u32_f32_e32 v13, v1
; GISEL-NEXT: v_cmp_le_u32_e32 vcc, s7, v10
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, -1, vcc
; GISEL-NEXT: v_cmp_le_u32_e32 vcc, s6, v0
; GISEL-NEXT: s_sub_u32 s6, 0, s8
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v13, 0
; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v6
; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, s7, v10
; GISEL-NEXT: v_cndmask_b32_e32 v10, v5, v14, vcc
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v15, v[1:2]
; GISEL-NEXT: s_subb_u32 s7, 0, s9
; GISEL-NEXT: v_add_i32_e32 v1, vcc, 1, v11
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s7, v13, v[5:6]
; GISEL-NEXT: v_addc_u32_e32 v14, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
; GISEL-NEXT: v_cndmask_b32_e32 v6, v11, v1, vcc
; GISEL-NEXT: v_mul_lo_u32 v1, v15, v0
; GISEL-NEXT: v_mul_lo_u32 v10, v13, v5
; GISEL-NEXT: v_cndmask_b32_e32 v11, v12, v14, vcc
; GISEL-NEXT: v_mul_hi_u32 v12, v13, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v15, v0
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v12
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v12, v15, v5
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; GISEL-NEXT: v_mul_hi_u32 v10, v13, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v12, v0
; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v12, v10
; GISEL-NEXT: v_mul_hi_u32 v5, v15, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v5, v1
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v13, v0
; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v15, v1, vcc
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v10, 0
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v7
; GISEL-NEXT: v_cndmask_b32_e32 v7, v8, v6, vcc
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v12, v[1:2]
; GISEL-NEXT: v_xor_b32_e32 v1, v7, v4
; GISEL-NEXT: v_ashrrev_i32_e32 v7, 31, v3
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s7, v10, v[5:6]
; GISEL-NEXT: v_cndmask_b32_e32 v8, v9, v11, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v7
; GISEL-NEXT: v_addc_u32_e32 v3, vcc, v3, v7, vcc
; GISEL-NEXT: v_xor_b32_e32 v9, v2, v7
; GISEL-NEXT: v_mul_lo_u32 v2, v12, v0
; GISEL-NEXT: v_mul_lo_u32 v6, v10, v5
; GISEL-NEXT: v_xor_b32_e32 v11, v3, v7
; GISEL-NEXT: v_mul_hi_u32 v3, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v12, v0
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v3, v12, v5
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v6, v2
; GISEL-NEXT: v_mul_hi_u32 v6, v10, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; GISEL-NEXT: v_mul_hi_u32 v5, v12, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0
; GISEL-NEXT: v_addc_u32_e32 v2, vcc, v12, v2, vcc
; GISEL-NEXT: v_mul_lo_u32 v3, v11, v0
; GISEL-NEXT: v_mul_lo_u32 v5, v9, v2
; GISEL-NEXT: v_mul_hi_u32 v6, v9, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v11, v0
; GISEL-NEXT: v_xor_b32_e32 v8, v8, v4
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v6, v11, v2
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; GISEL-NEXT: v_mul_hi_u32 v5, v9, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v0, v3
; GISEL-NEXT: v_mul_hi_u32 v6, v11, v2
; GISEL-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s8, v10, 0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v6, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v3
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s8, v12, v[0:1]
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v1, v4
; GISEL-NEXT: v_subb_u32_e32 v1, vcc, v8, v4, vcc
; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], s9, v10, v[5:6]
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v9, v2
; GISEL-NEXT: v_mov_b32_e32 v8, s9
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], v11, v3, vcc
; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v11, v3
; GISEL-NEXT: v_cmp_le_u32_e64 s[4:5], s9, v4
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v8, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_le_u32_e64 s[4:5], s8, v2
; GISEL-NEXT: v_subrev_i32_e32 v2, vcc, s8, v2
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], s9, v4
; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v4, v5, v6, s[4:5]
; GISEL-NEXT: v_add_i32_e32 v5, vcc, 1, v10
; GISEL-NEXT: v_addc_u32_e32 v6, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_le_u32_e32 vcc, s9, v3
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
; GISEL-NEXT: v_cmp_le_u32_e32 vcc, s8, v2
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, s9, v3
; GISEL-NEXT: v_cndmask_b32_e32 v2, v8, v2, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v5
; GISEL-NEXT: v_addc_u32_e32 v8, vcc, 0, v6, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; GISEL-NEXT: v_cndmask_b32_e32 v2, v5, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; GISEL-NEXT: v_cndmask_b32_e32 v2, v10, v2, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v12, v3, vcc
; GISEL-NEXT: v_xor_b32_e32 v2, v2, v7
; GISEL-NEXT: v_xor_b32_e32 v3, v3, v7
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v7
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v7, vcc
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_v2i64_pow2k_denom:
; CGP: ; %bb.0:
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CGP-NEXT: v_cvt_f32_u32_e32 v4, 0x1000
; CGP-NEXT: v_cvt_f32_ubyte0_e32 v5, 0
; CGP-NEXT: s_movk_i32 s6, 0xf000
; CGP-NEXT: s_movk_i32 s7, 0x1000
; CGP-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; CGP-NEXT: v_rcp_iflag_f32_e32 v4, v4
; CGP-NEXT: s_bfe_i32 s8, -1, 0x10000
; CGP-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; CGP-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; CGP-NEXT: v_trunc_f32_e32 v6, v5
; CGP-NEXT: v_mac_f32_e32 v4, 0xcf800000, v6
; CGP-NEXT: v_cvt_u32_f32_e32 v7, v4
; CGP-NEXT: v_mov_b32_e32 v4, 0xfffff000
; CGP-NEXT: v_cvt_u32_f32_e32 v8, v6
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v4, v7, 0
; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v8, v[5:6]
; CGP-NEXT: v_mul_hi_u32 v9, v7, v4
; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], -1, v7, v[5:6]
; CGP-NEXT: v_mul_lo_u32 v6, v8, v4
; CGP-NEXT: v_mul_hi_u32 v4, v8, v4
; CGP-NEXT: v_mul_lo_u32 v10, v7, v5
; CGP-NEXT: v_mul_lo_u32 v11, v8, v5
; CGP-NEXT: v_mul_hi_u32 v12, v7, v5
; CGP-NEXT: v_mul_hi_u32 v5, v8, v5
; CGP-NEXT: v_add_i32_e32 v6, vcc, v6, v10
; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v6, v9
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v10, v6
; CGP-NEXT: v_add_i32_e32 v4, vcc, v11, v4
; CGP-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v12
; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v9, vcc, v9, v10
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v9, v6
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; CGP-NEXT: v_add_i32_e32 v9, vcc, v7, v4
; CGP-NEXT: v_addc_u32_e32 v8, vcc, v8, v5, vcc
; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v9, 0
; CGP-NEXT: v_mov_b32_e32 v4, v6
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v8, v[4:5]
; CGP-NEXT: v_ashrrev_i32_e32 v4, 31, v1
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v4
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v9, v[6:7]
; CGP-NEXT: v_addc_u32_e32 v1, vcc, v1, v4, vcc
; CGP-NEXT: v_xor_b32_e32 v10, v0, v4
; CGP-NEXT: v_mul_lo_u32 v0, v8, v5
; CGP-NEXT: v_mul_lo_u32 v7, v9, v6
; CGP-NEXT: v_xor_b32_e32 v11, v1, v4
; CGP-NEXT: v_mul_hi_u32 v1, v9, v5
; CGP-NEXT: v_mul_hi_u32 v5, v8, v5
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v1, v8, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
; CGP-NEXT: v_mul_hi_u32 v7, v9, v6
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v5
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v7
; CGP-NEXT: v_mul_hi_u32 v6, v8, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v5, v1
; CGP-NEXT: v_add_i32_e32 v1, vcc, v6, v1
; CGP-NEXT: v_add_i32_e32 v0, vcc, v9, v0
; CGP-NEXT: v_addc_u32_e32 v1, vcc, v8, v1, vcc
; CGP-NEXT: v_mul_lo_u32 v5, v11, v0
; CGP-NEXT: v_mul_lo_u32 v6, v10, v1
; CGP-NEXT: v_mul_hi_u32 v7, v10, v0
; CGP-NEXT: v_mul_hi_u32 v0, v11, v0
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v7
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v7, v11, v1
; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; CGP-NEXT: v_mul_hi_u32 v6, v10, v1
; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v7, v6
; CGP-NEXT: v_add_i32_e32 v8, vcc, v0, v5
; CGP-NEXT: v_mul_hi_u32 v7, v11, v1
; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s7, v8, 0
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; CGP-NEXT: v_add_i32_e32 v9, vcc, v7, v5
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v9, v[1:2]
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v10, v0
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v8, v[6:7]
; CGP-NEXT: v_mov_b32_e32 v5, 0x1000
; CGP-NEXT: v_mov_b32_e32 v10, s8
; CGP-NEXT: v_subb_u32_e64 v1, s[4:5], v11, v6, vcc
; CGP-NEXT: v_sub_i32_e64 v6, s[4:5], v11, v6
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v5
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1
; CGP-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v6, vcc
; CGP-NEXT: v_cvt_f32_u32_e32 v6, 0x1000
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v5
; CGP-NEXT: v_cndmask_b32_e64 v10, v10, v7, s[4:5]
; CGP-NEXT: v_subbrev_u32_e32 v7, vcc, 0, v1, vcc
; CGP-NEXT: v_cvt_f32_ubyte0_e32 v1, 0
; CGP-NEXT: v_mac_f32_e32 v6, 0x4f800000, v1
; CGP-NEXT: v_rcp_iflag_f32_e32 v1, v6
; CGP-NEXT: v_add_i32_e32 v11, vcc, 1, v8
; CGP-NEXT: v_addc_u32_e32 v12, vcc, 0, v9, vcc
; CGP-NEXT: v_mul_f32_e32 v1, 0x5f7ffffc, v1
; CGP-NEXT: v_mul_f32_e32 v6, 0x2f800000, v1
; CGP-NEXT: v_trunc_f32_e32 v6, v6
; CGP-NEXT: v_mac_f32_e32 v1, 0xcf800000, v6
; CGP-NEXT: v_cvt_u32_f32_e32 v13, v1
; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v5
; CGP-NEXT: v_mov_b32_e32 v15, s4
; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v13, 0
; CGP-NEXT: v_cvt_u32_f32_e32 v16, v6
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v7
; CGP-NEXT: v_cndmask_b32_e32 v14, v15, v14, vcc
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v16, v[1:2]
; CGP-NEXT: v_add_i32_e32 v1, vcc, 1, v11
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v13, v[6:7]
; CGP-NEXT: v_addc_u32_e32 v15, vcc, 0, v12, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14
; CGP-NEXT: v_cndmask_b32_e32 v7, v11, v1, vcc
; CGP-NEXT: v_mul_lo_u32 v1, v16, v0
; CGP-NEXT: v_mul_lo_u32 v11, v13, v6
; CGP-NEXT: v_mul_hi_u32 v14, v13, v0
; CGP-NEXT: v_cndmask_b32_e32 v12, v12, v15, vcc
; CGP-NEXT: v_mul_hi_u32 v0, v16, v0
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v11
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v14
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v14, v16, v6
; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
; CGP-NEXT: v_mul_hi_u32 v11, v13, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v14, v0
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v11
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v11, vcc, v14, v11
; CGP-NEXT: v_mul_hi_u32 v6, v16, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
; CGP-NEXT: v_add_i32_e32 v1, vcc, v6, v1
; CGP-NEXT: v_add_i32_e32 v11, vcc, v13, v0
; CGP-NEXT: v_addc_u32_e32 v13, vcc, v16, v1, vcc
; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v11, 0
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
; CGP-NEXT: v_cndmask_b32_e32 v8, v8, v7, vcc
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v13, v[1:2]
; CGP-NEXT: v_xor_b32_e32 v1, v8, v4
; CGP-NEXT: v_ashrrev_i32_e32 v8, 31, v3
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v11, v[6:7]
; CGP-NEXT: v_cndmask_b32_e32 v9, v9, v12, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v8
; CGP-NEXT: v_addc_u32_e32 v3, vcc, v3, v8, vcc
; CGP-NEXT: v_xor_b32_e32 v10, v2, v8
; CGP-NEXT: v_mul_lo_u32 v2, v13, v0
; CGP-NEXT: v_mul_lo_u32 v7, v11, v6
; CGP-NEXT: v_xor_b32_e32 v12, v3, v8
; CGP-NEXT: v_mul_hi_u32 v3, v11, v0
; CGP-NEXT: v_mul_hi_u32 v0, v13, v0
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v3, v13, v6
; CGP-NEXT: v_add_i32_e32 v2, vcc, v7, v2
; CGP-NEXT: v_mul_hi_u32 v7, v11, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v7
; CGP-NEXT: v_mul_hi_u32 v6, v13, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CGP-NEXT: v_add_i32_e32 v2, vcc, v6, v2
; CGP-NEXT: v_add_i32_e32 v0, vcc, v11, v0
; CGP-NEXT: v_addc_u32_e32 v2, vcc, v13, v2, vcc
; CGP-NEXT: v_mul_lo_u32 v3, v12, v0
; CGP-NEXT: v_mul_lo_u32 v6, v10, v2
; CGP-NEXT: v_mul_hi_u32 v7, v10, v0
; CGP-NEXT: v_mul_hi_u32 v0, v12, v0
; CGP-NEXT: v_xor_b32_e32 v9, v9, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v7
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v7, v12, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v6, v3
; CGP-NEXT: v_mul_hi_u32 v6, v10, v2
; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v7, v6
; CGP-NEXT: v_add_i32_e32 v11, vcc, v0, v3
; CGP-NEXT: v_mul_hi_u32 v7, v12, v2
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s7, v11, 0
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v6, v0
; CGP-NEXT: v_add_i32_e32 v13, vcc, v7, v0
; CGP-NEXT: v_mov_b32_e32 v0, v3
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v13, v[0:1]
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v1, v4
; CGP-NEXT: v_subb_u32_e32 v1, vcc, v9, v4, vcc
; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], 0, v11, v[6:7]
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v10, v2
; CGP-NEXT: v_subb_u32_e64 v4, s[4:5], v12, v3, vcc
; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v12, v3
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; CGP-NEXT: s_bfe_i32 s6, -1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
; CGP-NEXT: v_mov_b32_e32 v7, s6
; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; CGP-NEXT: v_cndmask_b32_e64 v4, v7, v6, s[4:5]
; CGP-NEXT: v_add_i32_e32 v6, vcc, 1, v11
; CGP-NEXT: v_addc_u32_e32 v7, vcc, 0, v13, vcc
; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; CGP-NEXT: v_mov_b32_e32 v5, s4
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3
; CGP-NEXT: v_cndmask_b32_e32 v2, v5, v2, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, 1, v6
; CGP-NEXT: v_addc_u32_e32 v5, vcc, 0, v7, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; CGP-NEXT: v_cndmask_b32_e32 v2, v6, v3, vcc
; CGP-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; CGP-NEXT: v_cndmask_b32_e32 v2, v11, v2, vcc
; CGP-NEXT: v_cndmask_b32_e32 v3, v13, v3, vcc
; CGP-NEXT: v_xor_b32_e32 v2, v2, v8
; CGP-NEXT: v_xor_b32_e32 v3, v3, v8
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v8
; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v8, vcc
; CGP-NEXT: s_setpc_b64 s[30:31]
%result = sdiv <2 x i64> %num, <i64 4096, i64 4096>
ret <2 x i64> %result
}
define i64 @v_sdiv_i64_oddk_denom(i64 %num) {
; CHECK-LABEL: v_sdiv_i64_oddk_denom:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_cvt_f32_u32_e32 v2, 0x12d8fb
; CHECK-NEXT: v_cvt_f32_ubyte0_e32 v3, 0
; CHECK-NEXT: s_mov_b32 s6, 0xffed2705
; CHECK-NEXT: v_mac_f32_e32 v2, 0x4f800000, v3
; CHECK-NEXT: v_rcp_iflag_f32_e32 v2, v2
; CHECK-NEXT: v_mul_f32_e32 v2, 0x5f7ffffc, v2
; CHECK-NEXT: v_mul_f32_e32 v3, 0x2f800000, v2
; CHECK-NEXT: v_trunc_f32_e32 v4, v3
; CHECK-NEXT: v_mac_f32_e32 v2, 0xcf800000, v4
; CHECK-NEXT: v_cvt_u32_f32_e32 v5, v2
; CHECK-NEXT: v_mov_b32_e32 v2, 0xffed2705
; CHECK-NEXT: v_cvt_u32_f32_e32 v6, v4
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v2, v5, 0
; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], s6, v6, v[3:4]
; CHECK-NEXT: v_mul_hi_u32 v7, v5, v2
; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], -1, v5, v[3:4]
; CHECK-NEXT: v_mul_lo_u32 v4, v6, v2
; CHECK-NEXT: v_mul_hi_u32 v2, v6, v2
; CHECK-NEXT: v_mul_lo_u32 v8, v5, v3
; CHECK-NEXT: v_mul_lo_u32 v9, v6, v3
; CHECK-NEXT: v_mul_hi_u32 v10, v5, v3
; CHECK-NEXT: v_mul_hi_u32 v3, v6, v3
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v4, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v4, v7
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v8, v4
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v9, v2
; CHECK-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v10
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v8
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v4
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v4, vcc, v7, v4
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v4
; CHECK-NEXT: v_add_i32_e32 v5, vcc, v5, v2
; CHECK-NEXT: v_addc_u32_e32 v6, vcc, v6, v3, vcc
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s6, v5, 0
; CHECK-NEXT: v_ashrrev_i32_e32 v7, 31, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], s6, v6, v[3:4]
; CHECK-NEXT: v_addc_u32_e32 v1, vcc, v1, v7, vcc
; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], -1, v5, v[3:4]
; CHECK-NEXT: v_xor_b32_e32 v4, v0, v7
; CHECK-NEXT: v_mul_lo_u32 v0, v6, v2
; CHECK-NEXT: v_mul_lo_u32 v8, v5, v3
; CHECK-NEXT: v_xor_b32_e32 v9, v1, v7
; CHECK-NEXT: v_mul_hi_u32 v1, v5, v2
; CHECK-NEXT: v_mul_hi_u32 v2, v6, v2
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v1, v6, v3
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v8, v0
; CHECK-NEXT: v_mul_hi_u32 v8, v5, v3
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v2
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v8
; CHECK-NEXT: v_mul_hi_u32 v3, v6, v3
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v2, v1
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v3, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; CHECK-NEXT: v_addc_u32_e32 v1, vcc, v6, v1, vcc
; CHECK-NEXT: v_mul_lo_u32 v2, v9, v0
; CHECK-NEXT: v_mul_lo_u32 v3, v4, v1
; CHECK-NEXT: v_mul_hi_u32 v5, v4, v0
; CHECK-NEXT: v_mul_hi_u32 v0, v9, v0
; CHECK-NEXT: s_mov_b32 s6, 0x12d8fb
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v5
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v5, v9, v1
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CHECK-NEXT: v_mul_hi_u32 v3, v4, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v3
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; CHECK-NEXT: v_add_i32_e32 v5, vcc, v0, v2
; CHECK-NEXT: v_mul_hi_u32 v6, v9, v1
; CHECK-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v5, 0
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v6, v2
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], s6, v3, v[1:2]
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v4, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], 0, v5, v[1:2]
; CHECK-NEXT: v_mov_b32_e32 v6, 0x12d8fb
; CHECK-NEXT: s_bfe_i32 s6, -1, 0x10000
; CHECK-NEXT: v_subb_u32_e64 v2, s[4:5], v9, v1, vcc
; CHECK-NEXT: v_sub_i32_e64 v1, s[4:5], v9, v1
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v6
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v6
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, -1, s[4:5]
; CHECK-NEXT: v_mov_b32_e32 v8, s6
; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v2
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
; CHECK-NEXT: v_cndmask_b32_e64 v2, v8, v4, s[4:5]
; CHECK-NEXT: v_add_i32_e32 v4, vcc, 1, v5
; CHECK-NEXT: v_addc_u32_e32 v8, vcc, 0, v3, vcc
; CHECK-NEXT: s_bfe_i32 s4, -1, 0x10000
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v0, v6
; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
; CHECK-NEXT: v_mov_b32_e32 v6, s4
; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1
; CHECK-NEXT: v_cndmask_b32_e32 v0, v6, v0, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, 1, v4
; CHECK-NEXT: v_addc_u32_e32 v6, vcc, 0, v8, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0
; CHECK-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v1, v8, v6, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; CHECK-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
; CHECK-NEXT: v_xor_b32_e32 v0, v0, v7
; CHECK-NEXT: v_xor_b32_e32 v1, v1, v7
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v7
; CHECK-NEXT: v_subb_u32_e32 v1, vcc, v1, v7, vcc
; CHECK-NEXT: s_setpc_b64 s[30:31]
%result = sdiv i64 %num, 1235195
ret i64 %result
}
define <2 x i64> @v_sdiv_v2i64_oddk_denom(<2 x i64> %num) {
; GISEL-LABEL: v_sdiv_v2i64_oddk_denom:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: s_mov_b32 s8, 0
; GISEL-NEXT: s_add_u32 s4, 0x12d8fb, 0
; GISEL-NEXT: s_mov_b32 s9, s8
; GISEL-NEXT: s_addc_u32 s5, 0, 0
; GISEL-NEXT: s_xor_b64 s[6:7], s[4:5], s[8:9]
; GISEL-NEXT: v_cvt_f32_u32_e32 v4, s6
; GISEL-NEXT: v_cvt_f32_u32_e32 v5, s7
; GISEL-NEXT: s_sub_u32 s10, 0, s6
; GISEL-NEXT: s_subb_u32 s11, 0, s7
; GISEL-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; GISEL-NEXT: v_rcp_iflag_f32_e32 v4, v4
; GISEL-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; GISEL-NEXT: v_trunc_f32_e32 v6, v5
; GISEL-NEXT: v_mac_f32_e32 v4, 0xcf800000, v6
; GISEL-NEXT: v_cvt_u32_f32_e32 v7, v4
; GISEL-NEXT: v_cvt_u32_f32_e32 v8, v6
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], s10, v7, 0
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s10, v8, v[5:6]
; GISEL-NEXT: v_mul_hi_u32 v9, v7, v4
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s11, v7, v[5:6]
; GISEL-NEXT: v_mul_lo_u32 v6, v8, v4
; GISEL-NEXT: v_mul_hi_u32 v4, v8, v4
; GISEL-NEXT: v_mul_lo_u32 v10, v7, v5
; GISEL-NEXT: v_mul_lo_u32 v11, v8, v5
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v9
; GISEL-NEXT: v_mul_hi_u32 v9, v7, v5
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v10, v6
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v11, v4
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v9
; GISEL-NEXT: v_mul_hi_u32 v5, v8, v5
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v9, v6
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v7, v4
; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v8, v5, vcc
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s10, v9, 0
; GISEL-NEXT: v_mov_b32_e32 v4, v6
; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s10, v8, v[4:5]
; GISEL-NEXT: v_ashrrev_i32_e32 v4, 31, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4
; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s11, v9, v[6:7]
; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v1, v4, vcc
; GISEL-NEXT: v_xor_b32_e32 v7, v0, v4
; GISEL-NEXT: v_mul_lo_u32 v0, v8, v5
; GISEL-NEXT: v_mul_lo_u32 v10, v9, v6
; GISEL-NEXT: v_xor_b32_e32 v11, v1, v4
; GISEL-NEXT: v_mul_hi_u32 v1, v9, v5
; GISEL-NEXT: v_mul_hi_u32 v5, v8, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v1, v8, v6
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v10, v9, v6
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
; GISEL-NEXT: v_mul_hi_u32 v6, v8, v6
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v5, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v6, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v8, v1, vcc
; GISEL-NEXT: v_mul_lo_u32 v5, v11, v0
; GISEL-NEXT: v_mul_lo_u32 v6, v7, v1
; GISEL-NEXT: v_mul_hi_u32 v8, v7, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v11, v0
; GISEL-NEXT: v_mul_hi_u32 v9, v11, v1
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v8, v11, v1
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; GISEL-NEXT: v_mul_hi_u32 v6, v7, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v6, vcc, v8, v6
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v0, v5
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v8, 0
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v5
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v9, v[1:2]
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v7, v0
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s7, v8, v[5:6]
; GISEL-NEXT: v_mov_b32_e32 v1, s7
; GISEL-NEXT: v_subb_u32_e64 v6, s[4:5], v11, v5, vcc
; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], v11, v5
; GISEL-NEXT: v_cmp_le_u32_e64 s[4:5], s7, v6
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_le_u32_e64 s[4:5], s6, v0
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], s7, v6
; GISEL-NEXT: v_cndmask_b32_e64 v7, v7, v10, s[4:5]
; GISEL-NEXT: s_add_u32 s4, 0x12d8fb, 0
; GISEL-NEXT: s_addc_u32 s5, 0, 0
; GISEL-NEXT: s_xor_b64 s[8:9], s[4:5], s[8:9]
; GISEL-NEXT: v_subb_u32_e32 v1, vcc, v5, v1, vcc
; GISEL-NEXT: v_cvt_f32_u32_e32 v5, s8
; GISEL-NEXT: v_cvt_f32_u32_e32 v6, s9
; GISEL-NEXT: v_subrev_i32_e32 v0, vcc, s6, v0
; GISEL-NEXT: v_subbrev_u32_e32 v10, vcc, 0, v1, vcc
; GISEL-NEXT: v_mac_f32_e32 v5, 0x4f800000, v6
; GISEL-NEXT: v_rcp_iflag_f32_e32 v1, v5
; GISEL-NEXT: v_add_i32_e32 v11, vcc, 1, v8
; GISEL-NEXT: v_addc_u32_e32 v12, vcc, 0, v9, vcc
; GISEL-NEXT: v_mul_f32_e32 v1, 0x5f7ffffc, v1
; GISEL-NEXT: v_mul_f32_e32 v6, 0x2f800000, v1
; GISEL-NEXT: v_trunc_f32_e32 v6, v6
; GISEL-NEXT: v_mac_f32_e32 v1, 0xcf800000, v6
; GISEL-NEXT: v_cvt_u32_f32_e32 v13, v1
; GISEL-NEXT: v_cmp_le_u32_e32 vcc, s7, v10
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, -1, vcc
; GISEL-NEXT: v_cmp_le_u32_e32 vcc, s6, v0
; GISEL-NEXT: s_sub_u32 s6, 0, s8
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v13, 0
; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v6
; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, s7, v10
; GISEL-NEXT: v_cndmask_b32_e32 v10, v5, v14, vcc
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v15, v[1:2]
; GISEL-NEXT: s_subb_u32 s7, 0, s9
; GISEL-NEXT: v_add_i32_e32 v1, vcc, 1, v11
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s7, v13, v[5:6]
; GISEL-NEXT: v_addc_u32_e32 v14, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
; GISEL-NEXT: v_cndmask_b32_e32 v6, v11, v1, vcc
; GISEL-NEXT: v_mul_lo_u32 v1, v15, v0
; GISEL-NEXT: v_mul_lo_u32 v10, v13, v5
; GISEL-NEXT: v_cndmask_b32_e32 v11, v12, v14, vcc
; GISEL-NEXT: v_mul_hi_u32 v12, v13, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v15, v0
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v12
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v12, v15, v5
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; GISEL-NEXT: v_mul_hi_u32 v10, v13, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v12, v0
; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v12, v10
; GISEL-NEXT: v_mul_hi_u32 v5, v15, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v5, v1
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v13, v0
; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v15, v1, vcc
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v10, 0
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v7
; GISEL-NEXT: v_cndmask_b32_e32 v7, v8, v6, vcc
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v12, v[1:2]
; GISEL-NEXT: v_xor_b32_e32 v1, v7, v4
; GISEL-NEXT: v_ashrrev_i32_e32 v7, 31, v3
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s7, v10, v[5:6]
; GISEL-NEXT: v_cndmask_b32_e32 v8, v9, v11, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v7
; GISEL-NEXT: v_addc_u32_e32 v3, vcc, v3, v7, vcc
; GISEL-NEXT: v_xor_b32_e32 v9, v2, v7
; GISEL-NEXT: v_mul_lo_u32 v2, v12, v0
; GISEL-NEXT: v_mul_lo_u32 v6, v10, v5
; GISEL-NEXT: v_xor_b32_e32 v11, v3, v7
; GISEL-NEXT: v_mul_hi_u32 v3, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v12, v0
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v3, v12, v5
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v6, v2
; GISEL-NEXT: v_mul_hi_u32 v6, v10, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; GISEL-NEXT: v_mul_hi_u32 v5, v12, v5
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0
; GISEL-NEXT: v_addc_u32_e32 v2, vcc, v12, v2, vcc
; GISEL-NEXT: v_mul_lo_u32 v3, v11, v0
; GISEL-NEXT: v_mul_lo_u32 v5, v9, v2
; GISEL-NEXT: v_mul_hi_u32 v6, v9, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v11, v0
; GISEL-NEXT: v_xor_b32_e32 v8, v8, v4
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v6, v11, v2
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; GISEL-NEXT: v_mul_hi_u32 v5, v9, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v0, v3
; GISEL-NEXT: v_mul_hi_u32 v6, v11, v2
; GISEL-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s8, v10, 0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v6, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v3
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s8, v12, v[0:1]
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v1, v4
; GISEL-NEXT: v_subb_u32_e32 v1, vcc, v8, v4, vcc
; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], s9, v10, v[5:6]
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v9, v2
; GISEL-NEXT: v_mov_b32_e32 v8, s9
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], v11, v3, vcc
; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v11, v3
; GISEL-NEXT: v_cmp_le_u32_e64 s[4:5], s9, v4
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v8, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_le_u32_e64 s[4:5], s8, v2
; GISEL-NEXT: v_subrev_i32_e32 v2, vcc, s8, v2
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], s9, v4
; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v4, v5, v6, s[4:5]
; GISEL-NEXT: v_add_i32_e32 v5, vcc, 1, v10
; GISEL-NEXT: v_addc_u32_e32 v6, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_le_u32_e32 vcc, s9, v3
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
; GISEL-NEXT: v_cmp_le_u32_e32 vcc, s8, v2
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, s9, v3
; GISEL-NEXT: v_cndmask_b32_e32 v2, v8, v2, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v5
; GISEL-NEXT: v_addc_u32_e32 v8, vcc, 0, v6, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; GISEL-NEXT: v_cndmask_b32_e32 v2, v5, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; GISEL-NEXT: v_cndmask_b32_e32 v2, v10, v2, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v12, v3, vcc
; GISEL-NEXT: v_xor_b32_e32 v2, v2, v7
; GISEL-NEXT: v_xor_b32_e32 v3, v3, v7
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v7
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v7, vcc
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_v2i64_oddk_denom:
; CGP: ; %bb.0:
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CGP-NEXT: v_cvt_f32_u32_e32 v4, 0x12d8fb
; CGP-NEXT: v_cvt_f32_ubyte0_e32 v5, 0
; CGP-NEXT: s_mov_b32 s6, 0xffed2705
; CGP-NEXT: s_mov_b32 s7, 0x12d8fb
; CGP-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; CGP-NEXT: v_rcp_iflag_f32_e32 v4, v4
; CGP-NEXT: s_bfe_i32 s8, -1, 0x10000
; CGP-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; CGP-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; CGP-NEXT: v_trunc_f32_e32 v6, v5
; CGP-NEXT: v_mac_f32_e32 v4, 0xcf800000, v6
; CGP-NEXT: v_cvt_u32_f32_e32 v7, v4
; CGP-NEXT: v_mov_b32_e32 v4, 0xffed2705
; CGP-NEXT: v_cvt_u32_f32_e32 v8, v6
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v4, v7, 0
; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v8, v[5:6]
; CGP-NEXT: v_mul_hi_u32 v9, v7, v4
; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], -1, v7, v[5:6]
; CGP-NEXT: v_mul_lo_u32 v6, v8, v4
; CGP-NEXT: v_mul_hi_u32 v4, v8, v4
; CGP-NEXT: v_mul_lo_u32 v10, v7, v5
; CGP-NEXT: v_mul_lo_u32 v11, v8, v5
; CGP-NEXT: v_mul_hi_u32 v12, v7, v5
; CGP-NEXT: v_mul_hi_u32 v5, v8, v5
; CGP-NEXT: v_add_i32_e32 v6, vcc, v6, v10
; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v6, v9
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v10, v6
; CGP-NEXT: v_add_i32_e32 v4, vcc, v11, v4
; CGP-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v12
; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v9, vcc, v9, v10
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v9, v6
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; CGP-NEXT: v_add_i32_e32 v9, vcc, v7, v4
; CGP-NEXT: v_addc_u32_e32 v8, vcc, v8, v5, vcc
; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], s6, v9, 0
; CGP-NEXT: v_mov_b32_e32 v4, v6
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v8, v[4:5]
; CGP-NEXT: v_ashrrev_i32_e32 v4, 31, v1
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v4
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v9, v[6:7]
; CGP-NEXT: v_addc_u32_e32 v1, vcc, v1, v4, vcc
; CGP-NEXT: v_xor_b32_e32 v10, v0, v4
; CGP-NEXT: v_mul_lo_u32 v0, v8, v5
; CGP-NEXT: v_mul_lo_u32 v7, v9, v6
; CGP-NEXT: v_xor_b32_e32 v11, v1, v4
; CGP-NEXT: v_mul_hi_u32 v1, v9, v5
; CGP-NEXT: v_mul_hi_u32 v5, v8, v5
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v1, v8, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
; CGP-NEXT: v_mul_hi_u32 v7, v9, v6
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v5
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v7
; CGP-NEXT: v_mul_hi_u32 v6, v8, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v5, v1
; CGP-NEXT: v_add_i32_e32 v1, vcc, v6, v1
; CGP-NEXT: v_add_i32_e32 v0, vcc, v9, v0
; CGP-NEXT: v_addc_u32_e32 v1, vcc, v8, v1, vcc
; CGP-NEXT: v_mul_lo_u32 v5, v11, v0
; CGP-NEXT: v_mul_lo_u32 v6, v10, v1
; CGP-NEXT: v_mul_hi_u32 v7, v10, v0
; CGP-NEXT: v_mul_hi_u32 v0, v11, v0
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v7
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v7, v11, v1
; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; CGP-NEXT: v_mul_hi_u32 v6, v10, v1
; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v7, v6
; CGP-NEXT: v_add_i32_e32 v8, vcc, v0, v5
; CGP-NEXT: v_mul_hi_u32 v7, v11, v1
; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s7, v8, 0
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; CGP-NEXT: v_add_i32_e32 v9, vcc, v7, v5
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v9, v[1:2]
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v10, v0
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v8, v[6:7]
; CGP-NEXT: v_mov_b32_e32 v5, 0x12d8fb
; CGP-NEXT: v_mov_b32_e32 v10, s8
; CGP-NEXT: v_subb_u32_e64 v1, s[4:5], v11, v6, vcc
; CGP-NEXT: v_sub_i32_e64 v6, s[4:5], v11, v6
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v5
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1
; CGP-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v6, vcc
; CGP-NEXT: v_cvt_f32_u32_e32 v6, 0x12d8fb
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v5
; CGP-NEXT: v_cndmask_b32_e64 v10, v10, v7, s[4:5]
; CGP-NEXT: v_subbrev_u32_e32 v7, vcc, 0, v1, vcc
; CGP-NEXT: v_cvt_f32_ubyte0_e32 v1, 0
; CGP-NEXT: v_mac_f32_e32 v6, 0x4f800000, v1
; CGP-NEXT: v_rcp_iflag_f32_e32 v1, v6
; CGP-NEXT: v_add_i32_e32 v11, vcc, 1, v8
; CGP-NEXT: v_addc_u32_e32 v12, vcc, 0, v9, vcc
; CGP-NEXT: v_mul_f32_e32 v1, 0x5f7ffffc, v1
; CGP-NEXT: v_mul_f32_e32 v6, 0x2f800000, v1
; CGP-NEXT: v_trunc_f32_e32 v6, v6
; CGP-NEXT: v_mac_f32_e32 v1, 0xcf800000, v6
; CGP-NEXT: v_cvt_u32_f32_e32 v13, v1
; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v5
; CGP-NEXT: v_mov_b32_e32 v15, s4
; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v13, 0
; CGP-NEXT: v_cvt_u32_f32_e32 v16, v6
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v7
; CGP-NEXT: v_cndmask_b32_e32 v14, v15, v14, vcc
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v16, v[1:2]
; CGP-NEXT: v_add_i32_e32 v1, vcc, 1, v11
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v13, v[6:7]
; CGP-NEXT: v_addc_u32_e32 v15, vcc, 0, v12, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14
; CGP-NEXT: v_cndmask_b32_e32 v7, v11, v1, vcc
; CGP-NEXT: v_mul_lo_u32 v1, v16, v0
; CGP-NEXT: v_mul_lo_u32 v11, v13, v6
; CGP-NEXT: v_mul_hi_u32 v14, v13, v0
; CGP-NEXT: v_cndmask_b32_e32 v12, v12, v15, vcc
; CGP-NEXT: v_mul_hi_u32 v0, v16, v0
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v11
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v14
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v14, v16, v6
; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
; CGP-NEXT: v_mul_hi_u32 v11, v13, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v14, v0
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v11
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v11, vcc, v14, v11
; CGP-NEXT: v_mul_hi_u32 v6, v16, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
; CGP-NEXT: v_add_i32_e32 v1, vcc, v6, v1
; CGP-NEXT: v_add_i32_e32 v11, vcc, v13, v0
; CGP-NEXT: v_addc_u32_e32 v13, vcc, v16, v1, vcc
; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v11, 0
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
; CGP-NEXT: v_cndmask_b32_e32 v8, v8, v7, vcc
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v13, v[1:2]
; CGP-NEXT: v_xor_b32_e32 v1, v8, v4
; CGP-NEXT: v_ashrrev_i32_e32 v8, 31, v3
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v11, v[6:7]
; CGP-NEXT: v_cndmask_b32_e32 v9, v9, v12, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v8
; CGP-NEXT: v_addc_u32_e32 v3, vcc, v3, v8, vcc
; CGP-NEXT: v_xor_b32_e32 v10, v2, v8
; CGP-NEXT: v_mul_lo_u32 v2, v13, v0
; CGP-NEXT: v_mul_lo_u32 v7, v11, v6
; CGP-NEXT: v_xor_b32_e32 v12, v3, v8
; CGP-NEXT: v_mul_hi_u32 v3, v11, v0
; CGP-NEXT: v_mul_hi_u32 v0, v13, v0
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v3, v13, v6
; CGP-NEXT: v_add_i32_e32 v2, vcc, v7, v2
; CGP-NEXT: v_mul_hi_u32 v7, v11, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v7
; CGP-NEXT: v_mul_hi_u32 v6, v13, v6
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CGP-NEXT: v_add_i32_e32 v2, vcc, v6, v2
; CGP-NEXT: v_add_i32_e32 v0, vcc, v11, v0
; CGP-NEXT: v_addc_u32_e32 v2, vcc, v13, v2, vcc
; CGP-NEXT: v_mul_lo_u32 v3, v12, v0
; CGP-NEXT: v_mul_lo_u32 v6, v10, v2
; CGP-NEXT: v_mul_hi_u32 v7, v10, v0
; CGP-NEXT: v_mul_hi_u32 v0, v12, v0
; CGP-NEXT: v_xor_b32_e32 v9, v9, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v7
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v7, v12, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v6, v3
; CGP-NEXT: v_mul_hi_u32 v6, v10, v2
; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v6
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v6, vcc, v7, v6
; CGP-NEXT: v_add_i32_e32 v11, vcc, v0, v3
; CGP-NEXT: v_mul_hi_u32 v7, v12, v2
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s7, v11, 0
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v0, vcc, v6, v0
; CGP-NEXT: v_add_i32_e32 v13, vcc, v7, v0
; CGP-NEXT: v_mov_b32_e32 v0, v3
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v13, v[0:1]
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v1, v4
; CGP-NEXT: v_subb_u32_e32 v1, vcc, v9, v4, vcc
; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], 0, v11, v[6:7]
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v10, v2
; CGP-NEXT: v_subb_u32_e64 v4, s[4:5], v12, v3, vcc
; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v12, v3
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; CGP-NEXT: s_bfe_i32 s6, -1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
; CGP-NEXT: v_mov_b32_e32 v7, s6
; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; CGP-NEXT: v_cndmask_b32_e64 v4, v7, v6, s[4:5]
; CGP-NEXT: v_add_i32_e32 v6, vcc, 1, v11
; CGP-NEXT: v_addc_u32_e32 v7, vcc, 0, v13, vcc
; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; CGP-NEXT: v_mov_b32_e32 v5, s4
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3
; CGP-NEXT: v_cndmask_b32_e32 v2, v5, v2, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, 1, v6
; CGP-NEXT: v_addc_u32_e32 v5, vcc, 0, v7, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; CGP-NEXT: v_cndmask_b32_e32 v2, v6, v3, vcc
; CGP-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; CGP-NEXT: v_cndmask_b32_e32 v2, v11, v2, vcc
; CGP-NEXT: v_cndmask_b32_e32 v3, v13, v3, vcc
; CGP-NEXT: v_xor_b32_e32 v2, v2, v8
; CGP-NEXT: v_xor_b32_e32 v3, v3, v8
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v8
; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v8, vcc
; CGP-NEXT: s_setpc_b64 s[30:31]
%result = sdiv <2 x i64> %num, <i64 1235195, i64 1235195>
ret <2 x i64> %result
}
define i64 @v_sdiv_i64_pow2_shl_denom(i64 %x, i64 %y) {
; CHECK-LABEL: v_sdiv_i64_pow2_shl_denom:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: s_mov_b64 s[4:5], 0x1000
; CHECK-NEXT: v_lshl_b64 v[5:6], s[4:5], v2
; CHECK-NEXT: v_mov_b32_e32 v4, v1
; CHECK-NEXT: v_mov_b32_e32 v3, v0
; CHECK-NEXT: v_or_b32_e32 v1, v4, v6
; CHECK-NEXT: v_mov_b32_e32 v0, 0
; CHECK-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1
; CHECK-NEXT: s_and_saveexec_b64 s[4:5], vcc
; CHECK-NEXT: s_xor_b64 s[6:7], exec, s[4:5]
; CHECK-NEXT: s_cbranch_execz .LBB7_2
; CHECK-NEXT: ; %bb.1:
; CHECK-NEXT: v_ashrrev_i32_e32 v0, 31, v6
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v5, v0
; CHECK-NEXT: v_addc_u32_e32 v2, vcc, v6, v0, vcc
; CHECK-NEXT: v_xor_b32_e32 v7, v1, v0
; CHECK-NEXT: v_xor_b32_e32 v8, v2, v0
; CHECK-NEXT: v_cvt_f32_u32_e32 v1, v7
; CHECK-NEXT: v_cvt_f32_u32_e32 v2, v8
; CHECK-NEXT: v_sub_i32_e32 v10, vcc, 0, v7
; CHECK-NEXT: v_subb_u32_e32 v11, vcc, 0, v8, vcc
; CHECK-NEXT: v_mac_f32_e32 v1, 0x4f800000, v2
; CHECK-NEXT: v_rcp_iflag_f32_e32 v1, v1
; CHECK-NEXT: v_mul_f32_e32 v1, 0x5f7ffffc, v1
; CHECK-NEXT: v_mul_f32_e32 v2, 0x2f800000, v1
; CHECK-NEXT: v_trunc_f32_e32 v5, v2
; CHECK-NEXT: v_mac_f32_e32 v1, 0xcf800000, v5
; CHECK-NEXT: v_cvt_u32_f32_e32 v9, v1
; CHECK-NEXT: v_cvt_u32_f32_e32 v12, v5
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v10, v9, 0
; CHECK-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v10, v12, v[2:3]
; CHECK-NEXT: v_mul_lo_u32 v2, v12, v1
; CHECK-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v9, v[5:6]
; CHECK-NEXT: v_mul_hi_u32 v6, v9, v1
; CHECK-NEXT: v_mul_hi_u32 v1, v12, v1
; CHECK-NEXT: v_mul_lo_u32 v13, v9, v5
; CHECK-NEXT: v_mul_lo_u32 v14, v12, v5
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v13
; CHECK-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v6
; CHECK-NEXT: v_mul_hi_u32 v6, v9, v5
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v13, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v14, v1
; CHECK-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v6
; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v6, vcc, v13, v6
; CHECK-NEXT: v_mul_hi_u32 v5, v12, v5
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v2
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v6, v2
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; CHECK-NEXT: v_add_i32_e32 v9, vcc, v9, v1
; CHECK-NEXT: v_addc_u32_e32 v12, vcc, v12, v2, vcc
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v10, v9, 0
; CHECK-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v10, v12, v[2:3]
; CHECK-NEXT: v_ashrrev_i32_e32 v10, 31, v4
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v10
; CHECK-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v9, v[5:6]
; CHECK-NEXT: v_addc_u32_e32 v3, vcc, v4, v10, vcc
; CHECK-NEXT: v_xor_b32_e32 v4, v2, v10
; CHECK-NEXT: v_mul_lo_u32 v2, v12, v1
; CHECK-NEXT: v_mul_lo_u32 v6, v9, v5
; CHECK-NEXT: v_xor_b32_e32 v11, v3, v10
; CHECK-NEXT: v_mul_hi_u32 v3, v9, v1
; CHECK-NEXT: v_mul_hi_u32 v1, v12, v1
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v6
; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v3, v12, v5
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v6, v2
; CHECK-NEXT: v_mul_hi_u32 v6, v9, v5
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v3, v1
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v6
; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; CHECK-NEXT: v_mul_hi_u32 v5, v12, v5
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v2
; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CHECK-NEXT: v_add_i32_e32 v2, vcc, v5, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v9, v1
; CHECK-NEXT: v_addc_u32_e32 v2, vcc, v12, v2, vcc
; CHECK-NEXT: v_mul_lo_u32 v3, v11, v1
; CHECK-NEXT: v_mul_lo_u32 v5, v4, v2
; CHECK-NEXT: v_mul_hi_u32 v6, v4, v1
; CHECK-NEXT: v_mul_hi_u32 v1, v11, v1
; CHECK-NEXT: v_mul_hi_u32 v9, v11, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v5
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_mul_lo_u32 v6, v11, v2
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; CHECK-NEXT: v_mul_hi_u32 v5, v4, v2
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v6, v1
; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v5
; CHECK-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; CHECK-NEXT: v_add_i32_e32 v6, vcc, v1, v3
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v7, v6, 0
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; CHECK-NEXT: v_add_i32_e32 v5, vcc, v9, v3
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v7, v5, v[2:3]
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, v4, v1
; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v8, v6, v[2:3]
; CHECK-NEXT: v_subb_u32_e64 v3, s[4:5], v11, v2, vcc
; CHECK-NEXT: v_sub_i32_e64 v2, s[4:5], v11, v2
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v3, v8
; CHECK-NEXT: v_subb_u32_e32 v2, vcc, v2, v8, vcc
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, -1, s[4:5]
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v7
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, v1, v7
; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, -1, s[4:5]
; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], v3, v8
; CHECK-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
; CHECK-NEXT: v_cndmask_b32_e64 v3, v4, v9, s[4:5]
; CHECK-NEXT: v_add_i32_e32 v4, vcc, 1, v6
; CHECK-NEXT: v_addc_u32_e32 v9, vcc, 0, v5, vcc
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v2, v8
; CHECK-NEXT: v_cndmask_b32_e64 v11, 0, -1, vcc
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v1, v7
; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, -1, vcc
; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, v2, v8
; CHECK-NEXT: v_cndmask_b32_e32 v1, v11, v1, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, 1, v4
; CHECK-NEXT: v_addc_u32_e32 v7, vcc, 0, v9, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; CHECK-NEXT: v_cndmask_b32_e32 v1, v4, v2, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v2, v9, v7, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; CHECK-NEXT: v_cndmask_b32_e32 v1, v6, v1, vcc
; CHECK-NEXT: v_xor_b32_e32 v3, v10, v0
; CHECK-NEXT: v_cndmask_b32_e32 v2, v5, v2, vcc
; CHECK-NEXT: v_xor_b32_e32 v0, v1, v3
; CHECK-NEXT: v_xor_b32_e32 v1, v2, v3
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
; CHECK-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
; CHECK-NEXT: ; implicit-def: $vgpr5_vgpr6
; CHECK-NEXT: ; implicit-def: $vgpr3
; CHECK-NEXT: .LBB7_2: ; %Flow
; CHECK-NEXT: s_or_saveexec_b64 s[6:7], s[6:7]
; CHECK-NEXT: s_xor_b64 exec, exec, s[6:7]
; CHECK-NEXT: s_cbranch_execz .LBB7_4
; CHECK-NEXT: ; %bb.3:
; CHECK-NEXT: v_cvt_f32_u32_e32 v0, v5
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, 0, v5
; CHECK-NEXT: v_rcp_iflag_f32_e32 v0, v0
; CHECK-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0
; CHECK-NEXT: v_cvt_u32_f32_e32 v0, v0
; CHECK-NEXT: v_mul_lo_u32 v1, v1, v0
; CHECK-NEXT: v_mul_hi_u32 v1, v0, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CHECK-NEXT: v_mul_hi_u32 v0, v3, v0
; CHECK-NEXT: v_mul_lo_u32 v1, v0, v5
; CHECK-NEXT: v_add_i32_e32 v2, vcc, 1, v0
; CHECK-NEXT: v_sub_i32_e32 v1, vcc, v3, v1
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v1, v5
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; CHECK-NEXT: v_sub_i32_e64 v2, s[4:5], v1, v5
; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc
; CHECK-NEXT: v_add_i32_e32 v2, vcc, 1, v0
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v1, v5
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; CHECK-NEXT: v_mov_b32_e32 v1, 0
; CHECK-NEXT: .LBB7_4:
; CHECK-NEXT: s_or_b64 exec, exec, s[6:7]
; CHECK-NEXT: s_setpc_b64 s[30:31]
%shl.y = shl i64 4096, %y
%r = sdiv i64 %x, %shl.y
ret i64 %r
}
define <2 x i64> @v_sdiv_v2i64_pow2_shl_denom(<2 x i64> %x, <2 x i64> %y) {
; GISEL-LABEL: v_sdiv_v2i64_pow2_shl_denom:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: s_mov_b64 s[4:5], 0x1000
; GISEL-NEXT: v_lshl_b64 v[7:8], s[4:5], v4
; GISEL-NEXT: v_ashrrev_i32_e32 v4, 31, v8
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v4
; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v8, v4, vcc
; GISEL-NEXT: v_xor_b32_e32 v7, v5, v4
; GISEL-NEXT: v_xor_b32_e32 v5, v8, v4
; GISEL-NEXT: v_cvt_f32_u32_e32 v8, v7
; GISEL-NEXT: v_cvt_f32_u32_e32 v9, v5
; GISEL-NEXT: v_sub_i32_e32 v12, vcc, 0, v7
; GISEL-NEXT: v_subb_u32_e32 v13, vcc, 0, v5, vcc
; GISEL-NEXT: v_mac_f32_e32 v8, 0x4f800000, v9
; GISEL-NEXT: v_rcp_iflag_f32_e32 v8, v8
; GISEL-NEXT: v_mul_f32_e32 v8, 0x5f7ffffc, v8
; GISEL-NEXT: v_mul_f32_e32 v9, 0x2f800000, v8
; GISEL-NEXT: v_trunc_f32_e32 v10, v9
; GISEL-NEXT: v_mac_f32_e32 v8, 0xcf800000, v10
; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v8
; GISEL-NEXT: v_cvt_u32_f32_e32 v14, v10
; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[6:7], v12, v11, 0
; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[6:7], v12, v14, v[9:10]
; GISEL-NEXT: v_mul_hi_u32 v16, v11, v8
; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[6:7], v13, v11, v[9:10]
; GISEL-NEXT: v_mul_lo_u32 v10, v14, v8
; GISEL-NEXT: v_mul_hi_u32 v8, v14, v8
; GISEL-NEXT: v_mul_lo_u32 v15, v11, v9
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v15
; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v16
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v16, v14, v9
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v15, v10
; GISEL-NEXT: v_mul_hi_u32 v15, v11, v9
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v16, v8
; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v15
; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v16, v15
; GISEL-NEXT: v_mul_hi_u32 v9, v14, v9
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v15, v10
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v10
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v11, v8
; GISEL-NEXT: v_addc_u32_e32 v14, vcc, v14, v9, vcc
; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[6:7], v12, v15, 0
; GISEL-NEXT: v_mov_b32_e32 v8, v10
; GISEL-NEXT: v_mad_u64_u32 v[10:11], s[6:7], v12, v14, v[8:9]
; GISEL-NEXT: v_ashrrev_i32_e32 v8, 31, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v8
; GISEL-NEXT: v_mad_u64_u32 v[10:11], s[6:7], v13, v15, v[10:11]
; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v1, v8, vcc
; GISEL-NEXT: v_xor_b32_e32 v13, v0, v8
; GISEL-NEXT: v_mul_lo_u32 v0, v14, v9
; GISEL-NEXT: v_mul_lo_u32 v11, v15, v10
; GISEL-NEXT: v_xor_b32_e32 v16, v1, v8
; GISEL-NEXT: v_mul_hi_u32 v1, v15, v9
; GISEL-NEXT: v_mul_hi_u32 v9, v14, v9
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v11
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v1, v14, v10
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v11, v0
; GISEL-NEXT: v_mul_hi_u32 v11, v15, v10
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v9
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v11
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v11
; GISEL-NEXT: v_mul_hi_u32 v10, v14, v10
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v9, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v15, v0
; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v14, v1, vcc
; GISEL-NEXT: v_mul_lo_u32 v9, v16, v0
; GISEL-NEXT: v_mul_lo_u32 v10, v13, v1
; GISEL-NEXT: v_mul_hi_u32 v11, v13, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v16, v0
; GISEL-NEXT: v_xor_b32_e32 v8, v8, v4
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v11
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v11, v16, v1
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v9
; GISEL-NEXT: v_mul_hi_u32 v10, v13, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v11, v0
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10
; GISEL-NEXT: v_add_i32_e32 v14, vcc, v0, v9
; GISEL-NEXT: v_mul_hi_u32 v11, v16, v1
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[6:7], v7, v14, 0
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v9
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v11, v9
; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[6:7], v7, v15, v[1:2]
; GISEL-NEXT: v_lshl_b64 v[11:12], s[4:5], v6
; GISEL-NEXT: v_ashrrev_i32_e32 v6, 31, v12
; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[4:5], v5, v14, v[9:10]
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v11, v6
; GISEL-NEXT: v_addc_u32_e32 v10, vcc, v12, v6, vcc
; GISEL-NEXT: v_xor_b32_e32 v11, v1, v6
; GISEL-NEXT: v_xor_b32_e32 v12, v10, v6
; GISEL-NEXT: v_cvt_f32_u32_e32 v1, v11
; GISEL-NEXT: v_cvt_f32_u32_e32 v10, v12
; GISEL-NEXT: v_sub_i32_e32 v13, vcc, v13, v0
; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], v16, v9
; GISEL-NEXT: v_mac_f32_e32 v1, 0x4f800000, v10
; GISEL-NEXT: v_rcp_iflag_f32_e32 v1, v1
; GISEL-NEXT: v_subb_u32_e64 v17, s[4:5], v16, v9, vcc
; GISEL-NEXT: v_subb_u32_e32 v9, vcc, v0, v5, vcc
; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v1
; GISEL-NEXT: v_mul_f32_e32 v1, 0x2f800000, v0
; GISEL-NEXT: v_trunc_f32_e32 v10, v1
; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v10
; GISEL-NEXT: v_cvt_u32_f32_e32 v16, v0
; GISEL-NEXT: v_sub_i32_e32 v18, vcc, 0, v11
; GISEL-NEXT: v_cvt_u32_f32_e32 v21, v10
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[6:7], v18, v16, 0
; GISEL-NEXT: v_subb_u32_e32 v19, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v13, v7
; GISEL-NEXT: v_sub_i32_e32 v13, vcc, v13, v7
; GISEL-NEXT: v_subbrev_u32_e32 v20, vcc, 0, v9, vcc
; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[6:7], v18, v21, v[1:2]
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v13, v7
; GISEL-NEXT: v_mul_lo_u32 v1, v21, v0
; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[6:7], v19, v16, v[9:10]
; GISEL-NEXT: v_cmp_ge_u32_e64 s[8:9], v20, v5
; GISEL-NEXT: v_mul_lo_u32 v7, v16, v9
; GISEL-NEXT: v_add_i32_e64 v1, s[6:7], v1, v7
; GISEL-NEXT: v_mul_hi_u32 v7, v16, v0
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, s[6:7]
; GISEL-NEXT: v_mul_hi_u32 v0, v21, v0
; GISEL-NEXT: v_add_i32_e64 v1, s[6:7], v1, v7
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, -1, s[8:9]
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, v20, v5
; GISEL-NEXT: v_cmp_ge_u32_e64 s[8:9], v17, v5
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, -1, s[8:9]
; GISEL-NEXT: v_cndmask_b32_e64 v20, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], v17, v5
; GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v7, vcc
; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v14
; GISEL-NEXT: v_cndmask_b32_e64 v5, v13, v20, s[4:5]
; GISEL-NEXT: v_addc_u32_e32 v13, vcc, 0, v15, vcc
; GISEL-NEXT: v_add_i32_e32 v17, vcc, 1, v7
; GISEL-NEXT: v_addc_u32_e32 v20, vcc, 0, v13, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, s[6:7]
; GISEL-NEXT: v_cndmask_b32_e32 v7, v7, v17, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v13, v13, v20, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; GISEL-NEXT: v_mul_lo_u32 v10, v21, v9
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v10, v16, v9
; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
; GISEL-NEXT: v_mul_hi_u32 v9, v21, v9
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v10
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v17, v10
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v10, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v9, v1
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v16, v0
; GISEL-NEXT: v_addc_u32_e32 v10, vcc, v21, v1, vcc
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v18, v9, 0
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
; GISEL-NEXT: v_cndmask_b32_e32 v7, v14, v7, vcc
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v18, v10, v[1:2]
; GISEL-NEXT: v_xor_b32_e32 v1, v7, v8
; GISEL-NEXT: v_ashrrev_i32_e32 v7, 31, v3
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v19, v9, v[4:5]
; GISEL-NEXT: v_cndmask_b32_e32 v13, v15, v13, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v7
; GISEL-NEXT: v_addc_u32_e32 v3, vcc, v3, v7, vcc
; GISEL-NEXT: v_xor_b32_e32 v5, v2, v7
; GISEL-NEXT: v_mul_lo_u32 v2, v10, v0
; GISEL-NEXT: v_mul_lo_u32 v14, v9, v4
; GISEL-NEXT: v_xor_b32_e32 v15, v3, v7
; GISEL-NEXT: v_mul_hi_u32 v3, v9, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v10, v0
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v14
; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v3, v10, v4
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v14, v2
; GISEL-NEXT: v_mul_hi_u32 v14, v9, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v14
; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v14
; GISEL-NEXT: v_mul_hi_u32 v4, v10, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v4, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
; GISEL-NEXT: v_addc_u32_e32 v2, vcc, v10, v2, vcc
; GISEL-NEXT: v_mul_lo_u32 v3, v15, v0
; GISEL-NEXT: v_mul_lo_u32 v4, v5, v2
; GISEL-NEXT: v_mul_hi_u32 v10, v5, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v15, v0
; GISEL-NEXT: v_xor_b32_e32 v9, v13, v8
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v4
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v10
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v10, v15, v2
; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3
; GISEL-NEXT: v_mul_hi_u32 v4, v5, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0
; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v10, v4
; GISEL-NEXT: v_add_i32_e32 v10, vcc, v0, v3
; GISEL-NEXT: v_mul_hi_u32 v13, v15, v2
; GISEL-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v11, v10, 0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v3
; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v11, v13, v[0:1]
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v1, v8
; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v12, v10, v[3:4]
; GISEL-NEXT: v_subb_u32_e32 v1, vcc, v9, v8, vcc
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v5, v2
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], v15, v3, vcc
; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v15, v3
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v4, v12
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v12, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v11
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v11
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], v4, v12
; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v4, v5, v8, s[4:5]
; GISEL-NEXT: v_add_i32_e32 v5, vcc, 1, v10
; GISEL-NEXT: v_addc_u32_e32 v8, vcc, 0, v13, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v3, v12
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, -1, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v2, v11
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, v3, v12
; GISEL-NEXT: v_cndmask_b32_e32 v2, v9, v2, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v5
; GISEL-NEXT: v_addc_u32_e32 v9, vcc, 0, v8, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; GISEL-NEXT: v_cndmask_b32_e32 v2, v5, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v8, v9, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; GISEL-NEXT: v_cndmask_b32_e32 v2, v10, v2, vcc
; GISEL-NEXT: v_xor_b32_e32 v4, v7, v6
; GISEL-NEXT: v_cndmask_b32_e32 v3, v13, v3, vcc
; GISEL-NEXT: v_xor_b32_e32 v2, v2, v4
; GISEL-NEXT: v_xor_b32_e32 v3, v3, v4
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v4
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v4, vcc
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_v2i64_pow2_shl_denom:
; CGP: ; %bb.0:
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CGP-NEXT: s_mov_b64 s[6:7], 0x1000
; CGP-NEXT: v_mov_b32_e32 v5, v2
; CGP-NEXT: v_mov_b32_e32 v7, v3
; CGP-NEXT: v_lshl_b64 v[2:3], s[6:7], v4
; CGP-NEXT: v_mov_b32_e32 v9, v1
; CGP-NEXT: v_mov_b32_e32 v8, v0
; CGP-NEXT: v_or_b32_e32 v1, v9, v3
; CGP-NEXT: v_mov_b32_e32 v0, 0
; CGP-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[0:1]
; CGP-NEXT: ; implicit-def: $vgpr0_vgpr1
; CGP-NEXT: s_and_saveexec_b64 s[4:5], vcc
; CGP-NEXT: s_xor_b64 s[8:9], exec, s[4:5]
; CGP-NEXT: s_cbranch_execz .LBB8_2
; CGP-NEXT: ; %bb.1:
; CGP-NEXT: v_ashrrev_i32_e32 v0, 31, v3
; CGP-NEXT: v_add_i32_e32 v1, vcc, v2, v0
; CGP-NEXT: v_addc_u32_e32 v2, vcc, v3, v0, vcc
; CGP-NEXT: v_xor_b32_e32 v4, v1, v0
; CGP-NEXT: v_xor_b32_e32 v10, v2, v0
; CGP-NEXT: v_cvt_f32_u32_e32 v1, v4
; CGP-NEXT: v_cvt_f32_u32_e32 v2, v10
; CGP-NEXT: v_sub_i32_e32 v12, vcc, 0, v4
; CGP-NEXT: v_subb_u32_e32 v13, vcc, 0, v10, vcc
; CGP-NEXT: v_mac_f32_e32 v1, 0x4f800000, v2
; CGP-NEXT: v_rcp_iflag_f32_e32 v1, v1
; CGP-NEXT: v_mul_f32_e32 v1, 0x5f7ffffc, v1
; CGP-NEXT: v_mul_f32_e32 v2, 0x2f800000, v1
; CGP-NEXT: v_trunc_f32_e32 v3, v2
; CGP-NEXT: v_mac_f32_e32 v1, 0xcf800000, v3
; CGP-NEXT: v_cvt_u32_f32_e32 v11, v1
; CGP-NEXT: v_cvt_u32_f32_e32 v14, v3
; CGP-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v12, v11, 0
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v12, v14, v[2:3]
; CGP-NEXT: v_mul_hi_u32 v15, v11, v1
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v13, v11, v[2:3]
; CGP-NEXT: v_mul_lo_u32 v3, v14, v1
; CGP-NEXT: v_mul_hi_u32 v1, v14, v1
; CGP-NEXT: v_mul_lo_u32 v16, v11, v2
; CGP-NEXT: v_mul_lo_u32 v17, v14, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v16
; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v15
; CGP-NEXT: v_mul_hi_u32 v15, v11, v2
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v16, v3
; CGP-NEXT: v_add_i32_e32 v1, vcc, v17, v1
; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v15
; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v15, vcc, v16, v15
; CGP-NEXT: v_mul_hi_u32 v2, v14, v2
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v3
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v15, v3
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v1
; CGP-NEXT: v_addc_u32_e32 v14, vcc, v14, v2, vcc
; CGP-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v12, v11, 0
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v12, v14, v[2:3]
; CGP-NEXT: v_ashrrev_i32_e32 v12, 31, v9
; CGP-NEXT: v_mul_hi_u32 v15, v11, v1
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v13, v11, v[2:3]
; CGP-NEXT: v_add_i32_e32 v3, vcc, v8, v12
; CGP-NEXT: v_addc_u32_e32 v8, vcc, v9, v12, vcc
; CGP-NEXT: v_xor_b32_e32 v9, v3, v12
; CGP-NEXT: v_mul_lo_u32 v3, v14, v1
; CGP-NEXT: v_mul_lo_u32 v13, v11, v2
; CGP-NEXT: v_mul_hi_u32 v1, v14, v1
; CGP-NEXT: v_xor_b32_e32 v8, v8, v12
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v13
; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v15
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v15, v14, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v13, v3
; CGP-NEXT: v_mul_hi_u32 v13, v11, v2
; CGP-NEXT: v_add_i32_e32 v1, vcc, v15, v1
; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v13
; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v13, vcc, v15, v13
; CGP-NEXT: v_mul_hi_u32 v2, v14, v2
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v3
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v13, v3
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
; CGP-NEXT: v_addc_u32_e32 v2, vcc, v14, v2, vcc
; CGP-NEXT: v_mul_lo_u32 v3, v8, v1
; CGP-NEXT: v_mul_lo_u32 v11, v9, v2
; CGP-NEXT: v_mul_hi_u32 v13, v9, v1
; CGP-NEXT: v_mul_hi_u32 v1, v8, v1
; CGP-NEXT: v_mul_hi_u32 v14, v8, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v11
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v13
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v13, v8, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, v11, v3
; CGP-NEXT: v_mul_hi_u32 v11, v9, v2
; CGP-NEXT: v_add_i32_e32 v1, vcc, v13, v1
; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v11
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v11, vcc, v13, v11
; CGP-NEXT: v_add_i32_e32 v13, vcc, v1, v3
; CGP-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v4, v13, 0
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v11, v3
; CGP-NEXT: v_add_i32_e32 v11, vcc, v14, v3
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v4, v11, v[2:3]
; CGP-NEXT: v_sub_i32_e32 v1, vcc, v9, v1
; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v10, v13, v[2:3]
; CGP-NEXT: v_subb_u32_e64 v3, s[4:5], v8, v2, vcc
; CGP-NEXT: v_sub_i32_e64 v2, s[4:5], v8, v2
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v3, v10
; CGP-NEXT: v_subb_u32_e32 v2, vcc, v2, v10, vcc
; CGP-NEXT: v_cndmask_b32_e64 v8, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v1, v4
; CGP-NEXT: v_sub_i32_e32 v1, vcc, v1, v4
; CGP-NEXT: v_cndmask_b32_e64 v9, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], v3, v10
; CGP-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
; CGP-NEXT: v_cndmask_b32_e64 v3, v8, v9, s[4:5]
; CGP-NEXT: v_add_i32_e32 v8, vcc, 1, v13
; CGP-NEXT: v_addc_u32_e32 v9, vcc, 0, v11, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v2, v10
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, -1, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v4
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, v2, v10
; CGP-NEXT: v_cndmask_b32_e32 v1, v14, v1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, 1, v8
; CGP-NEXT: v_addc_u32_e32 v4, vcc, 0, v9, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; CGP-NEXT: v_cndmask_b32_e32 v1, v8, v2, vcc
; CGP-NEXT: v_cndmask_b32_e32 v2, v9, v4, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; CGP-NEXT: v_cndmask_b32_e32 v1, v13, v1, vcc
; CGP-NEXT: v_xor_b32_e32 v3, v12, v0
; CGP-NEXT: v_cndmask_b32_e32 v2, v11, v2, vcc
; CGP-NEXT: v_xor_b32_e32 v0, v1, v3
; CGP-NEXT: v_xor_b32_e32 v1, v2, v3
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
; CGP-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc
; CGP-NEXT: ; implicit-def: $vgpr2_vgpr3
; CGP-NEXT: ; implicit-def: $vgpr8
; CGP-NEXT: .LBB8_2: ; %Flow1
; CGP-NEXT: s_or_saveexec_b64 s[8:9], s[8:9]
; CGP-NEXT: v_lshl_b64 v[9:10], s[6:7], v6
; CGP-NEXT: s_xor_b64 exec, exec, s[8:9]
; CGP-NEXT: s_cbranch_execz .LBB8_4
; CGP-NEXT: ; %bb.3:
; CGP-NEXT: v_cvt_f32_u32_e32 v0, v2
; CGP-NEXT: v_sub_i32_e32 v1, vcc, 0, v2
; CGP-NEXT: v_rcp_iflag_f32_e32 v0, v0
; CGP-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0
; CGP-NEXT: v_cvt_u32_f32_e32 v0, v0
; CGP-NEXT: v_mul_lo_u32 v1, v1, v0
; CGP-NEXT: v_mul_hi_u32 v1, v0, v1
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CGP-NEXT: v_mul_hi_u32 v0, v8, v0
; CGP-NEXT: v_mul_lo_u32 v1, v0, v2
; CGP-NEXT: v_add_i32_e32 v3, vcc, 1, v0
; CGP-NEXT: v_sub_i32_e32 v1, vcc, v8, v1
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v2
; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v1, v2
; CGP-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, 1, v0
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v2
; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; CGP-NEXT: v_mov_b32_e32 v1, 0
; CGP-NEXT: .LBB8_4:
; CGP-NEXT: s_or_b64 exec, exec, s[8:9]
; CGP-NEXT: v_or_b32_e32 v3, v7, v10
; CGP-NEXT: v_mov_b32_e32 v2, 0
; CGP-NEXT: v_cmp_ne_u64_e32 vcc, 0, v[2:3]
; CGP-NEXT: ; implicit-def: $vgpr2_vgpr3
; CGP-NEXT: s_and_saveexec_b64 s[4:5], vcc
; CGP-NEXT: s_xor_b64 s[6:7], exec, s[4:5]
; CGP-NEXT: s_cbranch_execz .LBB8_6
; CGP-NEXT: ; %bb.5:
; CGP-NEXT: v_ashrrev_i32_e32 v2, 31, v10
; CGP-NEXT: v_add_i32_e32 v3, vcc, v9, v2
; CGP-NEXT: v_addc_u32_e32 v4, vcc, v10, v2, vcc
; CGP-NEXT: v_xor_b32_e32 v6, v3, v2
; CGP-NEXT: v_xor_b32_e32 v10, v4, v2
; CGP-NEXT: v_cvt_f32_u32_e32 v3, v6
; CGP-NEXT: v_cvt_f32_u32_e32 v4, v10
; CGP-NEXT: v_sub_i32_e32 v12, vcc, 0, v6
; CGP-NEXT: v_subb_u32_e32 v13, vcc, 0, v10, vcc
; CGP-NEXT: v_mac_f32_e32 v3, 0x4f800000, v4
; CGP-NEXT: v_rcp_iflag_f32_e32 v3, v3
; CGP-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3
; CGP-NEXT: v_mul_f32_e32 v4, 0x2f800000, v3
; CGP-NEXT: v_trunc_f32_e32 v8, v4
; CGP-NEXT: v_mac_f32_e32 v3, 0xcf800000, v8
; CGP-NEXT: v_cvt_u32_f32_e32 v11, v3
; CGP-NEXT: v_cvt_u32_f32_e32 v14, v8
; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v12, v11, 0
; CGP-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v12, v14, v[4:5]
; CGP-NEXT: v_mul_lo_u32 v4, v14, v3
; CGP-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v13, v11, v[8:9]
; CGP-NEXT: v_mul_hi_u32 v9, v11, v3
; CGP-NEXT: v_mul_hi_u32 v3, v14, v3
; CGP-NEXT: v_mul_lo_u32 v15, v11, v8
; CGP-NEXT: v_mul_lo_u32 v16, v14, v8
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v15
; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; CGP-NEXT: v_mul_hi_u32 v9, v11, v8
; CGP-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, v15, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v16, v3
; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v9
; CGP-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v9, vcc, v15, v9
; CGP-NEXT: v_mul_hi_u32 v8, v14, v8
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v4
; CGP-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, v9, v4
; CGP-NEXT: v_add_i32_e32 v4, vcc, v8, v4
; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v3
; CGP-NEXT: v_addc_u32_e32 v14, vcc, v14, v4, vcc
; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v12, v11, 0
; CGP-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v12, v14, v[4:5]
; CGP-NEXT: v_ashrrev_i32_e32 v12, 31, v7
; CGP-NEXT: v_add_i32_e32 v4, vcc, v5, v12
; CGP-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v13, v11, v[8:9]
; CGP-NEXT: v_addc_u32_e32 v5, vcc, v7, v12, vcc
; CGP-NEXT: v_xor_b32_e32 v7, v4, v12
; CGP-NEXT: v_mul_lo_u32 v4, v14, v3
; CGP-NEXT: v_mul_lo_u32 v9, v11, v8
; CGP-NEXT: v_xor_b32_e32 v13, v5, v12
; CGP-NEXT: v_mul_hi_u32 v5, v11, v3
; CGP-NEXT: v_mul_hi_u32 v3, v14, v3
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; CGP-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v5
; CGP-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v5, v14, v8
; CGP-NEXT: v_add_i32_e32 v4, vcc, v9, v4
; CGP-NEXT: v_mul_hi_u32 v9, v11, v8
; CGP-NEXT: v_add_i32_e32 v3, vcc, v5, v3
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v9
; CGP-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v9
; CGP-NEXT: v_mul_hi_u32 v8, v14, v8
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v4
; CGP-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, v5, v4
; CGP-NEXT: v_add_i32_e32 v4, vcc, v8, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v11, v3
; CGP-NEXT: v_addc_u32_e32 v4, vcc, v14, v4, vcc
; CGP-NEXT: v_mul_lo_u32 v5, v13, v3
; CGP-NEXT: v_mul_lo_u32 v8, v7, v4
; CGP-NEXT: v_mul_hi_u32 v9, v7, v3
; CGP-NEXT: v_mul_hi_u32 v3, v13, v3
; CGP-NEXT: v_mul_hi_u32 v11, v13, v4
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v8
; CGP-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v9
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_mul_lo_u32 v9, v13, v4
; CGP-NEXT: v_add_i32_e32 v5, vcc, v8, v5
; CGP-NEXT: v_mul_hi_u32 v8, v7, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v9, v3
; CGP-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v8
; CGP-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v8, vcc, v9, v8
; CGP-NEXT: v_add_i32_e32 v9, vcc, v3, v5
; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v6, v9, 0
; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v5, vcc, v8, v5
; CGP-NEXT: v_add_i32_e32 v8, vcc, v11, v5
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v6, v8, v[4:5]
; CGP-NEXT: v_sub_i32_e32 v3, vcc, v7, v3
; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v9, v[4:5]
; CGP-NEXT: v_subb_u32_e64 v5, s[4:5], v13, v4, vcc
; CGP-NEXT: v_sub_i32_e64 v4, s[4:5], v13, v4
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v5, v10
; CGP-NEXT: v_subb_u32_e32 v4, vcc, v4, v10, vcc
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v3, v6
; CGP-NEXT: v_sub_i32_e32 v3, vcc, v3, v6
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, -1, s[4:5]
; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], v5, v10
; CGP-NEXT: v_subbrev_u32_e32 v4, vcc, 0, v4, vcc
; CGP-NEXT: v_cndmask_b32_e64 v5, v7, v11, s[4:5]
; CGP-NEXT: v_add_i32_e32 v7, vcc, 1, v9
; CGP-NEXT: v_addc_u32_e32 v11, vcc, 0, v8, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v4, v10
; CGP-NEXT: v_cndmask_b32_e64 v13, 0, -1, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v3, v6
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, v4, v10
; CGP-NEXT: v_cndmask_b32_e32 v3, v13, v3, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, 1, v7
; CGP-NEXT: v_addc_u32_e32 v6, vcc, 0, v11, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; CGP-NEXT: v_cndmask_b32_e32 v3, v7, v4, vcc
; CGP-NEXT: v_cndmask_b32_e32 v4, v11, v6, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
; CGP-NEXT: v_cndmask_b32_e32 v3, v9, v3, vcc
; CGP-NEXT: v_xor_b32_e32 v5, v12, v2
; CGP-NEXT: v_cndmask_b32_e32 v4, v8, v4, vcc
; CGP-NEXT: v_xor_b32_e32 v2, v3, v5
; CGP-NEXT: v_xor_b32_e32 v3, v4, v5
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v5, vcc
; CGP-NEXT: ; implicit-def: $vgpr9_vgpr10
; CGP-NEXT: ; implicit-def: $vgpr5
; CGP-NEXT: .LBB8_6: ; %Flow
; CGP-NEXT: s_or_saveexec_b64 s[6:7], s[6:7]
; CGP-NEXT: s_xor_b64 exec, exec, s[6:7]
; CGP-NEXT: s_cbranch_execz .LBB8_8
; CGP-NEXT: ; %bb.7:
; CGP-NEXT: v_cvt_f32_u32_e32 v2, v9
; CGP-NEXT: v_sub_i32_e32 v3, vcc, 0, v9
; CGP-NEXT: v_rcp_iflag_f32_e32 v2, v2
; CGP-NEXT: v_mul_f32_e32 v2, 0x4f7ffffe, v2
; CGP-NEXT: v_cvt_u32_f32_e32 v2, v2
; CGP-NEXT: v_mul_lo_u32 v3, v3, v2
; CGP-NEXT: v_mul_hi_u32 v3, v2, v3
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_mul_hi_u32 v2, v5, v2
; CGP-NEXT: v_mul_lo_u32 v3, v2, v9
; CGP-NEXT: v_add_i32_e32 v4, vcc, 1, v2
; CGP-NEXT: v_sub_i32_e32 v3, vcc, v5, v3
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v3, v9
; CGP-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
; CGP-NEXT: v_sub_i32_e64 v4, s[4:5], v3, v9
; CGP-NEXT: v_cndmask_b32_e32 v3, v3, v4, vcc
; CGP-NEXT: v_add_i32_e32 v4, vcc, 1, v2
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v3, v9
; CGP-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
; CGP-NEXT: v_mov_b32_e32 v3, 0
; CGP-NEXT: .LBB8_8:
; CGP-NEXT: s_or_b64 exec, exec, s[6:7]
; CGP-NEXT: s_setpc_b64 s[30:31]
%shl.y = shl <2 x i64> <i64 4096, i64 4096>, %y
%r = sdiv <2 x i64> %x, %shl.y
ret <2 x i64> %r
}
define i64 @v_sdiv_i64_24bit(i64 %num, i64 %den) {
; GISEL-LABEL: v_sdiv_i64_24bit:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v2
; GISEL-NEXT: v_cvt_f32_u32_e32 v2, v1
; GISEL-NEXT: v_sub_i32_e32 v3, vcc, 0, v1
; GISEL-NEXT: v_and_b32_e32 v0, 0xffffff, v0
; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v2
; GISEL-NEXT: v_mul_f32_e32 v2, 0x4f7ffffe, v2
; GISEL-NEXT: v_cvt_u32_f32_e32 v2, v2
; GISEL-NEXT: v_mul_lo_u32 v3, v3, v2
; GISEL-NEXT: v_mul_hi_u32 v3, v2, v3
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; GISEL-NEXT: v_mul_hi_u32 v2, v0, v2
; GISEL-NEXT: v_mul_lo_u32 v3, v2, v1
; GISEL-NEXT: v_add_i32_e32 v4, vcc, 1, v2
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc
; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v0, v1
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v2
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1
; GISEL-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
; GISEL-NEXT: v_mov_b32_e32 v1, 0
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_i64_24bit:
; CGP: ; %bb.0:
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CGP-NEXT: v_and_b32_e32 v1, 0xffffff, v2
; CGP-NEXT: v_cvt_f32_i32_e32 v1, v1
; CGP-NEXT: v_and_b32_e32 v0, 0xffffff, v0
; CGP-NEXT: v_cvt_f32_i32_e32 v0, v0
; CGP-NEXT: v_rcp_f32_e32 v2, v1
; CGP-NEXT: v_mul_f32_e32 v2, v0, v2
; CGP-NEXT: v_trunc_f32_e32 v2, v2
; CGP-NEXT: v_mad_f32 v0, -v2, v1, v0
; CGP-NEXT: v_cvt_i32_f32_e32 v2, v2
; CGP-NEXT: v_cmp_ge_f32_e64 s[4:5], |v0|, |v1|
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
; CGP-NEXT: v_add_i32_e32 v0, vcc, v2, v0
; CGP-NEXT: v_bfe_i32 v0, v0, 0, 25
; CGP-NEXT: v_ashrrev_i32_e32 v1, 31, v0
; CGP-NEXT: s_setpc_b64 s[30:31]
%num.mask = and i64 %num, 16777215
%den.mask = and i64 %den, 16777215
%result = sdiv i64 %num.mask, %den.mask
ret i64 %result
}
define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-LABEL: v_sdiv_v2i64_24bit:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v4
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 0, v1
; GISEL-NEXT: v_addc_u32_e64 v1, s[4:5], 0, 0, vcc
; GISEL-NEXT: v_cvt_f32_u32_e32 v4, v3
; GISEL-NEXT: v_cvt_f32_u32_e32 v5, v1
; GISEL-NEXT: v_sub_i32_e32 v10, vcc, 0, v3
; GISEL-NEXT: v_subb_u32_e32 v11, vcc, 0, v1, vcc
; GISEL-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; GISEL-NEXT: v_rcp_iflag_f32_e32 v4, v4
; GISEL-NEXT: v_and_b32_e32 v0, 0xffffff, v0
; GISEL-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; GISEL-NEXT: v_trunc_f32_e32 v7, v5
; GISEL-NEXT: v_mac_f32_e32 v4, 0xcf800000, v7
; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v4
; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v7
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v9, 0
; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v10, v12, v[5:6]
; GISEL-NEXT: v_mul_lo_u32 v5, v12, v4
; GISEL-NEXT: v_mul_hi_u32 v13, v9, v4
; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v9, v[7:8]
; GISEL-NEXT: v_mul_hi_u32 v4, v12, v4
; GISEL-NEXT: v_mul_lo_u32 v8, v9, v7
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v13, v12, v7
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v8, v5
; GISEL-NEXT: v_mul_hi_u32 v8, v9, v7
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v13, v4
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8
; GISEL-NEXT: v_mul_hi_u32 v7, v12, v7
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v8, v5
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v4
; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v12, v5, vcc
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v9, 0
; GISEL-NEXT: v_and_b32_e32 v13, 0xffffff, v2
; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v10, v12, v[5:6]
; GISEL-NEXT: v_mul_lo_u32 v5, v12, v4
; GISEL-NEXT: v_add_i32_e32 v10, vcc, 0, v0
; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v9, v[7:8]
; GISEL-NEXT: v_mul_hi_u32 v0, v9, v4
; GISEL-NEXT: v_addc_u32_e64 v11, s[4:5], 0, 0, vcc
; GISEL-NEXT: v_mul_lo_u32 v8, v9, v7
; GISEL-NEXT: v_mul_hi_u32 v4, v12, v4
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v5, v12, v7
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
; GISEL-NEXT: v_mul_hi_u32 v8, v9, v7
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v4
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
; GISEL-NEXT: v_mul_hi_u32 v7, v12, v7
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v4
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v7, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0
; GISEL-NEXT: v_addc_u32_e32 v4, vcc, v12, v4, vcc
; GISEL-NEXT: v_mul_lo_u32 v5, v11, v0
; GISEL-NEXT: v_mul_lo_u32 v7, v10, v4
; GISEL-NEXT: v_mul_hi_u32 v8, v10, v0
; GISEL-NEXT: v_mul_hi_u32 v0, v11, v0
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v7
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v8, v11, v4
; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
; GISEL-NEXT: v_mul_hi_u32 v7, v10, v4
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v8, v7
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v0, v5
; GISEL-NEXT: v_mul_hi_u32 v8, v11, v4
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v9, 0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v7, v0
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v8, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v5
; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v3, v12, v[0:1]
; GISEL-NEXT: v_and_b32_e32 v0, 0xffffff, v6
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v10, v4
; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v9, v[7:8]
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], v11, v5, vcc
; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], v11, v5
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v4, v1
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v8, s[4:5], 0, v0
; GISEL-NEXT: v_addc_u32_e64 v10, s[4:5], 0, 0, s[4:5]
; GISEL-NEXT: v_cvt_f32_u32_e32 v0, v8
; GISEL-NEXT: v_cvt_f32_u32_e32 v11, v10
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], v4, v1
; GISEL-NEXT: v_subb_u32_e32 v4, vcc, v5, v1, vcc
; GISEL-NEXT: v_mac_f32_e32 v0, 0x4f800000, v11
; GISEL-NEXT: v_rcp_iflag_f32_e32 v0, v0
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[4:5]
; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v0
; GISEL-NEXT: v_subbrev_u32_e32 v7, vcc, 0, v4, vcc
; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v0
; GISEL-NEXT: v_trunc_f32_e32 v11, v4
; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v11
; GISEL-NEXT: v_cvt_u32_f32_e32 v14, v0
; GISEL-NEXT: v_sub_i32_e32 v15, vcc, 0, v8
; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v11
; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v15, v14, 0
; GISEL-NEXT: v_subb_u32_e32 v16, vcc, 0, v10, vcc
; GISEL-NEXT: v_add_i32_e32 v17, vcc, 1, v9
; GISEL-NEXT: v_addc_u32_e32 v18, vcc, 0, v12, vcc
; GISEL-NEXT: v_mov_b32_e32 v0, v5
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v2, v3
; GISEL-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v15, v11, v[0:1]
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v7, v1
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5]
; GISEL-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v16, v14, v[2:3]
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, v7, v1
; GISEL-NEXT: v_mul_lo_u32 v1, v11, v4
; GISEL-NEXT: v_mul_lo_u32 v5, v14, v2
; GISEL-NEXT: v_cndmask_b32_e32 v3, v0, v3, vcc
; GISEL-NEXT: v_mul_hi_u32 v0, v14, v4
; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v1, v11, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0
; GISEL-NEXT: v_mul_hi_u32 v5, v14, v2
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v4
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v5
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
; GISEL-NEXT: v_mul_hi_u32 v2, v11, v2
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v1, v0
; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v4, v1
; GISEL-NEXT: v_add_i32_e32 v1, vcc, v2, v1
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v14, v0
; GISEL-NEXT: v_addc_u32_e32 v5, vcc, v11, v1, vcc
; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v15, v4, 0
; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v17
; GISEL-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v15, v5, v[1:2]
; GISEL-NEXT: v_addc_u32_e32 v11, vcc, 0, v18, vcc
; GISEL-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v16, v4, v[1:2]
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; GISEL-NEXT: v_cndmask_b32_e32 v3, v17, v7, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v7, v18, v11, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
; GISEL-NEXT: v_cndmask_b32_e32 v6, v9, v3, vcc
; GISEL-NEXT: v_mul_lo_u32 v2, v5, v0
; GISEL-NEXT: v_mul_lo_u32 v3, v4, v1
; GISEL-NEXT: v_add_i32_e64 v9, s[4:5], 0, v13
; GISEL-NEXT: v_mul_hi_u32 v13, v4, v0
; GISEL-NEXT: v_addc_u32_e64 v11, s[4:5], 0, 0, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v2, s[4:5], v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v2, s[4:5], v2, v13
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[4:5]
; GISEL-NEXT: v_mul_lo_u32 v13, v5, v1
; GISEL-NEXT: v_mul_hi_u32 v0, v5, v0
; GISEL-NEXT: v_add_i32_e64 v2, s[4:5], v3, v2
; GISEL-NEXT: v_mul_hi_u32 v3, v4, v1
; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v13, v0
; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v0, v3
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v3, s[4:5], v13, v3
; GISEL-NEXT: v_mul_hi_u32 v1, v5, v1
; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v0, v2
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v2, s[4:5], v3, v2
; GISEL-NEXT: v_add_i32_e64 v1, s[4:5], v1, v2
; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v4, v0
; GISEL-NEXT: v_addc_u32_e64 v1, s[4:5], v5, v1, s[4:5]
; GISEL-NEXT: v_mul_lo_u32 v2, v11, v0
; GISEL-NEXT: v_mul_lo_u32 v3, v9, v1
; GISEL-NEXT: v_mul_hi_u32 v4, v9, v0
; GISEL-NEXT: v_cndmask_b32_e32 v5, v12, v7, vcc
; GISEL-NEXT: v_mul_hi_u32 v0, v11, v0
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v4
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; GISEL-NEXT: v_mul_lo_u32 v4, v11, v1
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; GISEL-NEXT: v_mul_hi_u32 v3, v9, v1
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v3
; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v3
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v0, v2
; GISEL-NEXT: v_mul_hi_u32 v1, v11, v1
; GISEL-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v8, v7, 0
; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v1, v0
; GISEL-NEXT: v_mov_b32_e32 v0, v3
; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v8, v12, v[0:1]
; GISEL-NEXT: v_subrev_i32_e32 v0, vcc, 0, v6
; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v10, v7, v[3:4]
; GISEL-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v5, vcc
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v9, v2
; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], v11, v3, vcc
; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v11, v3
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v4, v10
; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v10, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v8
; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v8
; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], v4, v10
; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e64 v4, v5, v6, s[4:5]
; GISEL-NEXT: v_add_i32_e32 v5, vcc, 1, v7
; GISEL-NEXT: v_addc_u32_e32 v6, vcc, 0, v12, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v3, v10
; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, -1, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v2, v8
; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, v3, v10
; GISEL-NEXT: v_cndmask_b32_e32 v2, v9, v2, vcc
; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v5
; GISEL-NEXT: v_addc_u32_e32 v8, vcc, 0, v6, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
; GISEL-NEXT: v_cndmask_b32_e32 v2, v5, v3, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v6, v8, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; GISEL-NEXT: v_cndmask_b32_e32 v2, v7, v2, vcc
; GISEL-NEXT: v_cndmask_b32_e32 v3, v12, v3, vcc
; GISEL-NEXT: v_subrev_i32_e32 v2, vcc, 0, v2
; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_v2i64_24bit:
; CGP: ; %bb.0:
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CGP-NEXT: v_and_b32_e32 v1, 0xffffff, v4
; CGP-NEXT: v_cvt_f32_i32_e32 v1, v1
; CGP-NEXT: v_and_b32_e32 v0, 0xffffff, v0
; CGP-NEXT: v_cvt_f32_i32_e32 v0, v0
; CGP-NEXT: v_and_b32_e32 v4, 0xffffff, v6
; CGP-NEXT: v_rcp_f32_e32 v3, v1
; CGP-NEXT: v_cvt_f32_i32_e32 v4, v4
; CGP-NEXT: v_and_b32_e32 v2, 0xffffff, v2
; CGP-NEXT: v_cvt_f32_i32_e32 v2, v2
; CGP-NEXT: v_mul_f32_e32 v3, v0, v3
; CGP-NEXT: v_trunc_f32_e32 v3, v3
; CGP-NEXT: v_mad_f32 v0, -v3, v1, v0
; CGP-NEXT: v_cvt_i32_f32_e32 v3, v3
; CGP-NEXT: v_rcp_f32_e32 v5, v4
; CGP-NEXT: v_cmp_ge_f32_e64 s[4:5], |v0|, |v1|
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5]
; CGP-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; CGP-NEXT: v_mul_f32_e32 v3, v2, v5
; CGP-NEXT: v_trunc_f32_e32 v3, v3
; CGP-NEXT: v_mad_f32 v2, -v3, v4, v2
; CGP-NEXT: v_cvt_i32_f32_e32 v3, v3
; CGP-NEXT: v_cmp_ge_f32_e64 s[4:5], |v2|, |v4|
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, s[4:5]
; CGP-NEXT: v_bfe_i32 v0, v0, 0, 25
; CGP-NEXT: v_add_i32_e32 v2, vcc, v3, v2
; CGP-NEXT: v_bfe_i32 v2, v2, 0, 25
; CGP-NEXT: v_ashrrev_i32_e32 v1, 31, v0
; CGP-NEXT: v_ashrrev_i32_e32 v3, 31, v2
; CGP-NEXT: s_setpc_b64 s[30:31]
%num.mask = and <2 x i64> %num, <i64 16777215, i64 16777215>
%den.mask = and <2 x i64> %den, <i64 16777215, i64 16777215>
%result = sdiv <2 x i64> %num.mask, %den.mask
ret <2 x i64> %result
}