/*
    set isa_file=%~1.isa
    set analysis_file=%~1.a
    set isa_file
    rga --define COMPILER_AMD_RGA=1 --source-kind hlsl --asic Pitcairn --profile cs_5_0 --function %2 --intrinsics --isa %isa_file% %1
 */
#if COMPILER_AMD_RGA
    #include "ags_shader_intrinsics_dx11.hlsl"

    uint2 ballot(bool pred)
    {
        return AmdDxExtShaderIntrinsics_Ballot(pred);
    }

    uint ReadFirstLane(uint x)
    {
        return AmdDxExtShaderIntrinsics_ReadfirstlaneU(x);
    }

    float ReadFirstLane(float x)
    {
        return AmdDxExtShaderIntrinsics_ReadfirstlaneF(x);
    }

    uint ReadLane(uint x, uint laneId)
    {
        return AmdDxExtShaderIntrinsics_ReadlaneU(x, laneId);
    }

    bool AnyExecSet()
    {
        return any(ballot(true));
    }

#endif

Texture2D<uint> InTexture : register(t0);
RWTexture2D<uint> OutTexture : register(u0);

[numthreads(8, 8, 1)]
void main(uint2 TexelId : SV_DispatchThreadId)
{
    const uint VgprData = InTexture[TexelId];
    uint NumDivergentGroups = 0;
#if 1
    for (; ; ++NumDivergentGroups)
    {
        const uint SgprData = ReadFirstLane(VgprData);
        if (SgprData == VgprData)
        {
            break;
        }
        if (AnyExecSet() == false)
        {
            break;
        }
    }
#endif
    OutTexture[TexelId] = NumDivergentGroups;
}

/*
; -------- Disassembly --------------------
shader main
  asic(SI)
  type(CS)

  v_mad_u32_u24  v0, s12, 8, v0                             // 00000000: D2860000 0401100C
  v_mad_u32_u24  v1, s13, 8, v1                             // 00000008: D2860001 0405100D
  s_load_dwordx8  s[12:19], s[2:3], 0x00                    // 00000010: C0C60300
  s_waitcnt     lgkmcnt(0)                                  // 00000014: BF8C007F
  image_load    v2, v[0:3], s[12:19] unorm                  // 00000018: F0001100 00030200
  s_mov_b64     s[0:1], exec                                // 00000020: BE80047E
  s_mov_b64     s[2:3], exec                                // 00000024: BE82047E
  v_mov_b32     v3, 0                                       // 00000028: 7E060280
  s_nop         0x0000                                      // 0000002C: BF800000
  s_nop         0x0000                                      // 00000030: BF800000
  s_nop         0x0000                                      // 00000034: BF800000
  s_nop         0x0000                                      // 00000038: BF800000
  s_nop         0x0000                                      // 0000003C: BF800000
label_0010:
  s_waitcnt     vmcnt(0)                                    // 00000040: BF8C0F70
  v_readfirstlane_b32  s12, v2                              // 00000044: 7E180502
  v_cmp_eq_i32  vcc, s12, v2                                // 00000048: 7D04040C
  s_and_saveexec_b64  s[12:13], vcc                         // 0000004C: BE8C246A
  s_andn2_b64   s[2:3], s[2:3], exec                        // 00000050: 8A827E02
  s_cbranch_scc0  label_001C                                // 00000054: BF840006
  s_and_b64     exec, s[12:13], s[2:3]                      // 00000058: 87FE020C
  s_or_b32      s12, exec_lo, exec_hi                       // 0000005C: 880C7F7E
  s_cmp_eq_i32  s12, 0                                      // 00000060: BF00800C
  s_cbranch_scc1  label_001C                                // 00000064: BF850002
  v_add_i32     v3, vcc, 1, v3                              // 00000068: 4A060681
  s_branch      label_0010                                  // 0000006C: BF82FFF4
label_001C:
  s_mov_b64     exec, s[0:1]                                // 00000070: BEFE0400
  v_mov_b32     v2, v3                                      // 00000074: 7E040303
  v_mov_b32     v4, v3                                      // 00000078: 7E080303
  v_mov_b32     v5, v3                                      // 0000007C: 7E0A0303
  image_store   v[2:5], v[0:3], s[4:11] dmask:0xf unorm glc // 00000080: F0203F00 00010200
  s_endpgm                                                  // 00000088: BF810000
end
*/