llvm/test/CodeGen/AMDGPU/code-size-estimate.ll

   1 ; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -show-mc-encoding < %s | FileCheck -check-prefixes=GFX9,NOT-GFX12 %s
   2 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1030 -show-mc-encoding < %s | FileCheck -check-prefixes=GFX10,NOT-GFX12 %s
   3 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -show-mc-encoding < %s | FileCheck -check-prefixes=GFX11,GFX1100,NOT-GFX12 %s
   4 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1150 -show-mc-encoding < %s | FileCheck -check-prefixes=GFX11,GFX1150,NOT-GFX12 %s
   5 ; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -show-mc-encoding < %s | FileCheck -check-prefixes=GFX1200 %s
   6
   7 declare float @llvm.fabs.f32(float)
   8 declare float @llvm.fma.f32(float, float, float)
   9
  10 define float @v_mul_f32_vop2(float %x, float %y) {
  11 ; GFX9-LABEL: v_mul_f32_vop2:
  12 ; GFX9:       ; %bb.0:
  13 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
  14 ; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v1 ; encoding: [0x00,0x03,0x00,0x0a]
  15 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
  16 ;
  17 ; GFX10-LABEL: v_mul_f32_vop2:
  18 ; GFX10:       ; %bb.0:
  19 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
  20 ; GFX10-NEXT:    v_mul_f32_e32 v0, v0, v1 ; encoding: [0x00,0x03,0x00,0x10]
  21 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
  22 ;
  23 ; GFX11-LABEL: v_mul_f32_vop2:
  24 ; GFX11:       ; %bb.0:
  25 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
  26 ; GFX11-NEXT:    v_mul_f32_e32 v0, v0, v1 ; encoding: [0x00,0x03,0x00,0x10]
  27 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
  28 ;
  29 ; GFX1200-LABEL: v_mul_f32_vop2:
  30 ; GFX1200:       ; %bb.0:
  31 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
  32 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
  33 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
  34 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
  35 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
  36 ; GFX1200-NEXT:    v_mul_f32_e32 v0, v0, v1 ; encoding: [0x00,0x03,0x00,0x10]
  37 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
  38   %mul = fmul float %x, %y
  39   ret float %mul
  40 }
  41 ; NOT-GFX12: codeLenInByte = 12
  42 ; GFX1200: codeLenInByte = 28
  43
  44 define float @v_mul_f32_vop2_inline_imm(float %x) {
  45 ; GFX9-LABEL: v_mul_f32_vop2_inline_imm:
  46 ; GFX9:       ; %bb.0:
  47 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
  48 ; GFX9-NEXT:    v_mul_f32_e32 v0, 4.0, v0 ; encoding: [0xf6,0x00,0x00,0x0a]
  49 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
  50 ;
  51 ; GFX10-LABEL: v_mul_f32_vop2_inline_imm:
  52 ; GFX10:       ; %bb.0:
  53 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
  54 ; GFX10-NEXT:    v_mul_f32_e32 v0, 4.0, v0 ; encoding: [0xf6,0x00,0x00,0x10]
  55 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
  56 ;
  57 ; GFX11-LABEL: v_mul_f32_vop2_inline_imm:
  58 ; GFX11:       ; %bb.0:
  59 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
  60 ; GFX11-NEXT:    v_mul_f32_e32 v0, 4.0, v0 ; encoding: [0xf6,0x00,0x00,0x10]
  61 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
  62 ;
  63 ; GFX1200-LABEL: v_mul_f32_vop2_inline_imm:
  64 ; GFX1200:       ; %bb.0:
  65 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
  66 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
  67 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
  68 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
  69 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
  70 ; GFX1200-NEXT:    v_mul_f32_e32 v0, 4.0, v0 ; encoding: [0xf6,0x00,0x00,0x10]
  71 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
  72   %mul = fmul float %x, 4.0
  73   ret float %mul
  74 }
  75 ; NOT-GFX12: codeLenInByte = 12
  76 ; GFX1200: codeLenInByte = 28
  77
  78 define float @v_mul_f32_vop2_literal(float %x) {
  79 ; GFX9-LABEL: v_mul_f32_vop2_literal:
  80 ; GFX9:       ; %bb.0:
  81 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
  82 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x42f60000, v0 ; encoding: [0xff,0x00,0x00,0x0a,0x00,0x00,0xf6,0x42]
  83 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
  84 ;
  85 ; GFX10-LABEL: v_mul_f32_vop2_literal:
  86 ; GFX10:       ; %bb.0:
  87 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
  88 ; GFX10-NEXT:    v_mul_f32_e32 v0, 0x42f60000, v0 ; encoding: [0xff,0x00,0x00,0x10,0x00,0x00,0xf6,0x42]
  89 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
  90 ;
  91 ; GFX11-LABEL: v_mul_f32_vop2_literal:
  92 ; GFX11:       ; %bb.0:
  93 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
  94 ; GFX11-NEXT:    v_mul_f32_e32 v0, 0x42f60000, v0 ; encoding: [0xff,0x00,0x00,0x10,0x00,0x00,0xf6,0x42]
  95 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
  96 ;
  97 ; GFX1200-LABEL: v_mul_f32_vop2_literal:
  98 ; GFX1200:       ; %bb.0:
  99 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 100 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 101 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 102 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 103 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 104 ; GFX1200-NEXT:    v_mul_f32_e32 v0, 0x42f60000, v0 ; encoding: [0xff,0x00,0x00,0x10,0x00,0x00,0xf6,0x42]
 105 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 106   %mul = fmul float %x, 123.0
 107   ret float %mul
 108 }
 109 ; NOT-GFX12: codeLenInByte = 16
 110 ; GFX1200: codeLenInByte = 32
 111
 112 define float @v_mul_f32_vop3_src_mods(float %x, float %y) {
 113 ; GFX9-LABEL: v_mul_f32_vop3_src_mods:
 114 ; GFX9:       ; %bb.0:
 115 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 116 ; GFX9-NEXT:    v_mul_f32_e64 v0, |v0|, v1 ; encoding: [0x00,0x01,0x05,0xd1,0x00,0x03,0x02,0x00]
 117 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 118 ;
 119 ; GFX10-LABEL: v_mul_f32_vop3_src_mods:
 120 ; GFX10:       ; %bb.0:
 121 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 122 ; GFX10-NEXT:    v_mul_f32_e64 v0, |v0|, v1 ; encoding: [0x00,0x01,0x08,0xd5,0x00,0x03,0x02,0x00]
 123 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 124 ;
 125 ; GFX11-LABEL: v_mul_f32_vop3_src_mods:
 126 ; GFX11:       ; %bb.0:
 127 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 128 ; GFX11-NEXT:    v_mul_f32_e64 v0, |v0|, v1 ; encoding: [0x00,0x01,0x08,0xd5,0x00,0x03,0x02,0x00]
 129 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 130 ;
 131 ; GFX1200-LABEL: v_mul_f32_vop3_src_mods:
 132 ; GFX1200:       ; %bb.0:
 133 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 134 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 135 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 136 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 137 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 138 ; GFX1200-NEXT:    v_mul_f32_e64 v0, |v0|, v1 ; encoding: [0x00,0x01,0x08,0xd5,0x00,0x03,0x02,0x00]
 139 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 140   %fabs.x = call float @llvm.fabs.f32(float %x)
 141   %mul = fmul float %fabs.x, %y
 142   ret float %mul
 143 }
 144 ; NOT-GFX12: codeLenInByte = 16
 145 ; GFX1200: codeLenInByte = 32
 146
 147 define float @v_mul_f32_vop3_src_mods_inline_imm(float %x, float %y) {
 148 ; GFX9-LABEL: v_mul_f32_vop3_src_mods_inline_imm:
 149 ; GFX9:       ; %bb.0:
 150 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 151 ; GFX9-NEXT:    v_mul_f32_e64 v0, |v0|, 4.0 ; encoding: [0x00,0x01,0x05,0xd1,0x00,0xed,0x01,0x00]
 152 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 153 ;
 154 ; GFX10-LABEL: v_mul_f32_vop3_src_mods_inline_imm:
 155 ; GFX10:       ; %bb.0:
 156 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 157 ; GFX10-NEXT:    v_mul_f32_e64 v0, |v0|, 4.0 ; encoding: [0x00,0x01,0x08,0xd5,0x00,0xed,0x01,0x00]
 158 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 159 ;
 160 ; GFX11-LABEL: v_mul_f32_vop3_src_mods_inline_imm:
 161 ; GFX11:       ; %bb.0:
 162 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 163 ; GFX11-NEXT:    v_mul_f32_e64 v0, |v0|, 4.0 ; encoding: [0x00,0x01,0x08,0xd5,0x00,0xed,0x01,0x00]
 164 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 165 ;
 166 ; GFX1200-LABEL: v_mul_f32_vop3_src_mods_inline_imm:
 167 ; GFX1200:       ; %bb.0:
 168 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 169 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 170 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 171 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 172 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 173 ; GFX1200-NEXT:    v_mul_f32_e64 v0, |v0|, 4.0 ; encoding: [0x00,0x01,0x08,0xd5,0x00,0xed,0x01,0x00]
 174 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 175   %fabs.x = call float @llvm.fabs.f32(float %x)
 176   %mul = fmul float %fabs.x, 4.0
 177   ret float %mul
 178 }
 179
 180 ; NOT-GFX12: codeLenInByte = 16
 181 ; GFX1200: codeLenInByte = 32
 182
 183 define float @v_mul_f32_vop3_src_mods_literal(float %x, float %y) {
 184 ; GFX9-LABEL: v_mul_f32_vop3_src_mods_literal:
 185 ; GFX9:       ; %bb.0:
 186 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 187 ; GFX9-NEXT:    s_mov_b32 s4, 0x42f60000 ; encoding: [0xff,0x00,0x84,0xbe,0x00,0x00,0xf6,0x42]
 188 ; GFX9-NEXT:    v_mul_f32_e64 v0, |v0|, s4 ; encoding: [0x00,0x01,0x05,0xd1,0x00,0x09,0x00,0x00]
 189 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 190 ;
 191 ; GFX10-LABEL: v_mul_f32_vop3_src_mods_literal:
 192 ; GFX10:       ; %bb.0:
 193 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 194 ; GFX10-NEXT:    v_mul_f32_e64 v0, 0x42f60000, |v0| ; encoding: [0x00,0x02,0x08,0xd5,0xff,0x00,0x02,0x00,0x00,0x00,0xf6,0x42]
 195 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 196 ;
 197 ; GFX11-LABEL: v_mul_f32_vop3_src_mods_literal:
 198 ; GFX11:       ; %bb.0:
 199 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 200 ; GFX11-NEXT:    v_mul_f32_e64 v0, 0x42f60000, |v0| ; encoding: [0x00,0x02,0x08,0xd5,0xff,0x00,0x02,0x00,0x00,0x00,0xf6,0x42]
 201 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 202 ;
 203 ; GFX1200-LABEL: v_mul_f32_vop3_src_mods_literal:
 204 ; GFX1200:       ; %bb.0:
 205 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 206 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 207 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 208 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 209 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 210 ; GFX1200-NEXT:    v_mul_f32_e64 v0, 0x42f60000, |v0| ; encoding: [0x00,0x02,0x08,0xd5,0xff,0x00,0x02,0x00,0x00,0x00,0xf6,0x42]
 211 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 212   %fabs.x = call float @llvm.fabs.f32(float %x)
 213   %mul = fmul float %fabs.x, 123.0
 214   ret float %mul
 215 }
 216
 217 ; GFX9: codeLenInByte = 24
 218 ; GFX10: codeLenInByte = 20
 219 ; GFX11: codeLenInByte = 20
 220 ; GFX1200: codeLenInByte = 36
 221
 222 define float @v_mul_f32_vop2_frame_index(float %x) {
 223 ; GFX9-LABEL: v_mul_f32_vop2_frame_index:
 224 ; GFX9:       ; %bb.0:
 225 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 226 ; GFX9-NEXT:    v_lshrrev_b32_e64 v1, 6, s32 ; encoding: [0x01,0x00,0x10,0xd1,0x86,0x40,0x00,0x00]
 227 ; GFX9-NEXT:    v_mul_f32_e32 v0, v1, v0 ; encoding: [0x01,0x01,0x00,0x0a]
 228 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 229 ;
 230 ; GFX10-LABEL: v_mul_f32_vop2_frame_index:
 231 ; GFX10:       ; %bb.0:
 232 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 233 ; GFX10-NEXT:    v_lshrrev_b32_e64 v1, 5, s32 ; encoding: [0x01,0x00,0x16,0xd5,0x85,0x40,0x00,0x00]
 234 ; GFX10-NEXT:    v_mul_f32_e32 v0, v1, v0 ; encoding: [0x01,0x01,0x00,0x10]
 235 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 236 ;
 237 ; GFX11-LABEL: v_mul_f32_vop2_frame_index:
 238 ; GFX11:       ; %bb.0:
 239 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 240 ; GFX11-NEXT:    v_mul_f32_e32 v0, s32, v0 ; encoding: [0x20,0x00,0x00,0x10]
 241 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 242 ;
 243 ; GFX1200-LABEL: v_mul_f32_vop2_frame_index:
 244 ; GFX1200:       ; %bb.0:
 245 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 246 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 247 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 248 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 249 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 250 ; GFX1200-NEXT:    v_mul_f32_e32 v0, s32, v0 ; encoding: [0x20,0x00,0x00,0x10]
 251 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 252   %alloca = alloca i32, addrspace(5)
 253   %ptrtoint = ptrtoint ptr addrspace(5) %alloca to i32
 254   %cast = bitcast i32 %ptrtoint to float
 255   %mul = fmul float %x, %cast
 256   ret float %mul
 257 }
 258
 259 ; GFX9: codeLenInByte = 20
 260 ; GFX10: codeLenInByte = 20
 261 ; GFX11: codeLenInByte = 12
 262 ; GFX1200: codeLenInByte = 28
 263
 264 define float @v_fma_f32(float %x, float %y, float %z) {
 265 ; GFX9-LABEL: v_fma_f32:
 266 ; GFX9:       ; %bb.0:
 267 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 268 ; GFX9-NEXT:    v_fma_f32 v0, v0, v1, v2 ; encoding: [0x00,0x00,0xcb,0xd1,0x00,0x03,0x0a,0x04]
 269 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 270 ;
 271 ; GFX10-LABEL: v_fma_f32:
 272 ; GFX10:       ; %bb.0:
 273 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 274 ; GFX10-NEXT:    v_fma_f32 v0, v0, v1, v2 ; encoding: [0x00,0x00,0x4b,0xd5,0x00,0x03,0x0a,0x04]
 275 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 276 ;
 277 ; GFX11-LABEL: v_fma_f32:
 278 ; GFX11:       ; %bb.0:
 279 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 280 ; GFX11-NEXT:    v_fma_f32 v0, v0, v1, v2 ; encoding: [0x00,0x00,0x13,0xd6,0x00,0x03,0x0a,0x04]
 281 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 282 ;
 283 ; GFX1200-LABEL: v_fma_f32:
 284 ; GFX1200:       ; %bb.0:
 285 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 286 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 287 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 288 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 289 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 290 ; GFX1200-NEXT:    v_fma_f32 v0, v0, v1, v2 ; encoding: [0x00,0x00,0x13,0xd6,0x00,0x03,0x0a,0x04]
 291 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 292   %fma = call float @llvm.fma.f32(float %x, float %y, float %z)
 293   ret float %fma
 294 }
 295
 296 ; NOT-GFX12: codeLenInByte = 16
 297 ; GFX1200: codeLenInByte = 32
 298
 299 define float @v_fma_f32_src_mods(float %x, float %y, float %z) {
 300 ; GFX9-LABEL: v_fma_f32_src_mods:
 301 ; GFX9:       ; %bb.0:
 302 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 303 ; GFX9-NEXT:    v_fma_f32 v0, |v0|, v1, v2 ; encoding: [0x00,0x01,0xcb,0xd1,0x00,0x03,0x0a,0x04]
 304 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 305 ;
 306 ; GFX10-LABEL: v_fma_f32_src_mods:
 307 ; GFX10:       ; %bb.0:
 308 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 309 ; GFX10-NEXT:    v_fma_f32 v0, |v0|, v1, v2 ; encoding: [0x00,0x01,0x4b,0xd5,0x00,0x03,0x0a,0x04]
 310 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 311 ;
 312 ; GFX11-LABEL: v_fma_f32_src_mods:
 313 ; GFX11:       ; %bb.0:
 314 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 315 ; GFX11-NEXT:    v_fma_f32 v0, |v0|, v1, v2 ; encoding: [0x00,0x01,0x13,0xd6,0x00,0x03,0x0a,0x04]
 316 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 317 ;
 318 ; GFX1200-LABEL: v_fma_f32_src_mods:
 319 ; GFX1200:       ; %bb.0:
 320 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 321 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 322 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 323 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 324 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 325 ; GFX1200-NEXT:    v_fma_f32 v0, |v0|, v1, v2 ; encoding: [0x00,0x01,0x13,0xd6,0x00,0x03,0x0a,0x04]
 326 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 327   %fabs.x = call float @llvm.fabs.f32(float %x)
 328   %fma = call float @llvm.fma.f32(float %fabs.x, float %y, float %z)
 329   ret float %fma
 330 }
 331
 332 ; NOT-GFX12: codeLenInByte = 16
 333 ; GFX1200: codeLenInByte = 32
 334
 335 define float @v_fmac_f32(float %x, float %y) {
 336 ; GFX9-LABEL: v_fmac_f32:
 337 ; GFX9:       ; %bb.0:
 338 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 339 ; GFX9-NEXT:    v_fma_f32 v0, v0, v1, v0 ; encoding: [0x00,0x00,0xcb,0xd1,0x00,0x03,0x02,0x04]
 340 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 341 ;
 342 ; GFX10-LABEL: v_fmac_f32:
 343 ; GFX10:       ; %bb.0:
 344 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 345 ; GFX10-NEXT:    v_fmac_f32_e32 v0, v0, v1 ; encoding: [0x00,0x03,0x00,0x56]
 346 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 347 ;
 348 ; GFX11-LABEL: v_fmac_f32:
 349 ; GFX11:       ; %bb.0:
 350 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 351 ; GFX11-NEXT:    v_fmac_f32_e32 v0, v0, v1 ; encoding: [0x00,0x03,0x00,0x56]
 352 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 353 ;
 354 ; GFX1200-LABEL: v_fmac_f32:
 355 ; GFX1200:       ; %bb.0:
 356 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 357 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 358 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 359 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 360 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 361 ; GFX1200-NEXT:    v_fmac_f32_e32 v0, v0, v1 ; encoding: [0x00,0x03,0x00,0x56]
 362 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 363   %fma = call float @llvm.fma.f32(float %x, float %y, float %x)
 364   ret float %fma
 365 }
 366
 367 ; GFX9: codeLenInByte = 16
 368 ; GFX10: codeLenInByte = 12
 369 ; GFX11: codeLenInByte = 12
 370 ; GFX1200: codeLenInByte = 28
 371
 372 define float @v_fmaak_f32(float %x, float %y) {
 373 ; GFX9-LABEL: v_fmaak_f32:
 374 ; GFX9:       ; %bb.0:
 375 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 376 ; GFX9-NEXT:    s_mov_b32 s4, 0x43800000 ; encoding: [0xff,0x00,0x84,0xbe,0x00,0x00,0x80,0x43]
 377 ; GFX9-NEXT:    v_fma_f32 v0, v0, v1, s4 ; encoding: [0x00,0x00,0xcb,0xd1,0x00,0x03,0x12,0x00]
 378 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 379 ;
 380 ; GFX10-LABEL: v_fmaak_f32:
 381 ; GFX10:       ; %bb.0:
 382 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 383 ; GFX10-NEXT:    v_fmaak_f32 v0, v0, v1, 0x43800000 ; encoding: [0x00,0x03,0x00,0x5a,0x00,0x00,0x80,0x43]
 384 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 385 ;
 386 ; GFX11-LABEL: v_fmaak_f32:
 387 ; GFX11:       ; %bb.0:
 388 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 389 ; GFX11-NEXT:    v_fmaak_f32 v0, v0, v1, 0x43800000 ; encoding: [0x00,0x03,0x00,0x5a,0x00,0x00,0x80,0x43]
 390 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 391 ;
 392 ; GFX1200-LABEL: v_fmaak_f32:
 393 ; GFX1200:       ; %bb.0:
 394 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 395 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 396 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 397 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 398 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 399 ; GFX1200-NEXT:    v_fmaak_f32 v0, v0, v1, 0x43800000 ; encoding: [0x00,0x03,0x00,0x5a,0x00,0x00,0x80,0x43]
 400 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 401   %fma = call float @llvm.fma.f32(float %x, float %y, float 256.0)
 402   ret float %fma
 403 }
 404
 405 ; GFX9: codeLenInByte = 24
 406 ; GFX10: codeLenInByte = 16
 407 ; GFX11: codeLenInByte = 16
 408 ; GFX1200: codeLenInByte = 32
 409
 410 define float @v_fma_k_f32_src_mods(float %x, float %y) {
 411 ; GFX9-LABEL: v_fma_k_f32_src_mods:
 412 ; GFX9:       ; %bb.0:
 413 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 414 ; GFX9-NEXT:    s_mov_b32 s4, 0x43800000 ; encoding: [0xff,0x00,0x84,0xbe,0x00,0x00,0x80,0x43]
 415 ; GFX9-NEXT:    v_fma_f32 v0, |v0|, v1, s4 ; encoding: [0x00,0x01,0xcb,0xd1,0x00,0x03,0x12,0x00]
 416 ; GFX9-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x1d,0x80,0xbe]
 417 ;
 418 ; GFX10-LABEL: v_fma_k_f32_src_mods:
 419 ; GFX10:       ; %bb.0:
 420 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x8c,0xbf]
 421 ; GFX10-NEXT:    v_fma_f32 v0, |v0|, v1, 0x43800000 ; encoding: [0x00,0x01,0x4b,0xd5,0x00,0x03,0xfe,0x03,0x00,0x00,0x80,0x43]
 422 ; GFX10-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x20,0x80,0xbe]
 423 ;
 424 ; GFX11-LABEL: v_fma_k_f32_src_mods:
 425 ; GFX11:       ; %bb.0:
 426 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; encoding: [0x00,0x00,0x89,0xbf]
 427 ; GFX11-NEXT:    v_fma_f32 v0, |v0|, v1, 0x43800000 ; encoding: [0x00,0x01,0x13,0xd6,0x00,0x03,0xfe,0x03,0x00,0x00,0x80,0x43]
 428 ; GFX11-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 429 ;
 430 ; GFX1200-LABEL: v_fma_k_f32_src_mods:
 431 ; GFX1200:       ; %bb.0:
 432 ; GFX1200-NEXT:    s_wait_loadcnt_dscnt 0x0 ; encoding: [0x00,0x00,0xc8,0xbf]
 433 ; GFX1200-NEXT:    s_wait_expcnt 0x0 ; encoding: [0x00,0x00,0xc4,0xbf]
 434 ; GFX1200-NEXT:    s_wait_samplecnt 0x0 ; encoding: [0x00,0x00,0xc2,0xbf]
 435 ; GFX1200-NEXT:    s_wait_bvhcnt 0x0 ; encoding: [0x00,0x00,0xc3,0xbf]
 436 ; GFX1200-NEXT:    s_wait_kmcnt 0x0 ; encoding: [0x00,0x00,0xc7,0xbf]
 437 ; GFX1200-NEXT:    v_fma_f32 v0, |v0|, v1, 0x43800000 ; encoding: [0x00,0x01,0x13,0xd6,0x00,0x03,0xfe,0x03,0x00,0x00,0x80,0x43]
 438 ; GFX1200-NEXT:    s_setpc_b64 s[30:31] ; encoding: [0x1e,0x48,0x80,0xbe]
 439   %fabs.x = call float @llvm.fabs.f32(float %x)
 440   %fma = call float @llvm.fma.f32(float %fabs.x, float %y, float 256.0)
 441   ret float %fma
 442 }
 443
 444 ; GFX9: codeLenInByte = 24
 445 ; GFX10: codeLenInByte = 20
 446 ; GFX11: codeLenInByte = 20
 447 ; GFX1200: codeLenInByte = 36
 448
 449 define amdgpu_ps float @s_fmaak_f32(float inreg %x, float inreg %y) {
 450 ; GFX9-LABEL: s_fmaak_f32:
 451 ; GFX9:       ; %bb.0:
 452 ; GFX9-NEXT:    v_mov_b32_e32 v0, s1 ; encoding: [0x01,0x02,0x00,0x7e]
 453 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0x43800000 ; encoding: [0xff,0x02,0x02,0x7e,0x00,0x00,0x80,0x43]
 454 ; GFX9-NEXT:    v_fma_f32 v0, s0, v0, v1 ; encoding: [0x00,0x00,0xcb,0xd1,0x00,0x00,0x06,0x04]
 455 ; GFX9-NEXT:    ; return to shader part epilog
 456 ;
 457 ; GFX10-LABEL: s_fmaak_f32:
 458 ; GFX10:       ; %bb.0:
 459 ; GFX10-NEXT:    v_mov_b32_e32 v0, s1 ; encoding: [0x01,0x02,0x00,0x7e]
 460 ; GFX10-NEXT:    v_fmaak_f32 v0, s0, v0, 0x43800000 ; encoding: [0x00,0x00,0x00,0x5a,0x00,0x00,0x80,0x43]
 461 ; GFX10-NEXT:    ; return to shader part epilog
 462 ;
 463 ; GFX1100-LABEL: s_fmaak_f32:
 464 ; GFX1100:       ; %bb.0:
 465 ; GFX1100-NEXT:    v_mov_b32_e32 v0, s1 ; encoding: [0x01,0x02,0x00,0x7e]
 466 ; GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1) ; encoding: [0x01,0x00,0x87,0xbf]
 467 ; GFX1100-NEXT:    v_fmaak_f32 v0, s0, v0, 0x43800000 ; encoding: [0x00,0x00,0x00,0x5a,0x00,0x00,0x80,0x43]
 468 ; GFX1100-NEXT:    ; return to shader part epilog
 469 ;
 470 ; GFX1150-LABEL: s_fmaak_f32:
 471 ; GFX1150:       ; %bb.0:
 472 ; GFX1150-NEXT:    s_fmaak_f32 s0, s0, s1, 0x43800000 ; encoding: [0x00,0x01,0x80,0xa2,0x00,0x00,0x80,0x43]
 473 ; GFX1150-NEXT:    s_delay_alu instid0(SALU_CYCLE_3) ; encoding: [0x0b,0x00,0x87,0xbf]
 474 ; GFX1150-NEXT:    v_mov_b32_e32 v0, s0 ; encoding: [0x00,0x02,0x00,0x7e]
 475 ; GFX1150-NEXT:    ; return to shader part epilog
 476 ;
 477 ; GFX1200-LABEL: s_fmaak_f32:
 478 ; GFX1200:       ; %bb.0:
 479 ; GFX1200-NEXT:    s_fmaak_f32 s0, s0, s1, 0x43800000 ; encoding: [0x00,0x01,0x80,0xa2,0x00,0x00,0x80,0x43]
 480 ; GFX1200-NEXT:    s_delay_alu instid0(SALU_CYCLE_3) ; encoding: [0x0b,0x00,0x87,0xbf]
 481 ; GFX1200-NEXT:    v_mov_b32_e32 v0, s0 ; encoding: [0x00,0x02,0x00,0x7e]
 482 ; GFX1200-NEXT:    ; return to shader part epilog
 483   %fma = call float @llvm.fma.f32(float %x, float %y, float 256.0)
 484   ret float %fma
 485 }
 486
 487 ; GFX9: codeLenInByte = 20
 488 ; GFX10: codeLenInByte = 12
 489 ; GFX1100: codeLenInByte = 16
 490 ; GFX1150: codeLenInByte = 16
 491 ; GFX1200: codeLenInByte = 16