xref: /llvm-project/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.readlane.ptr.ll (revision 5feb32ba929f9e517c530217cabb09d1d734a763)

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji -verify-machineinstrs < %s | FileCheck --check-prefix=CHECK-SDAG -enable-var-scope %s

define void @test_readlane_p0(ptr addrspace(1) %out, ptr %src, i32 %src1) {
; CHECK-SDAG-LABEL: test_readlane_p0:
; CHECK-SDAG:       ; %bb.0:
; CHECK-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-SDAG-NEXT:    v_readfirstlane_b32 s4, v4
; CHECK-SDAG-NEXT:    s_nop 3
; CHECK-SDAG-NEXT:    v_readlane_b32 s5, v3, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s4, v2, s4
; CHECK-SDAG-NEXT:    ;;#ASMSTART
; CHECK-SDAG-NEXT:    ; use s[4:5]
; CHECK-SDAG-NEXT:    ;;#ASMEND
; CHECK-SDAG-NEXT:    s_setpc_b64 s[30:31]
  %x = call ptr @llvm.amdgcn.readlane.p0(ptr %src, i32 %src1)
  call void asm sideeffect "; use $0", "s"(ptr %x)
  ret void
}

define void @test_readlane_v3p0(ptr addrspace(1) %out, <3 x ptr> %src, i32 %src1) {
; CHECK-SDAG-LABEL: test_readlane_v3p0:
; CHECK-SDAG:       ; %bb.0:
; CHECK-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-SDAG-NEXT:    v_readfirstlane_b32 s4, v8
; CHECK-SDAG-NEXT:    s_nop 3
; CHECK-SDAG-NEXT:    v_readlane_b32 s9, v7, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s8, v6, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s7, v5, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s6, v4, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s5, v3, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s4, v2, s4
; CHECK-SDAG-NEXT:    ;;#ASMSTART
; CHECK-SDAG-NEXT:    ; use s[4:9]
; CHECK-SDAG-NEXT:    ;;#ASMEND
; CHECK-SDAG-NEXT:    s_setpc_b64 s[30:31]
  %x = call <3 x ptr> @llvm.amdgcn.readlane.v3p0(<3 x ptr> %src, i32 %src1)
  call void asm sideeffect "; use $0", "s"(<3 x ptr> %x)
  ret void
}

define void @test_readlane_p3(ptr addrspace(1) %out, ptr addrspace(3) %src, i32 %src1) {
; CHECK-SDAG-LABEL: test_readlane_p3:
; CHECK-SDAG:       ; %bb.0:
; CHECK-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-SDAG-NEXT:    v_readfirstlane_b32 s4, v3
; CHECK-SDAG-NEXT:    s_nop 3
; CHECK-SDAG-NEXT:    v_readlane_b32 s4, v2, s4
; CHECK-SDAG-NEXT:    ;;#ASMSTART
; CHECK-SDAG-NEXT:    ; use s4
; CHECK-SDAG-NEXT:    ;;#ASMEND
; CHECK-SDAG-NEXT:    s_setpc_b64 s[30:31]
  %x = call ptr addrspace(3) @llvm.amdgcn.readlane.p3(ptr addrspace(3) %src, i32 %src1)
  call void asm sideeffect "; use $0", "s"(ptr addrspace(3) %x)
  ret void
}

define void @test_readlane_v3p3(ptr addrspace(1) %out, <3 x ptr addrspace(3)> %src, i32 %src1) {
; CHECK-SDAG-LABEL: test_readlane_v3p3:
; CHECK-SDAG:       ; %bb.0:
; CHECK-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-SDAG-NEXT:    v_readfirstlane_b32 s4, v5
; CHECK-SDAG-NEXT:    s_nop 3
; CHECK-SDAG-NEXT:    v_readlane_b32 s6, v4, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s5, v3, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s4, v2, s4
; CHECK-SDAG-NEXT:    ;;#ASMSTART
; CHECK-SDAG-NEXT:    ; use s[4:6]
; CHECK-SDAG-NEXT:    ;;#ASMEND
; CHECK-SDAG-NEXT:    s_setpc_b64 s[30:31]
  %x = call <3 x ptr addrspace(3)> @llvm.amdgcn.readlane.v3p3(<3 x ptr addrspace(3)> %src, i32 %src1)
  call void asm sideeffect "; use $0", "s"(<3 x ptr addrspace(3)> %x)
  ret void
}

define void @test_readlane_p5(ptr addrspace(1) %out, ptr addrspace(5) %src, i32 %src1) {
; CHECK-SDAG-LABEL: test_readlane_p5:
; CHECK-SDAG:       ; %bb.0:
; CHECK-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-SDAG-NEXT:    v_readfirstlane_b32 s4, v3
; CHECK-SDAG-NEXT:    s_nop 3
; CHECK-SDAG-NEXT:    v_readlane_b32 s4, v2, s4
; CHECK-SDAG-NEXT:    ;;#ASMSTART
; CHECK-SDAG-NEXT:    ; use s4
; CHECK-SDAG-NEXT:    ;;#ASMEND
; CHECK-SDAG-NEXT:    s_setpc_b64 s[30:31]
  %x = call ptr addrspace(5) @llvm.amdgcn.readlane.p5(ptr addrspace(5) %src, i32 %src1)
  call void asm sideeffect "; use $0", "s"(ptr addrspace(5) %x)
  ret void
}

define void @test_readlane_v3p5(ptr addrspace(1) %out, <3 x ptr addrspace(5)> %src, i32 %src1) {
; CHECK-SDAG-LABEL: test_readlane_v3p5:
; CHECK-SDAG:       ; %bb.0:
; CHECK-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-SDAG-NEXT:    v_readfirstlane_b32 s4, v5
; CHECK-SDAG-NEXT:    s_nop 3
; CHECK-SDAG-NEXT:    v_readlane_b32 s6, v4, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s5, v3, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s4, v2, s4
; CHECK-SDAG-NEXT:    ;;#ASMSTART
; CHECK-SDAG-NEXT:    ; use s[4:6]
; CHECK-SDAG-NEXT:    ;;#ASMEND
; CHECK-SDAG-NEXT:    s_setpc_b64 s[30:31]
  %x = call <3 x ptr addrspace(5)> @llvm.amdgcn.readlane.v3p5(<3 x ptr addrspace(5)> %src, i32 %src1)
  call void asm sideeffect "; use $0", "s"(<3 x ptr addrspace(5)> %x)
  ret void
}

define void @test_readlane_p6(ptr addrspace(1) %out, ptr addrspace(6) %src, i32 %src1) {
; CHECK-SDAG-LABEL: test_readlane_p6:
; CHECK-SDAG:       ; %bb.0:
; CHECK-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-SDAG-NEXT:    v_readfirstlane_b32 s4, v3
; CHECK-SDAG-NEXT:    s_nop 3
; CHECK-SDAG-NEXT:    v_readlane_b32 s4, v2, s4
; CHECK-SDAG-NEXT:    ;;#ASMSTART
; CHECK-SDAG-NEXT:    ; use s4
; CHECK-SDAG-NEXT:    ;;#ASMEND
; CHECK-SDAG-NEXT:    s_setpc_b64 s[30:31]
  %x = call ptr addrspace(6) @llvm.amdgcn.readlane.p6(ptr addrspace(6) %src, i32 %src1)
  call void asm sideeffect "; use $0", "s"(ptr addrspace(6) %x)
  ret void
}

define void @test_readlane_v3p6(ptr addrspace(1) %out, <3 x ptr addrspace(6)> %src, i32 %src1) {
; CHECK-SDAG-LABEL: test_readlane_v3p6:
; CHECK-SDAG:       ; %bb.0:
; CHECK-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-SDAG-NEXT:    v_readfirstlane_b32 s4, v5
; CHECK-SDAG-NEXT:    s_nop 3
; CHECK-SDAG-NEXT:    v_readlane_b32 s6, v4, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s5, v3, s4
; CHECK-SDAG-NEXT:    v_readlane_b32 s4, v2, s4
; CHECK-SDAG-NEXT:    ;;#ASMSTART
; CHECK-SDAG-NEXT:    ; use s[4:6]
; CHECK-SDAG-NEXT:    ;;#ASMEND
; CHECK-SDAG-NEXT:    s_setpc_b64 s[30:31]
  %x = call <3 x ptr addrspace(6)> @llvm.amdgcn.readlane.v3p6(<3 x ptr addrspace(6)> %src, i32 %src1)
  call void asm sideeffect "; use $0", "s"(<3 x ptr addrspace(6)> %x)
  ret void
}