AMDGPU/GlobalISel: Select G_CTLZ_ZERO_UNDEF

Directly select this rather than going through the intermediate
instruction, which may provide some combine value in the future.

Author: arsenm

llvm/lib/Target/AMDGPU/AMDGPUGISel.td

@@ -140,7 +140,7 @@ def : GINodeEquiv<G_ATOMICRMW_UMIN, atomic_load_umin_glue>;
 def : GINodeEquiv<G_ATOMICRMW_UMAX, atomic_load_umax_glue>;
 def : GINodeEquiv<G_ATOMICRMW_FADD, atomic_load_fadd_glue>;
 
-def : GINodeEquiv<G_AMDGPU_FFBH_U32, AMDGPUffbh_u32>;
+def : GINodeEquiv<G_AMDGPU_FFBH_U32, AMDGPUffbh_u32_impl>;
 def : GINodeEquiv<G_AMDGPU_FMIN_LEGACY, AMDGPUfmin_legacy>;
 def : GINodeEquiv<G_AMDGPU_FMAX_LEGACY, AMDGPUfmax_legacy>;
 

llvm/lib/Target/AMDGPU/AMDGPUInstrInfo.td

@@ -283,8 +283,8 @@ def AMDGPUbfe_i32 : SDNode<"AMDGPUISD::BFE_I32", AMDGPUDTIntTernaryOp>;
 def AMDGPUbfi : SDNode<"AMDGPUISD::BFI", AMDGPUDTIntTernaryOp>;
 def AMDGPUbfm : SDNode<"AMDGPUISD::BFM", SDTIntBinOp>;
 
-def AMDGPUffbh_u32 : SDNode<"AMDGPUISD::FFBH_U32", SDTIntUnaryOp>;
-def AMDGPUffbh_i32_impl : SDNode<"AMDGPUISD::FFBH_I32", SDTIntUnaryOp>;
+def AMDGPUffbh_u32_impl : SDNode<"AMDGPUISD::FFBH_U32", SDTIntBitCountUnaryOp>;
+def AMDGPUffbh_i32_impl : SDNode<"AMDGPUISD::FFBH_I32", SDTIntBitCountUnaryOp>;
 
 def AMDGPUffbl_b32 : SDNode<"AMDGPUISD::FFBL_B32", SDTIntUnaryOp>;
 
@@ -421,6 +421,10 @@ def AMDGPUffbh_i32 : PatFrags<(ops node:$src),
   [(int_amdgcn_sffbh node:$src),
    (AMDGPUffbh_i32_impl node:$src)]>;
 
+def AMDGPUffbh_u32 : PatFrags<(ops node:$src),
+  [(ctlz_zero_undef node:$src),
+   (AMDGPUffbh_u32_impl node:$src)]>;
+
 def AMDGPUpkrtz_f16_f32 : PatFrags<(ops node:$src0, node:$src1),
   [(int_amdgcn_cvt_pkrtz node:$src0, node:$src1),
   (AMDGPUpkrtz_f16_f32_impl node:$src0, node:$src1)]>;

llvm/lib/Target/AMDGPU/SOPInstructions.td

@@ -220,7 +220,9 @@ def S_FLBIT_I32_B32 : SOP1_32 <"s_flbit_i32_b32",
   [(set i32:$sdst, (AMDGPUffbh_u32 i32:$src0))]
 >;
 
-def S_FLBIT_I32_B64 : SOP1_32_64 <"s_flbit_i32_b64">;
+def S_FLBIT_I32_B64 : SOP1_32_64 <"s_flbit_i32_b64",
+  [(set i32:$sdst, (AMDGPUffbh_u32 i64:$src0))]
+>;
 def S_FLBIT_I32 : SOP1_32 <"s_flbit_i32",
   [(set i32:$sdst, (AMDGPUffbh_i32 i32:$src0))]
 >;

llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-ctlz-zero-undef.mir

@@ -0,0 +1,82 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s
+
+---
+name: ctlz_zero_undef_s32_ss
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+
+    ; CHECK-LABEL: name: ctlz_zero_undef_s32_ss
+    ; CHECK: liveins: $sgpr0
+    ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
+    ; CHECK: [[S_FLBIT_I32_B32_:%[0-9]+]]:sreg_32 = S_FLBIT_I32_B32 [[COPY]]
+    ; CHECK: S_ENDPGM 0, implicit [[S_FLBIT_I32_B32_]]
+    %0:sgpr(s32) = COPY $sgpr0
+    %1:sgpr(s32) = G_CTLZ_ZERO_UNDEF %0
+    S_ENDPGM 0, implicit %1
+...
+
+---
+name: ctlz_zero_undef_s32_vs
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+
+    ; CHECK-LABEL: name: ctlz_zero_undef_s32_vs
+    ; CHECK: liveins: $sgpr0
+    ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
+    ; CHECK: [[V_FFBH_U32_e64_:%[0-9]+]]:vgpr_32 = V_FFBH_U32_e64 [[COPY]], implicit $exec
+    ; CHECK: S_ENDPGM 0, implicit [[V_FFBH_U32_e64_]]
+    %0:sgpr(s32) = COPY $sgpr0
+    %1:vgpr(s32) = G_CTLZ_ZERO_UNDEF %0
+    S_ENDPGM 0, implicit %1
+...
+
+---
+name: ctlz_zero_undef_s32_vv
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; CHECK-LABEL: name: ctlz_zero_undef_s32_vv
+    ; CHECK: liveins: $vgpr0
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; CHECK: [[V_FFBH_U32_e64_:%[0-9]+]]:vgpr_32 = V_FFBH_U32_e64 [[COPY]], implicit $exec
+    ; CHECK: S_ENDPGM 0, implicit [[V_FFBH_U32_e64_]]
+    %0:vgpr(s32) = COPY $vgpr0
+    %1:vgpr(s32) = G_CTLZ_ZERO_UNDEF %0
+    S_ENDPGM 0, implicit %1
+...
+
+---
+name: ctlz_zero_undef_s64_ss
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    ; CHECK-LABEL: name: ctlz_zero_undef_s64_ss
+    ; CHECK: liveins: $sgpr0_sgpr1
+    ; CHECK: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
+    ; CHECK: [[S_FLBIT_I32_B64_:%[0-9]+]]:sreg_32 = S_FLBIT_I32_B64 [[COPY]]
+    ; CHECK: S_ENDPGM 0, implicit [[S_FLBIT_I32_B64_]]
+    %0:sgpr(s64) = COPY $sgpr0_sgpr1
+    %1:sgpr(s32) = G_CTLZ_ZERO_UNDEF %0
+    S_ENDPGM 0, implicit %1
+...