[llvm] r364758 - GlobalISel: Add DAG compat for G_FCANONICALIZE
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Mon Jul 1 06:22:00 PDT 2019
Author: arsenm
Date: Mon Jul 1 06:22:00 2019
New Revision: 364758
URL: http://llvm.org/viewvc/llvm-project?rev=364758&view=rev
Log:
GlobalISel: Add DAG compat for G_FCANONICALIZE
Added:
llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
Modified:
llvm/trunk/include/llvm/Target/GlobalISel/SelectionDAGCompat.td
Modified: llvm/trunk/include/llvm/Target/GlobalISel/SelectionDAGCompat.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Target/GlobalISel/SelectionDAGCompat.td?rev=364758&r1=364757&r2=364758&view=diff
==============================================================================
--- llvm/trunk/include/llvm/Target/GlobalISel/SelectionDAGCompat.td (original)
+++ llvm/trunk/include/llvm/Target/GlobalISel/SelectionDAGCompat.td Mon Jul 1 06:22:00 2019
@@ -76,6 +76,7 @@ def : GINodeEquiv<G_FREM, frem>;
def : GINodeEquiv<G_FPOW, fpow>;
def : GINodeEquiv<G_FEXP2, fexp2>;
def : GINodeEquiv<G_FLOG2, flog2>;
+def : GINodeEquiv<G_FCANONICALIZE, fcanonicalize>;
def : GINodeEquiv<G_INTRINSIC, intrinsic_wo_chain>;
// ISD::INTRINSIC_VOID can also be handled with G_INTRINSIC_W_SIDE_EFFECTS.
def : GINodeEquiv<G_INTRINSIC_W_SIDE_EFFECTS, intrinsic_void>;
Added: llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir?rev=364758&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir (added)
+++ llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcanonicalize.mir Mon Jul 1 06:22:00 2019
@@ -0,0 +1,169 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=+fp32-denormals,+fp64-fp16-denormals -run-pass=instruction-select -global-isel-abort=2 -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9-DENORM %s
+# RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-fp32-denormals,-fp64-fp16-denormals -run-pass=instruction-select -global-isel-abort=2 -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9-FLUSH %s
+
+---
+
+name: fcanonicalize_f16
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX9-DENORM-LABEL: name: fcanonicalize_f16
+ ; GFX9-DENORM: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX9-DENORM: [[V_MAX_F16_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F16_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX9-DENORM: S_ENDPGM 0, implicit [[V_MAX_F16_e64_]]
+ ; GFX9-FLUSH-LABEL: name: fcanonicalize_f16
+ ; GFX9-FLUSH: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX9-FLUSH: [[V_MUL_F16_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F16_e64 0, 15360, 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX9-FLUSH: S_ENDPGM 0, implicit [[V_MUL_F16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s16) = G_TRUNC %0
+ %2:vgpr(s16) = G_FCANONICALIZE %1
+ S_ENDPGM 0, implicit %2
+...
+
+---
+
+name: fcanonicalize_f32
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX9-DENORM-LABEL: name: fcanonicalize_f32
+ ; GFX9-DENORM: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX9-DENORM: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32 = V_MAX_F32_e64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX9-DENORM: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]]
+ ; GFX9-FLUSH-LABEL: name: fcanonicalize_f32
+ ; GFX9-FLUSH: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX9-FLUSH: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 1065353216, 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX9-FLUSH: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = G_FCANONICALIZE %0
+ S_ENDPGM 0, implicit %1
+...
+
+---
+
+name: fcanonicalize_v2f16
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX9-DENORM-LABEL: name: fcanonicalize_v2f16
+ ; GFX9-DENORM: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+ ; GFX9-DENORM: [[FCANONICALIZE:%[0-9]+]]:vgpr(<2 x s16>) = G_FCANONICALIZE [[COPY]]
+ ; GFX9-DENORM: S_ENDPGM 0, implicit [[FCANONICALIZE]](<2 x s16>)
+ ; GFX9-FLUSH-LABEL: name: fcanonicalize_v2f16
+ ; GFX9-FLUSH: [[COPY:%[0-9]+]]:vgpr(<2 x s16>) = COPY $vgpr0
+ ; GFX9-FLUSH: [[FCANONICALIZE:%[0-9]+]]:vgpr(<2 x s16>) = G_FCANONICALIZE [[COPY]]
+ ; GFX9-FLUSH: S_ENDPGM 0, implicit [[FCANONICALIZE]](<2 x s16>)
+ %0:vgpr(<2 x s16>) = COPY $vgpr0
+ %1:vgpr(<2 x s16>) = G_FCANONICALIZE %0
+ S_ENDPGM 0, implicit %1
+...
+
+---
+
+name: fcanonicalize_f64
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0_vgpr1
+ ; GFX9-DENORM-LABEL: name: fcanonicalize_f64
+ ; GFX9-DENORM: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
+ ; GFX9-DENORM: [[V_MAX_F64_:%[0-9]+]]:vreg_64 = V_MAX_F64 0, [[COPY]], 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX9-DENORM: S_ENDPGM 0, implicit [[V_MAX_F64_]]
+ ; GFX9-FLUSH-LABEL: name: fcanonicalize_f64
+ ; GFX9-FLUSH: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
+ ; GFX9-FLUSH: [[V_MUL_F64_:%[0-9]+]]:vreg_64 = V_MUL_F64 0, 4607182418800017408, 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX9-FLUSH: S_ENDPGM 0, implicit [[V_MUL_F64_]]
+ %0:vgpr(s64) = COPY $vgpr0_vgpr1
+ %1:vgpr(s64) = G_FCANONICALIZE %0
+ S_ENDPGM 0, implicit %1
+...
+
+---
+
+name: fcanonicalize_fabs_f32
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX9-DENORM-LABEL: name: fcanonicalize_fabs_f32
+ ; GFX9-DENORM: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; GFX9-DENORM: [[FABS:%[0-9]+]]:vgpr_32(s32) = G_FABS [[COPY]]
+ ; GFX9-DENORM: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32(s32) = V_MAX_F32_e64 0, [[FABS]](s32), 0, [[FABS]](s32), 0, 0, implicit $exec
+ ; GFX9-DENORM: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]](s32)
+ ; GFX9-FLUSH-LABEL: name: fcanonicalize_fabs_f32
+ ; GFX9-FLUSH: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; GFX9-FLUSH: [[FABS:%[0-9]+]]:vgpr_32(s32) = G_FABS [[COPY]]
+ ; GFX9-FLUSH: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32(s32) = V_MUL_F32_e64 0, 1065353216, 0, [[FABS]](s32), 0, 0, implicit $exec
+ ; GFX9-FLUSH: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]](s32)
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = G_FABS %0
+ %2:vgpr(s32) = G_FCANONICALIZE %1
+ S_ENDPGM 0, implicit %2
+...
+
+---
+
+name: fcanonicalize_fneg_f32
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX9-DENORM-LABEL: name: fcanonicalize_fneg_f32
+ ; GFX9-DENORM: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; GFX9-DENORM: [[FNEG:%[0-9]+]]:vgpr_32(s32) = G_FNEG [[COPY]]
+ ; GFX9-DENORM: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32(s32) = V_MAX_F32_e64 0, [[FNEG]](s32), 0, [[FNEG]](s32), 0, 0, implicit $exec
+ ; GFX9-DENORM: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]](s32)
+ ; GFX9-FLUSH-LABEL: name: fcanonicalize_fneg_f32
+ ; GFX9-FLUSH: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX9-FLUSH: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_F32_e64 0, 3212836864, 0, [[COPY]], 0, 0, implicit $exec
+ ; GFX9-FLUSH: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = G_FNEG %0
+ %2:vgpr(s32) = G_FCANONICALIZE %1
+ S_ENDPGM 0, implicit %2
+...
+
+---
+
+name: fcanonicalize_fneg_fabs_f32
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0
+ ; GFX9-DENORM-LABEL: name: fcanonicalize_fneg_fabs_f32
+ ; GFX9-DENORM: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; GFX9-DENORM: [[FNEG:%[0-9]+]]:vgpr(s32) = G_FNEG [[COPY]]
+ ; GFX9-DENORM: [[FABS:%[0-9]+]]:vgpr_32(s32) = G_FABS [[FNEG]]
+ ; GFX9-DENORM: [[V_MAX_F32_e64_:%[0-9]+]]:vgpr_32(s32) = V_MAX_F32_e64 0, [[FABS]](s32), 0, [[FABS]](s32), 0, 0, implicit $exec
+ ; GFX9-DENORM: S_ENDPGM 0, implicit [[V_MAX_F32_e64_]](s32)
+ ; GFX9-FLUSH-LABEL: name: fcanonicalize_fneg_fabs_f32
+ ; GFX9-FLUSH: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; GFX9-FLUSH: [[FNEG:%[0-9]+]]:vgpr(s32) = G_FNEG [[COPY]]
+ ; GFX9-FLUSH: [[FABS:%[0-9]+]]:vgpr_32(s32) = G_FABS [[FNEG]]
+ ; GFX9-FLUSH: [[V_MUL_F32_e64_:%[0-9]+]]:vgpr_32(s32) = V_MUL_F32_e64 0, 1065353216, 0, [[FABS]](s32), 0, 0, implicit $exec
+ ; GFX9-FLUSH: S_ENDPGM 0, implicit [[V_MUL_F32_e64_]](s32)
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = G_FNEG %0
+ %2:vgpr(s32) = G_FABS %1
+ %3:vgpr(s32) = G_FCANONICALIZE %2
+ S_ENDPGM 0, implicit %3
+...
More information about the llvm-commits
mailing list