[llvm] [AMDGPU][True16][Codegen] true16 codegen for FPtoI1 (PR #125120)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Jan 30 17:53:07 PST 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Brox Chen (broxigarchen)
<details>
<summary>Changes</summary>
True16 codegen for FPtoi1.
It seems tablegen figured out the pattern even without this pat in place, and the fptoui/fptosi.ll already got the right transformation. Aditionally updated the mir file and split it to pre-gfx11 and post-gfx11.
---
Patch is 43.13 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/125120.diff
4 Files Affected:
- (modified) llvm/lib/Target/AMDGPU/SIInstructions.td (+9-4)
- (added) llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.gfx11plus-fake16.mir (+374)
- (added) llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.gfx11plus.mir (+402)
- (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.mir (-117)
``````````diff
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 5af46989aca97b..f1ad9383ae7597 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -2680,15 +2680,20 @@ class FPToI1Pat<Instruction Inst, int KOne, ValueType kone_type, ValueType vt, S
(i1 (Inst 0, (kone_type KOne), $src0_modifiers, $src0, DSTCLAMP.NONE))
>;
-let OtherPredicates = [NotHasTrue16BitInsts] in {
+let True16Predicate = NotHasTrue16BitInsts in {
def : FPToI1Pat<V_CMP_EQ_F16_e64, CONST.FP16_ONE, i16, f16, fp_to_uint>;
def : FPToI1Pat<V_CMP_EQ_F16_e64, CONST.FP16_NEG_ONE, i16, f16, fp_to_sint>;
-} // end OtherPredicates = [NotHasTrue16BitInsts]
+} // end True16Predicate = NotHasTrue16BitInsts
+
+let True16Predicate = UseRealTrue16Insts in {
+ def : FPToI1Pat<V_CMP_EQ_F16_t16_e64, CONST.FP16_ONE, i16, f16, fp_to_uint>;
+ def : FPToI1Pat<V_CMP_EQ_F16_t16_e64, CONST.FP16_NEG_ONE, i16, f16, fp_to_sint>;
+} // end True16Predicate = UseRealTrue16BitInsts
-let OtherPredicates = [HasTrue16BitInsts] in {
+let True16Predicate = UseFakeTrue16Insts in {
def : FPToI1Pat<V_CMP_EQ_F16_fake16_e64, CONST.FP16_ONE, i16, f16, fp_to_uint>;
def : FPToI1Pat<V_CMP_EQ_F16_fake16_e64, CONST.FP16_NEG_ONE, i16, f16, fp_to_sint>;
-} // end OtherPredicates = [HasTrue16BitInsts]
+} // end True16Predicate = UseFakeTrue16BitInsts
def : FPToI1Pat<V_CMP_EQ_F32_e64, CONST.FP32_ONE, i32, f32, fp_to_uint>;
def : FPToI1Pat<V_CMP_EQ_F32_e64, CONST.FP32_NEG_ONE, i32, f32, fp_to_sint>;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.gfx11plus-fake16.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.gfx11plus-fake16.mir
new file mode 100644
index 00000000000000..23da26d96b6291
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.gfx11plus-fake16.mir
@@ -0,0 +1,374 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -run-pass=instruction-select -mattr=-real-true16 -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX11 %s
+
+---
+name: fcmp_false_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_false_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; GFX11-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
+ ; GFX11-NEXT: [[TRUNC1:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY1]](s32)
+ ; GFX11-NEXT: [[FCMP:%[0-9]+]]:vcc(s1) = G_FCMP floatpred(false), [[TRUNC]](s16), [[TRUNC1]]
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[FCMP]](s1)
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(false), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_oeq_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_oeq_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_EQ_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_EQ_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_EQ_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(oeq), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ogt_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ogt_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_GT_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_GT_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_GT_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(ogt), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_oge_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_oge_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_GE_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_GE_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_GE_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(oge), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_olt_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_olt_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_LT_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_LT_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_LT_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(olt), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ole_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ole_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_LE_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_LE_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_LE_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(ole), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+---
+name: fcmp_one_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_one_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_LG_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_LG_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_LG_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(one), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ord_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ord_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_LG_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_LG_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_LG_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(one), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_uno_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_uno_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_U_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_U_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_U_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(uno), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ueq_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ueq_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_NLG_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_NLG_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_NLG_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(ueq), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ugt_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ugt_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_NLE_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_NLE_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_NLE_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(ugt), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_uge_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_uge_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_NLT_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_NLT_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_NLT_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(uge), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ult_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ult_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_NGE_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_NGE_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_NGE_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(ult), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ule_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ule_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_NGT_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_NGT_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_NGT_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(ule), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_une_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_une_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[V_CMP_NEQ_F16_fake16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_NEQ_F16_fake16_e64 0, [[COPY]], 0, [[COPY1]], 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_NEQ_F16_fake16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(une), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_true_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_true_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; GFX11-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
+ ; GFX11-NEXT: [[TRUNC1:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY1]](s32)
+ ; GFX11-NEXT: [[FCMP:%[0-9]+]]:vcc(s1) = G_FCMP floatpred(true), [[TRUNC]](s16), [[TRUNC1]]
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[FCMP]](s1)
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(true), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.gfx11plus.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.gfx11plus.mir
new file mode 100644
index 00000000000000..a7140e6a74fd4d
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-fcmp.s16.gfx11plus.mir
@@ -0,0 +1,402 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -run-pass=instruction-select -mattr=+real-true16 -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX11 %s
+
+---
+name: fcmp_false_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_false_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+ ; GFX11-NEXT: [[TRUNC:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY]](s32)
+ ; GFX11-NEXT: [[TRUNC1:%[0-9]+]]:vgpr(s16) = G_TRUNC [[COPY1]](s32)
+ ; GFX11-NEXT: [[FCMP:%[0-9]+]]:vcc(s1) = G_FCMP floatpred(false), [[TRUNC]](s16), [[TRUNC1]]
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[FCMP]](s1)
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(false), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_oeq_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_oeq_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
+ ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_16 = COPY [[COPY1]].lo16
+ ; GFX11-NEXT: [[V_CMP_EQ_F16_t16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_EQ_F16_t16_e64 0, [[COPY2]], 0, [[COPY3]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_EQ_F16_t16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(oeq), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ogt_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ogt_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
+ ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_16 = COPY [[COPY1]].lo16
+ ; GFX11-NEXT: [[V_CMP_GT_F16_t16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_GT_F16_t16_e64 0, [[COPY2]], 0, [[COPY3]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_GT_F16_t16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(ogt), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_oge_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_oge_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
+ ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_16 = COPY [[COPY1]].lo16
+ ; GFX11-NEXT: [[V_CMP_GE_F16_t16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_GE_F16_t16_e64 0, [[COPY2]], 0, [[COPY3]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_GE_F16_t16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(oge), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_olt_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_olt_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = COPY [[COPY]].lo16
+ ; GFX11-NEXT: [[COPY3:%[0-9]+]]:vgpr_16 = COPY [[COPY1]].lo16
+ ; GFX11-NEXT: [[V_CMP_LT_F16_t16_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_LT_F16_t16_e64 0, [[COPY2]], 0, [[COPY3]], 0, 0, implicit $mode, implicit $exec
+ ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_CMP_LT_F16_t16_e64_]]
+ %0:vgpr(s32) = COPY $vgpr0
+ %1:vgpr(s32) = COPY $vgpr1
+ %2:vgpr(s16) = G_TRUNC %0
+ %3:vgpr(s16) = G_TRUNC %1
+ %4:vcc(s1) = G_FCMP floatpred(olt), %2, %3
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: fcmp_ole_s16_vv
+legalized: true
+regBankSelected: true
+
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1
+ ; GFX11-LABEL: name: fcmp_ole_s16_vv
+ ; GFX11: liveins: $vgpr0, $vgpr1
+ ; GFX11-NEXT: {{ $}}
+ ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+ ; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+ ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_16 = CO...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/125120
More information about the llvm-commits
mailing list