AMDGPU/GlobalISel: Legalize s64 min/max by lowering

Reviewers: arsenm, rampitec

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, rovka, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D75108

5 files changed, 57 insertions, 42 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 722b803..c0c1fab 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -635,15 +635,17 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
         .legalFor({S32, S16, V2S16})
         .moreElementsIf(isSmallOddVector(0), oneMoreElement(0))
         .clampMaxNumElements(0, S16, 2)
-        .clampScalar(0, S16, S32)
+        .minScalar(0, S16)
         .widenScalarToNextPow2(0)
-        .scalarize(0);
+        .scalarize(0)
+        .lower();
     } else {
       getActionDefinitionsBuilder({G_SMIN, G_SMAX, G_UMIN, G_UMAX})
         .legalFor({S32, S16})
         .widenScalarToNextPow2(0)
-        .clampScalar(0, S16, S32)
-        .scalarize(0);
+        .minScalar(0, S16)
+        .scalarize(0)
+        .lower();
     }
   } else {
     // TODO: Should have same legality without v_perm_b32
@@ -659,9 +661,10 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
 
     getActionDefinitionsBuilder({G_SMIN, G_SMAX, G_UMIN, G_UMAX})
       .legalFor({S32})
-      .clampScalar(0, S32, S32)
+      .minScalar(0, S32)
       .widenScalarToNextPow2(0)
-      .scalarize(0);
+      .scalarize(0)
+      .lower();
   }
 
   getActionDefinitionsBuilder(G_INTTOPTR)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir
index 9774b542..a108705 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir
@@ -1,7 +1,7 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=SI  %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=VI %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=GFX9  %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer %s -o - | FileCheck -check-prefix=SI  %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer %s -o - | FileCheck -check-prefix=VI %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9  %s
 
 ---
 name: test_smax_s32
@@ -39,18 +39,21 @@ body: |
     ; SI-LABEL: name: test_smax_s64
     ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; SI: [[SMAX:%[0-9]+]]:_(s64) = G_SMAX [[COPY]], [[COPY1]]
-    ; SI: $vgpr0_vgpr1 = COPY [[SMAX]](s64)
+    ; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(sgt), [[COPY]](s64), [[COPY1]]
+    ; SI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; SI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     ; VI-LABEL: name: test_smax_s64
     ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; VI: [[SMAX:%[0-9]+]]:_(s64) = G_SMAX [[COPY]], [[COPY1]]
-    ; VI: $vgpr0_vgpr1 = COPY [[SMAX]](s64)
+    ; VI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(sgt), [[COPY]](s64), [[COPY1]]
+    ; VI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; VI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     ; GFX9-LABEL: name: test_smax_s64
     ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; GFX9: [[SMAX:%[0-9]+]]:_(s64) = G_SMAX [[COPY]], [[COPY1]]
-    ; GFX9: $vgpr0_vgpr1 = COPY [[SMAX]](s64)
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(sgt), [[COPY]](s64), [[COPY1]]
+    ; GFX9: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; GFX9: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     %0:_(s64) = COPY $vgpr0_vgpr1
     %1:_(s64) = COPY $vgpr2_vgpr3
     %2:_(s64) = G_SMAX %0, %1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smin.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smin.mir
index bef9c62..483681f 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smin.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smin.mir
@@ -1,7 +1,7 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=SI  %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=VI %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=GFX9  %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer %s -o - | FileCheck -check-prefix=SI  %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer %s -o - | FileCheck -check-prefix=VI %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9  %s
 
 ---
 name: test_smin_s32
@@ -39,18 +39,21 @@ body: |
     ; SI-LABEL: name: test_smin_s64
     ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; SI: [[SMIN:%[0-9]+]]:_(s64) = G_SMIN [[COPY]], [[COPY1]]
-    ; SI: $vgpr0_vgpr1 = COPY [[SMIN]](s64)
+    ; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(slt), [[COPY]](s64), [[COPY1]]
+    ; SI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; SI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     ; VI-LABEL: name: test_smin_s64
     ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; VI: [[SMIN:%[0-9]+]]:_(s64) = G_SMIN [[COPY]], [[COPY1]]
-    ; VI: $vgpr0_vgpr1 = COPY [[SMIN]](s64)
+    ; VI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(slt), [[COPY]](s64), [[COPY1]]
+    ; VI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; VI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     ; GFX9-LABEL: name: test_smin_s64
     ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; GFX9: [[SMIN:%[0-9]+]]:_(s64) = G_SMIN [[COPY]], [[COPY1]]
-    ; GFX9: $vgpr0_vgpr1 = COPY [[SMIN]](s64)
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(slt), [[COPY]](s64), [[COPY1]]
+    ; GFX9: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; GFX9: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     %0:_(s64) = COPY $vgpr0_vgpr1
     %1:_(s64) = COPY $vgpr2_vgpr3
     %2:_(s64) = G_SMIN %0, %1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umax.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umax.mir
index da9ce8b..7a1bf1b 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umax.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umax.mir
@@ -1,7 +1,7 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=SI  %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=VI %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=GFX9  %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer %s -o - | FileCheck -check-prefix=SI  %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer %s -o - | FileCheck -check-prefix=VI %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9  %s
 
 ---
 name: test_umax_s32
@@ -39,18 +39,21 @@ body: |
     ; SI-LABEL: name: test_umax_s64
     ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; SI: [[UMAX:%[0-9]+]]:_(s64) = G_UMAX [[COPY]], [[COPY1]]
-    ; SI: $vgpr0_vgpr1 = COPY [[UMAX]](s64)
+    ; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[COPY]](s64), [[COPY1]]
+    ; SI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; SI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     ; VI-LABEL: name: test_umax_s64
     ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; VI: [[UMAX:%[0-9]+]]:_(s64) = G_UMAX [[COPY]], [[COPY1]]
-    ; VI: $vgpr0_vgpr1 = COPY [[UMAX]](s64)
+    ; VI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[COPY]](s64), [[COPY1]]
+    ; VI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; VI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     ; GFX9-LABEL: name: test_umax_s64
     ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; GFX9: [[UMAX:%[0-9]+]]:_(s64) = G_UMAX [[COPY]], [[COPY1]]
-    ; GFX9: $vgpr0_vgpr1 = COPY [[UMAX]](s64)
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ugt), [[COPY]](s64), [[COPY1]]
+    ; GFX9: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; GFX9: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     %0:_(s64) = COPY $vgpr0_vgpr1
     %1:_(s64) = COPY $vgpr2_vgpr3
     %2:_(s64) = G_UMAX %0, %1
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umin.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umin.mir
index c1fd2cd..de480c0 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umin.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-umin.mir
@@ -1,7 +1,7 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=SI  %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=VI %s
-# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=GFX9  %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer %s -o - | FileCheck -check-prefix=SI  %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer %s -o - | FileCheck -check-prefix=VI %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9  %s
 
 ---
 name: test_umin_s32
@@ -39,18 +39,21 @@ body: |
     ; SI-LABEL: name: test_umin_s64
     ; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; SI: [[UMIN:%[0-9]+]]:_(s64) = G_UMIN [[COPY]], [[COPY1]]
-    ; SI: $vgpr0_vgpr1 = COPY [[UMIN]](s64)
+    ; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ult), [[COPY]](s64), [[COPY1]]
+    ; SI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; SI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     ; VI-LABEL: name: test_umin_s64
     ; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; VI: [[UMIN:%[0-9]+]]:_(s64) = G_UMIN [[COPY]], [[COPY1]]
-    ; VI: $vgpr0_vgpr1 = COPY [[UMIN]](s64)
+    ; VI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ult), [[COPY]](s64), [[COPY1]]
+    ; VI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; VI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     ; GFX9-LABEL: name: test_umin_s64
     ; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
     ; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
-    ; GFX9: [[UMIN:%[0-9]+]]:_(s64) = G_UMIN [[COPY]], [[COPY1]]
-    ; GFX9: $vgpr0_vgpr1 = COPY [[UMIN]](s64)
+    ; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ult), [[COPY]](s64), [[COPY1]]
+    ; GFX9: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
+    ; GFX9: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
     %0:_(s64) = COPY $vgpr0_vgpr1
     %1:_(s64) = COPY $vgpr2_vgpr3
     %2:_(s64) = G_UMIN %0, %1