[MIPS GlobalISel] Select MSA vector generic and builtin fsqrt

selectImpl is able to select G_FSQRT when we set bank for vector
operands to fprb. Add detailed tests.
Note: G_FSQRT is generated from llvm-ir intrinsics llvm.sqrt.*,
and at the moment MIPS is not able to generate this intrinsic for
vector type (some targets generate vector llvm.sqrt.* from calls
to a builtin function).
__builtin_msa_fsqrt_<format> will be transformed into G_FSQRT
in legalizeIntrinsic and selected in the same way.

Differential Revision: https://reviews.llvm.org/D69376

6 files changed, 303 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/Mips/GlobalISel/instruction-select/fsqrt_vec.mir b/llvm/test/CodeGen/Mips/GlobalISel/instruction-select/fsqrt_vec.mir
new file mode 100644
index 00000000000..5a81540f294
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/GlobalISel/instruction-select/fsqrt_vec.mir
@@ -0,0 +1,60 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=mipsel-linux-gnu -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=P5600
+--- |
+
+  define void @sqrt_v4f32(<4 x float>* %a, <4 x float>* %c) { entry: ret void }
+  define void @sqrt_v2f64(<2 x double>* %a, <2 x double>* %c) { entry: ret void }
+
+...
+---
+name:            sqrt_v4f32
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: sqrt_v4f32
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:gpr32 = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:gpr32 = COPY $a1
+    ; P5600: [[LD_W:%[0-9]+]]:msa128w = LD_W [[COPY]], 0 :: (load 16 from %ir.a)
+    ; P5600: [[FSQRT_W:%[0-9]+]]:msa128w = FSQRT_W [[LD_W]]
+    ; P5600: ST_W [[FSQRT_W]], [[COPY1]], 0 :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:gprb(p0) = COPY $a0
+    %1:gprb(p0) = COPY $a1
+    %2:fprb(<4 x s32>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:fprb(<4 x s32>) = G_FSQRT %2
+    G_STORE %3(<4 x s32>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+---
+name:            sqrt_v2f64
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: sqrt_v2f64
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:gpr32 = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:gpr32 = COPY $a1
+    ; P5600: [[LD_D:%[0-9]+]]:msa128d = LD_D [[COPY]], 0 :: (load 16 from %ir.a)
+    ; P5600: [[FSQRT_D:%[0-9]+]]:msa128d = FSQRT_D [[LD_D]]
+    ; P5600: ST_D [[FSQRT_D]], [[COPY1]], 0 :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:gprb(p0) = COPY $a0
+    %1:gprb(p0) = COPY $a1
+    %2:fprb(<2 x s64>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:fprb(<2 x s64>) = G_FSQRT %2
+    G_STORE %3(<2 x s64>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
diff --git a/llvm/test/CodeGen/Mips/GlobalISel/legalizer/fsqrt_vec.mir b/llvm/test/CodeGen/Mips/GlobalISel/legalizer/fsqrt_vec.mir
new file mode 100644
index 00000000000..a5994b6e88f
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/GlobalISel/legalizer/fsqrt_vec.mir
@@ -0,0 +1,56 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=mipsel-linux-gnu -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=P5600
+--- |
+
+  define void @sqrt_v4f32(<4 x float>* %a, <4 x float>* %c) { entry: ret void }
+  define void @sqrt_v2f64(<2 x double>* %a, <2 x double>* %c) { entry: ret void }
+
+...
+---
+name:            sqrt_v4f32
+alignment:       4
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: sqrt_v4f32
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:_(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:_(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FSQRT:%[0-9]+]]:_(<4 x s32>) = G_FSQRT [[LOAD]]
+    ; P5600: G_STORE [[FSQRT]](<4 x s32>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<4 x s32>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<4 x s32>) = G_FSQRT %2
+    G_STORE %3(<4 x s32>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+---
+name:            sqrt_v2f64
+alignment:       4
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: sqrt_v2f64
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:_(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:_(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FSQRT:%[0-9]+]]:_(<2 x s64>) = G_FSQRT [[LOAD]]
+    ; P5600: G_STORE [[FSQRT]](<2 x s64>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<2 x s64>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<2 x s64>) = G_FSQRT %2
+    G_STORE %3(<2 x s64>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
diff --git a/llvm/test/CodeGen/Mips/GlobalISel/legalizer/fsqrt_vec_builtin.mir b/llvm/test/CodeGen/Mips/GlobalISel/legalizer/fsqrt_vec_builtin.mir
new file mode 100644
index 00000000000..e6d31789a48
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/GlobalISel/legalizer/fsqrt_vec_builtin.mir
@@ -0,0 +1,60 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=mipsel-linux-gnu -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=P5600
+--- |
+
+  declare <4 x float> @llvm.mips.fsqrt.w(<4 x float>)
+  define void @fsqrt_v4f32_builtin(<4 x float>* %a, <4 x float>* %c) { entry: ret void }
+
+  declare <2 x double> @llvm.mips.fsqrt.d(<2 x double>)
+  define void @fsqrt_v2f64_builtin(<2 x double>* %a, <2 x double>* %c) { entry: ret void }
+
+...
+---
+name:            fsqrt_v4f32_builtin
+alignment:       4
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fsqrt_v4f32_builtin
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:_(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:_(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:_(<4 x s32>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FSQRT:%[0-9]+]]:_(<4 x s32>) = G_FSQRT [[LOAD]]
+    ; P5600: G_STORE [[FSQRT]](<4 x s32>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<4 x s32>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<4 x s32>) = G_INTRINSIC intrinsic(@llvm.mips.fsqrt.w), %2(<4 x s32>)
+    G_STORE %3(<4 x s32>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+---
+name:            fsqrt_v2f64_builtin
+alignment:       4
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: fsqrt_v2f64_builtin
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:_(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:_(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:_(<2 x s64>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FSQRT:%[0-9]+]]:_(<2 x s64>) = G_FSQRT [[LOAD]]
+    ; P5600: G_STORE [[FSQRT]](<2 x s64>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<2 x s64>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<2 x s64>) = G_INTRINSIC intrinsic(@llvm.mips.fsqrt.d), %2(<2 x s64>)
+    G_STORE %3(<2 x s64>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+
diff --git a/llvm/test/CodeGen/Mips/GlobalISel/llvm-ir/fsqrt_vec.ll b/llvm/test/CodeGen/Mips/GlobalISel/llvm-ir/fsqrt_vec.ll
new file mode 100644
index 00000000000..27bbb9aa59c
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/GlobalISel/llvm-ir/fsqrt_vec.ll
@@ -0,0 +1,34 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -O0 -mtriple=mipsel-linux-gnu -global-isel -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -verify-machineinstrs %s -o -| FileCheck %s -check-prefixes=P5600
+
+declare <4 x float>  @llvm.sqrt.v4f32(<4 x float>  %Val)
+define void @sqrt_v4f32(<4 x float>* %a, <4 x float>* %c) {
+; P5600-LABEL: sqrt_v4f32:
+; P5600:       # %bb.0: # %entry
+; P5600-NEXT:    ld.w $w0, 0($4)
+; P5600-NEXT:    fsqrt.w $w0, $w0
+; P5600-NEXT:    st.w $w0, 0($5)
+; P5600-NEXT:    jr $ra
+; P5600-NEXT:    nop
+entry:
+  %0 = load <4 x float>, <4 x float>* %a, align 16
+  %sqrt = call <4 x float> @llvm.sqrt.v4f32 (<4 x float> %0)
+  store <4 x float> %sqrt, <4 x float>* %c, align 16
+  ret void
+}
+
+declare <2 x double> @llvm.sqrt.v2f64(<2 x double> %Val)
+define void @sqrt_v2f64(<2 x double>* %a, <2 x double>* %c) {
+; P5600-LABEL: sqrt_v2f64:
+; P5600:       # %bb.0: # %entry
+; P5600-NEXT:    ld.d $w0, 0($4)
+; P5600-NEXT:    fsqrt.d $w0, $w0
+; P5600-NEXT:    st.d $w0, 0($5)
+; P5600-NEXT:    jr $ra
+; P5600-NEXT:    nop
+entry:
+  %0 = load <2 x double>, <2 x double>* %a, align 16
+  %sqrt = call <2 x double> @llvm.sqrt.v2f64 (<2 x double> %0)
+  store <2 x double> %sqrt, <2 x double>* %c, align 16
+  ret void
+}
diff --git a/llvm/test/CodeGen/Mips/GlobalISel/llvm-ir/fsqrt_vec_builtin.ll b/llvm/test/CodeGen/Mips/GlobalISel/llvm-ir/fsqrt_vec_builtin.ll
new file mode 100644
index 00000000000..a765591d42f
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/GlobalISel/llvm-ir/fsqrt_vec_builtin.ll
@@ -0,0 +1,35 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -O0 -mtriple=mipsel-linux-gnu -global-isel -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -verify-machineinstrs %s -o -| FileCheck %s -check-prefixes=P5600
+
+declare <4 x float> @llvm.mips.fsqrt.w(<4 x float>)
+define void @fsqrt_v4f32_builtin(<4 x float>* %a, <4 x float>* %c) {
+; P5600-LABEL: fsqrt_v4f32_builtin:
+; P5600:       # %bb.0: # %entry
+; P5600-NEXT:    ld.w $w0, 0($4)
+; P5600-NEXT:    fsqrt.w $w0, $w0
+; P5600-NEXT:    st.w $w0, 0($5)
+; P5600-NEXT:    jr $ra
+; P5600-NEXT:    nop
+entry:
+  %0 = load <4 x float>, <4 x float>* %a, align 16
+  %1 = tail call <4 x float> @llvm.mips.fsqrt.w(<4 x float> %0)
+  store <4 x float> %1, <4 x float>* %c, align 16
+  ret void
+}
+
+declare <2 x double> @llvm.mips.fsqrt.d(<2 x double>)
+define void @fsqrt_v2f64_builtin(<2 x double>* %a, <2 x double>* %c) {
+; P5600-LABEL: fsqrt_v2f64_builtin:
+; P5600:       # %bb.0: # %entry
+; P5600-NEXT:    ld.d $w0, 0($4)
+; P5600-NEXT:    fsqrt.d $w0, $w0
+; P5600-NEXT:    st.d $w0, 0($5)
+; P5600-NEXT:    jr $ra
+; P5600-NEXT:    nop
+entry:
+  %0 = load <2 x double>, <2 x double>* %a, align 16
+  %1 = tail call <2 x double> @llvm.mips.fsqrt.d(<2 x double> %0)
+  store <2 x double> %1, <2 x double>* %c, align 16
+  ret void
+}
+
diff --git a/llvm/test/CodeGen/Mips/GlobalISel/regbankselect/fsqrt_vec.mir b/llvm/test/CodeGen/Mips/GlobalISel/regbankselect/fsqrt_vec.mir
new file mode 100644
index 00000000000..d36a0e51977
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/GlobalISel/regbankselect/fsqrt_vec.mir
@@ -0,0 +1,58 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=mipsel-linux-gnu -mcpu=mips32r5 -mattr=+msa,+fp64,+nan2008 -run-pass=regbankselect -verify-machineinstrs %s -o - | FileCheck %s -check-prefixes=P5600
+--- |
+
+  define void @sqrt_v4f32(<4 x float>* %a, <4 x float>* %c) { entry: ret void }
+  define void @sqrt_v2f64(<2 x double>* %a, <2 x double>* %c) { entry: ret void }
+
+...
+---
+name:            sqrt_v4f32
+alignment:       4
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: sqrt_v4f32
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:gprb(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:gprb(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:fprb(<4 x s32>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FSQRT:%[0-9]+]]:fprb(<4 x s32>) = G_FSQRT [[LOAD]]
+    ; P5600: G_STORE [[FSQRT]](<4 x s32>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<4 x s32>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<4 x s32>) = G_FSQRT %2
+    G_STORE %3(<4 x s32>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...
+---
+name:            sqrt_v2f64
+alignment:       4
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.1.entry:
+    liveins: $a0, $a1
+
+    ; P5600-LABEL: name: sqrt_v2f64
+    ; P5600: liveins: $a0, $a1
+    ; P5600: [[COPY:%[0-9]+]]:gprb(p0) = COPY $a0
+    ; P5600: [[COPY1:%[0-9]+]]:gprb(p0) = COPY $a1
+    ; P5600: [[LOAD:%[0-9]+]]:fprb(<2 x s64>) = G_LOAD [[COPY]](p0) :: (load 16 from %ir.a)
+    ; P5600: [[FSQRT:%[0-9]+]]:fprb(<2 x s64>) = G_FSQRT [[LOAD]]
+    ; P5600: G_STORE [[FSQRT]](<2 x s64>), [[COPY1]](p0) :: (store 16 into %ir.c)
+    ; P5600: RetRA
+    %0:_(p0) = COPY $a0
+    %1:_(p0) = COPY $a1
+    %2:_(<2 x s64>) = G_LOAD %0(p0) :: (load 16 from %ir.a)
+    %3:_(<2 x s64>) = G_FSQRT %2
+    G_STORE %3(<2 x s64>), %1(p0) :: (store 16 into %ir.c)
+    RetRA
+
+...