[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-11-02 Thread Florian Hahn via Phabricator via cfe-commits
This revision was landed with ongoing or failed builds.
This revision was automatically updated to reflect the committed changes.
Closed by commit rG7999355106fb: [Clang] Add min/max reduction builtins. 
(authored by fhahn).

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

Files:
  clang/include/clang/Basic/Builtins.def
  clang/include/clang/Basic/DiagnosticSemaKinds.td
  clang/include/clang/Sema/Sema.h
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/lib/Sema/SemaChecking.cpp
  clang/test/CodeGen/builtins-reduction-math.c
  clang/test/Sema/builtins-reduction-math.c

Index: clang/test/Sema/builtins-reduction-math.c
===
--- /dev/null
+++ clang/test/Sema/builtins-reduction-math.c
@@ -0,0 +1,37 @@
+// RUN: %clang_cc1 %s -pedantic -verify -triple=x86_64-apple-darwin9
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef int int3 __attribute__((ext_vector_type(3)));
+typedef unsigned unsigned4 __attribute__((ext_vector_type(4)));
+
+struct Foo {
+  char *p;
+};
+
+void test_builtin_reduce_max(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_max(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_max(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_max();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_max(i);
+  // expected-error@-1 {{1st argument must be a vector type (was 'int')}}
+}
+
+void test_builtin_reduce_min(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_min(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_min(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_min();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_min(i);
+  // expected-error@-1 {{1st argument must be a vector type (was 'int')}}
+}
Index: clang/test/CodeGen/builtins-reduction-math.c
===
--- /dev/null
+++ clang/test/CodeGen/builtins-reduction-math.c
@@ -0,0 +1,59 @@
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -emit-llvm -disable-llvm-passes -o - | FileCheck %s
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef short int si8 __attribute__((ext_vector_type(8)));
+typedef unsigned int u4 __attribute__((ext_vector_type(4)));
+
+__attribute__((address_space(1))) float4 vf1_as_one;
+
+void test_builtin_reduce_max(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_max(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_max(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_max(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_max(vu1);
+
+  // CHECK:  [[VF1_AS1:%.+]] = load <4 x float>, <4 x float> addrspace(1)* @vf1_as_one, align 16
+  // CHECK-NEXT: [[RDX1:%.+]] = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1_AS1]])
+  // CHECK-NEXT: fpext float [[RDX1]] to double
+  const double r4 = __builtin_reduce_max(vf1_as_one);
+
+  // CHECK:  [[CVI1:%.+]] = load <8 x i16>, <8 x i16>* %cvi1, align 16
+  // CHECK-NEXT: [[RDX2:%.+]] = call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[CVI1]])
+  // CHECK-NEXT: sext i16 [[RDX2]] to i64
+  const si8 cvi1 = vi1;
+  unsigned long long r5 = __builtin_reduce_max(cvi1);
+}
+
+void test_builtin_reduce_min(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_min(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmin.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_min(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smin.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_min(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_min(vu1);
+
+  // CHECK:  [[VF1_AS1:%.+]] = load <4 x float>, <4 x float> addrspace(1)* @vf1_as_one, align 16
+  // CHECK-NEXT: [[RDX1:%.+]] = call float @llvm.vect

[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-11-02 Thread Aaron Ballman via Phabricator via cfe-commits
aaron.ballman accepted this revision.
aaron.ballman added a comment.
This revision is now accepted and ready to land.

LGTM!


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits


[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-11-01 Thread Florian Hahn via Phabricator via cfe-commits
fhahn updated this revision to Diff 383900.
fhahn added a comment.

In D112001#3099965 , @aaron.ballman 
wrote:

> Precommit CI looks to be failing.
>
>   
> /var/lib/buildkite-agent/builds/llvm-project/clang/test/CodeGen/builtins-reduction-math.c:13:9:
>  error: initializing 'float' with an expression of incompatible type 'float4' 
> (vector of 4 'float' values)
>   
> float r1 = __builtin_reduce_max(vf1);

The last version of the patch did not correctly set the return type to the 
element type. That should be fixed now!


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

Files:
  clang/include/clang/Basic/Builtins.def
  clang/include/clang/Basic/DiagnosticSemaKinds.td
  clang/include/clang/Sema/Sema.h
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/lib/Sema/SemaChecking.cpp
  clang/test/CodeGen/builtins-reduction-math.c
  clang/test/Sema/builtins-reduction-math.c

Index: clang/test/Sema/builtins-reduction-math.c
===
--- /dev/null
+++ clang/test/Sema/builtins-reduction-math.c
@@ -0,0 +1,37 @@
+// RUN: %clang_cc1 %s -pedantic -verify -triple=x86_64-apple-darwin9
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef int int3 __attribute__((ext_vector_type(3)));
+typedef unsigned unsigned4 __attribute__((ext_vector_type(4)));
+
+struct Foo {
+  char *p;
+};
+
+void test_builtin_reduce_max(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_max(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_max(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_max();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_max(i);
+  // expected-error@-1 {{1st argument must be a vector type (was 'int')}}
+}
+
+void test_builtin_reduce_min(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_min(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_min(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_min();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_min(i);
+  // expected-error@-1 {{1st argument must be a vector type (was 'int')}}
+}
Index: clang/test/CodeGen/builtins-reduction-math.c
===
--- /dev/null
+++ clang/test/CodeGen/builtins-reduction-math.c
@@ -0,0 +1,59 @@
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -emit-llvm -disable-llvm-passes -o - | FileCheck %s
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef short int si8 __attribute__((ext_vector_type(8)));
+typedef unsigned int u4 __attribute__((ext_vector_type(4)));
+
+__attribute__((address_space(1))) float4 vf1_as_one;
+
+void test_builtin_reduce_max(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_max(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_max(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_max(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_max(vu1);
+
+  // CHECK:  [[VF1_AS1:%.+]] = load <4 x float>, <4 x float> addrspace(1)* @vf1_as_one, align 16
+  // CHECK-NEXT: [[RDX1:%.+]] = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1_AS1]])
+  // CHECK-NEXT: fpext float [[RDX1]] to double
+  const double r4 = __builtin_reduce_max(vf1_as_one);
+
+  // CHECK:  [[CVI1:%.+]] = load <8 x i16>, <8 x i16>* %cvi1, align 16
+  // CHECK-NEXT: [[RDX2:%.+]] = call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[CVI1]])
+  // CHECK-NEXT: sext i16 [[RDX2]] to i64
+  const si8 cvi1 = vi1;
+  unsigned long long r5 = __builtin_reduce_max(cvi1);
+}
+
+void test_builtin_reduce_min(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_min(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmin.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_min(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smin.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_min(vi1);
+
+  // CH

[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-11-01 Thread Aaron Ballman via Phabricator via cfe-commits
aaron.ballman added a comment.

Precommit CI looks to be failing.

  
/var/lib/buildkite-agent/builds/llvm-project/clang/test/CodeGen/builtins-reduction-math.c:13:9:
 error: initializing 'float' with an expression of incompatible type 'float4' 
(vector of 4 'float' values)
  
float r1 = __builtin_reduce_max(vf1);


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits


[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-11-01 Thread Florian Hahn via Phabricator via cfe-commits
fhahn updated this revision to Diff 383756.
fhahn added a comment.

rebase on top of recent changes.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

Files:
  clang/include/clang/Basic/Builtins.def
  clang/include/clang/Basic/DiagnosticSemaKinds.td
  clang/include/clang/Sema/Sema.h
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/lib/Sema/SemaChecking.cpp
  clang/test/CodeGen/builtins-reduction-math.c
  clang/test/Sema/builtins-reduction-math.c

Index: clang/test/Sema/builtins-reduction-math.c
===
--- /dev/null
+++ clang/test/Sema/builtins-reduction-math.c
@@ -0,0 +1,37 @@
+// RUN: %clang_cc1 %s -pedantic -verify -triple=x86_64-apple-darwin9
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef int int3 __attribute__((ext_vector_type(3)));
+typedef unsigned unsigned4 __attribute__((ext_vector_type(4)));
+
+struct Foo {
+  char *p;
+};
+
+void test_builtin_reduce_max(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_max(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int3' (vector of 3 'int' values)}}
+
+  i = __builtin_reduce_max(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_max();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_max(i);
+  // expected-error@-1 {{1st argument must be a vector type (was 'int')}}
+}
+
+void test_builtin_reduce_min(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_min(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int3' (vector of 3 'int' values)}}
+
+  i = __builtin_reduce_min(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_min();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_min(i);
+  // expected-error@-1 {{1st argument must be a vector type (was 'int')}}
+}
Index: clang/test/CodeGen/builtins-reduction-math.c
===
--- /dev/null
+++ clang/test/CodeGen/builtins-reduction-math.c
@@ -0,0 +1,59 @@
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -emit-llvm -disable-llvm-passes -o - | FileCheck %s
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef short int si8 __attribute__((ext_vector_type(8)));
+typedef unsigned int u4 __attribute__((ext_vector_type(4)));
+
+__attribute__((address_space(1))) float4 vf1_as_one;
+
+void test_builtin_reduce_max(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_max(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_max(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_max(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_max(vu1);
+
+  // CHECK:  [[VF1_AS1:%.+]] = load <4 x float>, <4 x float> addrspace(1)* @vf1_as_one, align 16
+  // CHECK-NEXT: [[RDX1:%.+]] = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1_AS1]])
+  // CHECK-NEXT: fpext float [[RDX1]] to double
+  const double r4 = __builtin_reduce_max(vf1_as_one);
+
+  // CHECK:  [[CVI1:%.+]] = load <8 x i16>, <8 x i16>* %cvi1, align 16
+  // CHECK-NEXT: [[RDX2:%.+]] = call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[CVI1]])
+  // CHECK-NEXT: sext i16 [[RDX2]] to i64
+  const si8 cvi1 = vi1;
+  unsigned long long r5 = __builtin_reduce_max(cvi1);
+}
+
+void test_builtin_reduce_min(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_min(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmin.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_min(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smin.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_min(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_min(vu1);
+
+  // CHECK:  [[VF1_AS1:%.+]] = load <4 x float>, <4 x float> addrspace(1)* @vf1_as_one, align 16
+  // CHECK-NEXT: [[RDX1:%.+]] = call float @llvm.vector.reduce.fmin.v4f32(<4 x float> [[VF1_AS1]])
+  // CHECK-NEXT: fpe

[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-10-26 Thread Aaron Ballman via Phabricator via cfe-commits
aaron.ballman added inline comments.



Comment at: clang/lib/Sema/SemaChecking.cpp:16606-16609
+  const VectorType *VecTy = TyA->getAs();
+  if (!VecTy)
+return Diag(A->getBeginLoc(), 
diag::err_elementwise_math_invalid_arg_type_2)
+   << "vector" << TyA;

What about type sugar?



Comment at: clang/lib/Sema/SemaChecking.cpp:16611
+
+  TheCall->setType(VecTy->getElementType());
+  return false;

Should this type undergo the usual conversions?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits


[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-10-26 Thread Florian Hahn via Phabricator via cfe-commits
fhahn updated this revision to Diff 382240.
fhahn added a comment.

Rebased after recent changes in D111985  
https://reviews.llvm.org/D111985: changes the return type of 
SemaBuiltinReduceMath to bool and drop the CallResult argument.

Also added tests with const and variables with address space and moved tests to 
builtins-reduction-math.c instead of builtins-elementwise-math.c


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

Files:
  clang/include/clang/Basic/Builtins.def
  clang/include/clang/Sema/Sema.h
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/lib/Sema/SemaChecking.cpp
  clang/test/CodeGen/builtins-reduction-math.c
  clang/test/Sema/builtins-reduction-math.c

Index: clang/test/Sema/builtins-reduction-math.c
===
--- /dev/null
+++ clang/test/Sema/builtins-reduction-math.c
@@ -0,0 +1,37 @@
+// RUN: %clang_cc1 %s -pedantic -verify -triple=x86_64-apple-darwin9
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef int int3 __attribute__((ext_vector_type(3)));
+typedef unsigned unsigned4 __attribute__((ext_vector_type(4)));
+
+struct Foo {
+  char *p;
+};
+
+void test_builtin_reduce_max(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_max(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_max(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_max();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_max(i);
+  // expected-error@-1 {{argument must have a vector type, but was 'int'}}
+}
+
+void test_builtin_reduce_min(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_min(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_min(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_min();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_min(i);
+  // expected-error@-1 {{argument must have a vector type, but was 'int'}}
+}
Index: clang/test/CodeGen/builtins-reduction-math.c
===
--- /dev/null
+++ clang/test/CodeGen/builtins-reduction-math.c
@@ -0,0 +1,59 @@
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -emit-llvm -disable-llvm-passes -o - | FileCheck %s
+
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef short int si8 __attribute__((ext_vector_type(8)));
+typedef unsigned int u4 __attribute__((ext_vector_type(4)));
+
+__attribute__((address_space(1))) float4 vf1_as_one;
+
+void test_builtin_reduce_max(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_max(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_max(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_max(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_max(vu1);
+
+  // CHECK:  [[VF1_AS1:%.+]] = load <4 x float>, <4 x float> addrspace(1)* @vf1_as_one, align 16
+  // CHECK-NEXT: [[RDX1:%.+]] = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1_AS1]])
+  // CHECK-NEXT: fpext float [[RDX1]] to double
+  const double r4 = __builtin_reduce_max(vf1_as_one);
+
+  // CHECK:  [[CVI1:%.+]] = load <8 x i16>, <8 x i16>* %cvi1, align 16
+  // CHECK-NEXT: [[RDX2:%.+]] = call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[CVI1]])
+  // CHECK-NEXT: sext i16 [[RDX2]] to i64
+  const si8 cvi1 = vi1;
+  unsigned long long r5 = __builtin_reduce_max(cvi1);
+}
+
+void test_builtin_reduce_min(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_min(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmin.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_min(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smin.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_min(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_min(vu1);
+
+  // CHECK:

[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-10-20 Thread Florian Hahn via Phabricator via cfe-commits
fhahn updated this revision to Diff 380948.
fhahn marked an inline comment as done.
fhahn added a comment.

rebase on top of recently updated patches.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

Files:
  clang/include/clang/Basic/Builtins.def
  clang/include/clang/Sema/Sema.h
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/lib/Sema/SemaChecking.cpp
  clang/test/CodeGen/builtins-elementwise-math.c
  clang/test/Sema/builtins-elementwise-math.c

Index: clang/test/Sema/builtins-elementwise-math.c
===
--- clang/test/Sema/builtins-elementwise-math.c
+++ clang/test/Sema/builtins-elementwise-math.c
@@ -61,3 +61,31 @@
   i = __builtin_elementwise_min(v, iv);
   // expected-error@-1 {{argument types do not match, 'float4' (vector of 4 'float' values) != 'int3' (vector of 3 'int' values)}}
 }
+
+void test_builtin_reduce_max(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_max(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_max(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_max();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_max(i);
+  // expected-error@-1 {{argument must have a vector type, but was 'int'}}
+}
+
+void test_builtin_reduce_min(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_min(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_min(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_min();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_min(i);
+  // expected-error@-1 {{argument must have a vector type, but was 'int'}}
+}
Index: clang/test/CodeGen/builtins-elementwise-math.c
===
--- clang/test/CodeGen/builtins-elementwise-math.c
+++ clang/test/CodeGen/builtins-elementwise-math.c
@@ -110,3 +110,33 @@
   // CHECK-NEXT: call <4 x i32> @llvm.umin.v4i32(<4 x i32> [[VU1]], <4 x i32> [[VU2]])
   vu1 = __builtin_elementwise_min(vu1, vu2);
 }
+
+void test_builtin_reduce_max(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_max(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_max(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_max(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_max(vu1);
+}
+
+void test_builtin_reduce_min(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_min(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmin.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_min(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smin.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_min(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_min(vu1);
+}
Index: clang/lib/Sema/SemaChecking.cpp
===
--- clang/lib/Sema/SemaChecking.cpp
+++ clang/lib/Sema/SemaChecking.cpp
@@ -1978,11 +1978,14 @@
 
   case Builtin::BI__builtin_elementwise_abs:
 return SemaBuiltinElementwiseMathOneArg(TheCall, TheCallResult);
-
   case Builtin::BI__builtin_elementwise_min:
   case Builtin::BI__builtin_elementwise_max:
 return SemaBuiltinElementwiseMathTwoArgs(TheCall, TheCallResult);
 
+  case Builtin::BI__builtin_reduce_max:
+  case Builtin::BI__builtin_reduce_min:
+return SemaBuiltinReduceMath(TheCall, TheCallResult);
+
   case Builtin::BI__builtin_matrix_transpose:
 return SemaBuiltinMatrixTranspose(TheCall, TheCallResult);
   case Builtin::BI__builtin_matrix_column_major_load:
@@ -16712,6 +16715,23 @@
   return CallResult;
 }
 
+ExprResult Sema::SemaBuiltinReduceMath(CallExpr *TheCall,
+   ExprResult CallResult) {
+  if (checkArgCount(*this, TheCall, 1))
+return ExprError();
+
+  Expr *A = TheCall->getArg(0);
+  QualType TyA = A->getType

[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-10-18 Thread Florian Hahn via Phabricator via cfe-commits
fhahn marked an inline comment as done.
fhahn added inline comments.



Comment at: clang/lib/Sema/SemaChecking.cpp:16673
+  if (!VecTy)
+  if (!TyA->getAs())
+  return Diag(A->getBeginLoc(), 
diag::err_elementwise_math_invalid_arg_type_2)

craig.topper wrote:
> Is this indented incorrectly? There appear to be 2 ifs at the same level
Yes they were, thanks! The `if (!TyA->getAs())` was a left-over 
from an earlier iteration. Removed.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits


[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-10-18 Thread Florian Hahn via Phabricator via cfe-commits
fhahn updated this revision to Diff 380486.
fhahn added a comment.

Remove stray `!TyA->getAs()` check.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

Files:
  clang/include/clang/Basic/Builtins.def
  clang/include/clang/Sema/Sema.h
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/lib/Sema/SemaChecking.cpp
  clang/test/CodeGen/builtins-elementwise-math.c
  clang/test/Sema/builtins-elementwise-math.c

Index: clang/test/Sema/builtins-elementwise-math.c
===
--- clang/test/Sema/builtins-elementwise-math.c
+++ clang/test/Sema/builtins-elementwise-math.c
@@ -61,3 +61,31 @@
   i = __builtin_elementwise_min(v, iv);
   // expected-error@-1 {{argument types do not match, 'float4' (vector of 4 'float' values) != 'int3' (vector of 3 'int' values)}}
 }
+
+void test_builtin_reduce_max(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_max(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_max(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_max();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_max(i);
+  // expected-error@-1 {{argument must have a vector type, but was 'int'}}
+}
+
+void test_builtin_reduce_min(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_min(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_min(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_min();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_min(i);
+  // expected-error@-1 {{argument must have a vector type, but was 'int'}}
+}
Index: clang/test/CodeGen/builtins-elementwise-math.c
===
--- clang/test/CodeGen/builtins-elementwise-math.c
+++ clang/test/CodeGen/builtins-elementwise-math.c
@@ -110,3 +110,33 @@
   // CHECK-NEXT: call <4 x i32> @llvm.umin.v4i32(<4 x i32> [[VU1]], <4 x i32> [[VU2]])
   vu1 = __builtin_elementwise_min(vu1, vu2);
 }
+
+void test_builtin_reduce_max(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_max(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_max(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_max(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_max(vu1);
+}
+
+void test_builtin_reduce_min(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_min(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmin.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_min(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smin.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_min(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_min(vu1);
+}
Index: clang/lib/Sema/SemaChecking.cpp
===
--- clang/lib/Sema/SemaChecking.cpp
+++ clang/lib/Sema/SemaChecking.cpp
@@ -1978,11 +1978,14 @@
 
   case Builtin::BI__builtin_elementwise_abs:
 return SemaBuiltinElementwiseMathOneArg(TheCall, TheCallResult);
-
   case Builtin::BI__builtin_elementwise_min:
   case Builtin::BI__builtin_elementwise_max:
 return SemaBuiltinElementwiseMathTwoArgs(TheCall, TheCallResult);
 
+  case Builtin::BI__builtin_reduce_max:
+  case Builtin::BI__builtin_reduce_min:
+return SemaBuiltinReduceMath(TheCall, TheCallResult);
+
   case Builtin::BI__builtin_matrix_transpose:
 return SemaBuiltinMatrixTranspose(TheCall, TheCallResult);
   case Builtin::BI__builtin_matrix_column_major_load:
@@ -16656,6 +16659,23 @@
  _2, _3, _4));
 }
 
+ExprResult Sema::SemaBuiltinReduceMath(CallExpr *TheCall,
+   ExprResult CallResult) {
+  if (checkArgCount(*this, TheCall, 1))
+return ExprError();
+
+  Expr *A = TheCall->getArg(0);
+  QualType TyA = A->getType();
+
+  const VectorType *VecTy 

[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-10-18 Thread Craig Topper via Phabricator via cfe-commits
craig.topper added inline comments.



Comment at: clang/lib/Sema/SemaChecking.cpp:16673
+  if (!VecTy)
+  if (!TyA->getAs())
+  return Diag(A->getBeginLoc(), 
diag::err_elementwise_math_invalid_arg_type_2)

Is this indented incorrectly? There appear to be 2 ifs at the same level


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112001/new/

https://reviews.llvm.org/D112001

___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits


[PATCH] D112001: [Clang] Add min/max reduction builtins.

2021-10-18 Thread Florian Hahn via Phabricator via cfe-commits
fhahn created this revision.
fhahn added reviewers: aaron.ballman, scanon, craig.topper, rjmccall, 
erichkeane.
fhahn requested review of this revision.
Herald added a project: clang.

This patch implements __builtin_reduce_max and __builtin_reduce_min as
specified in D111529 .

The order of operations does not matter for min or max reductions and
they can be directly lowered to the corresponding
llvm.vector.reduce.{fmin,fmax,umin,umax,smin,smax} intrinsic calls.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D112001

Files:
  clang/include/clang/Basic/Builtins.def
  clang/include/clang/Sema/Sema.h
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/lib/Sema/SemaChecking.cpp
  clang/test/CodeGen/builtins-elementwise-math.c
  clang/test/Sema/builtins-elementwise-math.c

Index: clang/test/Sema/builtins-elementwise-math.c
===
--- clang/test/Sema/builtins-elementwise-math.c
+++ clang/test/Sema/builtins-elementwise-math.c
@@ -61,3 +61,31 @@
   i = __builtin_elementwise_min(v, iv);
   // expected-error@-1 {{argument types do not match, 'float4' (vector of 4 'float' values) != 'int3' (vector of 3 'int' values)}}
 }
+
+void test_builtin_reduce_max(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_max(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_max(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_max();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_max(i);
+  // expected-error@-1 {{argument must have a vector type, but was 'int'}}
+}
+
+void test_builtin_reduce_min(int i, float4 v, int3 iv) {
+  struct Foo s = __builtin_reduce_min(iv);
+  // expected-error@-1 {{initializing 'struct Foo' with an expression of incompatible type 'int'}}
+
+  i = __builtin_reduce_min(v, v);
+  // expected-error@-1 {{too many arguments to function call, expected 1, have 2}}
+
+  i = __builtin_reduce_min();
+  // expected-error@-1 {{too few arguments to function call, expected 1, have 0}}
+
+  i = __builtin_reduce_min(i);
+  // expected-error@-1 {{argument must have a vector type, but was 'int'}}
+}
Index: clang/test/CodeGen/builtins-elementwise-math.c
===
--- clang/test/CodeGen/builtins-elementwise-math.c
+++ clang/test/CodeGen/builtins-elementwise-math.c
@@ -110,3 +110,33 @@
   // CHECK-NEXT: call <4 x i32> @llvm.umin.v4i32(<4 x i32> [[VU1]], <4 x i32> [[VU2]])
   vu1 = __builtin_elementwise_min(vu1, vu2);
 }
+
+void test_builtin_reduce_max(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_max(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmax.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_max(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smax.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_max(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_max(vu1);
+}
+
+void test_builtin_reduce_min(float4 vf1, si8 vi1, u4 vu1) {
+  // CHECK-LABEL: define void @test_builtin_reduce_min(
+  // CHECK:  [[VF1:%.+]] = load <4 x float>, <4 x float>* %vf1.addr, align 16
+  // CHECK-NEXT: call float @llvm.vector.reduce.fmin.v4f32(<4 x float> [[VF1]])
+  float r1 = __builtin_reduce_min(vf1);
+
+  // CHECK:  [[VI1:%.+]] = load <8 x i16>, <8 x i16>* %vi1.addr, align 16
+  // CHECK-NEXT: call i16 @llvm.vector.reduce.smin.v8i16(<8 x i16> [[VI1]])
+  short r2 = __builtin_reduce_min(vi1);
+
+  // CHECK:  [[VU1:%.+]] = load <4 x i32>, <4 x i32>* %vu1.addr, align 16
+  // CHECK-NEXT: call i32 @llvm.vector.reduce.umin.v4i32(<4 x i32> [[VU1]])
+  unsigned r3 = __builtin_reduce_min(vu1);
+}
Index: clang/lib/Sema/SemaChecking.cpp
===
--- clang/lib/Sema/SemaChecking.cpp
+++ clang/lib/Sema/SemaChecking.cpp
@@ -1978,11 +1978,15 @@
 
   case Builtin::BI__builtin_elementwise_abs:
 return SemaBuiltinElementwiseMathOneArg(TheCall, TheCallResult);
-
   case Builtin::BI__builtin_elementwise_min:
   case Builtin::BI__builtin_elementwise_max:
 return SemaBuiltinElementwiseMathTwoArgs(TheCall, TheCallResult);
 
+  case Builtin::BI__builtin_reduce_max:
+  case Builtin::BI__builtin_reduce_min:
+return SemaBuiltinReduceMath(TheCall, TheCallResult);
+
+
   case Builtin::BI__builtin_matrix_transpose:
 return SemaBuiltinMatrixTranspose(TheCall, TheCallResult);
   case Builtin::BI__builtin_matrix_column_major_load:
@@ -16656,6 +16660,