[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Heejin Ahn via cfe-commits 

https://github.com/aheejin closed 
https://github.com/llvm/llvm-project/pull/91545
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Brendan Dahl via cfe-commits 

https://github.com/brendandahl updated 
https://github.com/llvm/llvm-project/pull/91545

>From adcb77e15d09f466f217d754f6f80aeb729aadc4 Mon Sep 17 00:00:00 2001
From: Brendan Dahl 
Date: Wed, 8 May 2024 23:10:07 +
Subject: [PATCH 1/5] [WebAssembly] Implement prototype f32.store_f16
 instruction.

Adds a builtin and intrinsic for the f32.store_f16 instruction.

The instruction stores an f32 value as an f16 memory.
Specified at:
https://github.com/WebAssembly/half-precision/blob/29a9b9462c9285d4ccc1a5dc39214ddfd1892658/proposals/half-precision/Overview.md

Note: the current spec has f32.store_f16 as opcode 0xFD0121, but this is 
incorrect
and will be changed to 0xFC31 soon.
---
 .../clang/Basic/BuiltinsWebAssembly.def   |  1 +
 clang/lib/CodeGen/CGBuiltin.cpp   |  6 +
 clang/test/CodeGen/builtins-wasm.c|  6 +
 llvm/include/llvm/IR/IntrinsicsWebAssembly.td |  5 
 .../MCTargetDesc/WebAssemblyMCTargetDesc.h|  1 +
 .../WebAssembly/WebAssemblyISelLowering.cpp   |  8 ++
 .../WebAssembly/WebAssemblyInstrMemory.td |  4 +++
 .../CodeGen/WebAssembly/half-precision.ll |  9 +++
 llvm/test/CodeGen/WebAssembly/offset.ll   | 27 +++
 llvm/test/MC/WebAssembly/simd-encodings.s |  3 +++
 10 files changed, 70 insertions(+)

diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def 
b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index cf54f8f4422f8..41fadd10e9432 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")
 
 // Reference Types builtins
 // Some builtins are custom type-checked - see 't' as part of the third 
argument,
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e8a6bd050e17e..abb644d8eb506 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -21308,6 +21308,12 @@ Value 
*CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_loadf16_f32);
 return Builder.CreateCall(Callee, {Addr});
   }
+  case WebAssembly::BI__builtin_wasm_storef16_f32: {
+Value *Val = EmitScalarExpr(E->getArg(0));
+Value *Addr = EmitScalarExpr(E->getArg(1));
+Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_storef16_f32);
+return Builder.CreateCall(Callee, {Val, Addr});
+  }
   case WebAssembly::BI__builtin_wasm_table_get: {
 assert(E->getArg(0)->getType()->isArrayType());
 Value *Table = EmitArrayToPointerDecay(E->getArg(0)).emitRawPointer(*this);
diff --git a/clang/test/CodeGen/builtins-wasm.c 
b/clang/test/CodeGen/builtins-wasm.c
index ab1c6cd494ae5..bcb15969de1c5 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -807,6 +807,12 @@ float load_f16_f32(__fp16 *addr) {
   // WEBASSEMBLY: call float @llvm.wasm.loadf16.f32(ptr %{{.*}})
 }
 
+void store_f16_f32(float val, __fp16 *addr) {
+  return __builtin_wasm_storef16_f32(val, addr);
+  // WEBASSEMBLY: tail call void @llvm.wasm.storef16.f32(float %val, ptr 
%{{.*}})
+  // WEBASSEMBLY-NEXT: ret
+}
+
 __externref_t externref_null() {
   return __builtin_wasm_ref_null_extern();
   // WEBASSEMBLY: tail call ptr addrspace(10) @llvm.wasm.ref.null.extern()
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td 
b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index f8142a8ca9e93..572d334ac9552 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -332,6 +332,11 @@ def int_wasm_loadf16_f32:
 [llvm_ptr_ty],
 [IntrReadMem, IntrArgMemOnly],
  "", [SDNPMemOperand]>;
+def int_wasm_storef16_f32:
+  Intrinsic<[],
+[llvm_float_ty, llvm_ptr_ty],
+[IntrWriteMem, IntrArgMemOnly],
+ "", [SDNPMemOperand]>;
 
 
 
//===--===//
diff --git a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
index d3b496ae59179..d4e9fb057c44d 100644
--- a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
+++ b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
@@ -207,6 +207,7 @@ inline unsigned GetDefaultP2AlignAny(unsigned Opc) {
   WASM_LOAD_STORE(LOAD_LANE_I16x8)
   WASM_LOAD_STORE(STORE_LANE_I16x8)
   WASM_LOAD_STORE(LOAD_F16_F32)
+  WASM_LOAD_STORE(STORE_F16_F32)
   return 1;
   WASM_LOAD_STORE(LOAD_I32)
   WASM_LOAD_STORE(LOAD_F32)
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp 
b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Heejin Ahn via cfe-commits 

https://github.com/aheejin approved this pull request.

LGTM with removing the pure attribute

https://github.com/llvm/llvm-project/pull/91545
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Brendan Dahl via cfe-commits 

https://github.com/brendandahl updated 
https://github.com/llvm/llvm-project/pull/91545

>From adcb77e15d09f466f217d754f6f80aeb729aadc4 Mon Sep 17 00:00:00 2001
From: Brendan Dahl 
Date: Wed, 8 May 2024 23:10:07 +
Subject: [PATCH 1/4] [WebAssembly] Implement prototype f32.store_f16
 instruction.

Adds a builtin and intrinsic for the f32.store_f16 instruction.

The instruction stores an f32 value as an f16 memory.
Specified at:
https://github.com/WebAssembly/half-precision/blob/29a9b9462c9285d4ccc1a5dc39214ddfd1892658/proposals/half-precision/Overview.md

Note: the current spec has f32.store_f16 as opcode 0xFD0121, but this is 
incorrect
and will be changed to 0xFC31 soon.
---
 .../clang/Basic/BuiltinsWebAssembly.def   |  1 +
 clang/lib/CodeGen/CGBuiltin.cpp   |  6 +
 clang/test/CodeGen/builtins-wasm.c|  6 +
 llvm/include/llvm/IR/IntrinsicsWebAssembly.td |  5 
 .../MCTargetDesc/WebAssemblyMCTargetDesc.h|  1 +
 .../WebAssembly/WebAssemblyISelLowering.cpp   |  8 ++
 .../WebAssembly/WebAssemblyInstrMemory.td |  4 +++
 .../CodeGen/WebAssembly/half-precision.ll |  9 +++
 llvm/test/CodeGen/WebAssembly/offset.ll   | 27 +++
 llvm/test/MC/WebAssembly/simd-encodings.s |  3 +++
 10 files changed, 70 insertions(+)

diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def 
b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index cf54f8f4422f8..41fadd10e9432 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")
 
 // Reference Types builtins
 // Some builtins are custom type-checked - see 't' as part of the third 
argument,
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e8a6bd050e17e..abb644d8eb506 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -21308,6 +21308,12 @@ Value 
*CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_loadf16_f32);
 return Builder.CreateCall(Callee, {Addr});
   }
+  case WebAssembly::BI__builtin_wasm_storef16_f32: {
+Value *Val = EmitScalarExpr(E->getArg(0));
+Value *Addr = EmitScalarExpr(E->getArg(1));
+Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_storef16_f32);
+return Builder.CreateCall(Callee, {Val, Addr});
+  }
   case WebAssembly::BI__builtin_wasm_table_get: {
 assert(E->getArg(0)->getType()->isArrayType());
 Value *Table = EmitArrayToPointerDecay(E->getArg(0)).emitRawPointer(*this);
diff --git a/clang/test/CodeGen/builtins-wasm.c 
b/clang/test/CodeGen/builtins-wasm.c
index ab1c6cd494ae5..bcb15969de1c5 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -807,6 +807,12 @@ float load_f16_f32(__fp16 *addr) {
   // WEBASSEMBLY: call float @llvm.wasm.loadf16.f32(ptr %{{.*}})
 }
 
+void store_f16_f32(float val, __fp16 *addr) {
+  return __builtin_wasm_storef16_f32(val, addr);
+  // WEBASSEMBLY: tail call void @llvm.wasm.storef16.f32(float %val, ptr 
%{{.*}})
+  // WEBASSEMBLY-NEXT: ret
+}
+
 __externref_t externref_null() {
   return __builtin_wasm_ref_null_extern();
   // WEBASSEMBLY: tail call ptr addrspace(10) @llvm.wasm.ref.null.extern()
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td 
b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index f8142a8ca9e93..572d334ac9552 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -332,6 +332,11 @@ def int_wasm_loadf16_f32:
 [llvm_ptr_ty],
 [IntrReadMem, IntrArgMemOnly],
  "", [SDNPMemOperand]>;
+def int_wasm_storef16_f32:
+  Intrinsic<[],
+[llvm_float_ty, llvm_ptr_ty],
+[IntrWriteMem, IntrArgMemOnly],
+ "", [SDNPMemOperand]>;
 
 
 
//===--===//
diff --git a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
index d3b496ae59179..d4e9fb057c44d 100644
--- a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
+++ b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
@@ -207,6 +207,7 @@ inline unsigned GetDefaultP2AlignAny(unsigned Opc) {
   WASM_LOAD_STORE(LOAD_LANE_I16x8)
   WASM_LOAD_STORE(STORE_LANE_I16x8)
   WASM_LOAD_STORE(LOAD_F16_F32)
+  WASM_LOAD_STORE(STORE_F16_F32)
   return 1;
   WASM_LOAD_STORE(LOAD_I32)
   WASM_LOAD_STORE(LOAD_F32)
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp 
b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Heejin Ahn via cfe-commits 


@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")

aheejin wrote:

I'm not exactly sure about the definition, but Builtins.def says it should be 
in sync with `Builtin::Context` class:
https://github.com/llvm/llvm-project/blob/b942c24845a39e6161c8623b1efc4e2083d879e9/clang/include/clang/Basic/Builtins.def#L67-L68

And `Builtin::Context` says this:
https://github.com/llvm/llvm-project/blob/b942c24845a39e6161c8623b1efc4e2083d879e9/clang/include/clang/Basic/Builtins.h#L115-L118

So I guess stores cannot be considered as pure, because they have side effects.

https://github.com/llvm/llvm-project/pull/91545
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Heejin Ahn via cfe-commits 


@@ -171,12 +171,16 @@ defm STORE8_I64 : WebAssemblyStore;
 defm STORE16_I64 : WebAssemblyStore;
 defm STORE32_I64 : WebAssemblyStore;
 
+defm STORE_F16_F32 : WebAssemblyStore;

aheejin wrote:

How about adding one for `LOAD` too?

https://github.com/llvm/llvm-project/pull/91545
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Derek Schuff via cfe-commits 


@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")

dschuff wrote:

The only documentation I know of for this is 
https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html
Clang has https://clang.llvm.org/docs/AttributeReference.html but it's not 
helpful here.

I guess a store shouldn't be pure, since it has an observable side effect, and 
in any case it doesn't return a value.

https://github.com/llvm/llvm-project/pull/91545
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Brendan Dahl via cfe-commits 

https://github.com/brendandahl updated 
https://github.com/llvm/llvm-project/pull/91545

>From adcb77e15d09f466f217d754f6f80aeb729aadc4 Mon Sep 17 00:00:00 2001
From: Brendan Dahl 
Date: Wed, 8 May 2024 23:10:07 +
Subject: [PATCH 1/3] [WebAssembly] Implement prototype f32.store_f16
 instruction.

Adds a builtin and intrinsic for the f32.store_f16 instruction.

The instruction stores an f32 value as an f16 memory.
Specified at:
https://github.com/WebAssembly/half-precision/blob/29a9b9462c9285d4ccc1a5dc39214ddfd1892658/proposals/half-precision/Overview.md

Note: the current spec has f32.store_f16 as opcode 0xFD0121, but this is 
incorrect
and will be changed to 0xFC31 soon.
---
 .../clang/Basic/BuiltinsWebAssembly.def   |  1 +
 clang/lib/CodeGen/CGBuiltin.cpp   |  6 +
 clang/test/CodeGen/builtins-wasm.c|  6 +
 llvm/include/llvm/IR/IntrinsicsWebAssembly.td |  5 
 .../MCTargetDesc/WebAssemblyMCTargetDesc.h|  1 +
 .../WebAssembly/WebAssemblyISelLowering.cpp   |  8 ++
 .../WebAssembly/WebAssemblyInstrMemory.td |  4 +++
 .../CodeGen/WebAssembly/half-precision.ll |  9 +++
 llvm/test/CodeGen/WebAssembly/offset.ll   | 27 +++
 llvm/test/MC/WebAssembly/simd-encodings.s |  3 +++
 10 files changed, 70 insertions(+)

diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def 
b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index cf54f8f4422f..41fadd10e943 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")
 
 // Reference Types builtins
 // Some builtins are custom type-checked - see 't' as part of the third 
argument,
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e8a6bd050e17..abb644d8eb50 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -21308,6 +21308,12 @@ Value 
*CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_loadf16_f32);
 return Builder.CreateCall(Callee, {Addr});
   }
+  case WebAssembly::BI__builtin_wasm_storef16_f32: {
+Value *Val = EmitScalarExpr(E->getArg(0));
+Value *Addr = EmitScalarExpr(E->getArg(1));
+Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_storef16_f32);
+return Builder.CreateCall(Callee, {Val, Addr});
+  }
   case WebAssembly::BI__builtin_wasm_table_get: {
 assert(E->getArg(0)->getType()->isArrayType());
 Value *Table = EmitArrayToPointerDecay(E->getArg(0)).emitRawPointer(*this);
diff --git a/clang/test/CodeGen/builtins-wasm.c 
b/clang/test/CodeGen/builtins-wasm.c
index ab1c6cd494ae..bcb15969de1c 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -807,6 +807,12 @@ float load_f16_f32(__fp16 *addr) {
   // WEBASSEMBLY: call float @llvm.wasm.loadf16.f32(ptr %{{.*}})
 }
 
+void store_f16_f32(float val, __fp16 *addr) {
+  return __builtin_wasm_storef16_f32(val, addr);
+  // WEBASSEMBLY: tail call void @llvm.wasm.storef16.f32(float %val, ptr 
%{{.*}})
+  // WEBASSEMBLY-NEXT: ret
+}
+
 __externref_t externref_null() {
   return __builtin_wasm_ref_null_extern();
   // WEBASSEMBLY: tail call ptr addrspace(10) @llvm.wasm.ref.null.extern()
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td 
b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index f8142a8ca9e9..572d334ac955 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -332,6 +332,11 @@ def int_wasm_loadf16_f32:
 [llvm_ptr_ty],
 [IntrReadMem, IntrArgMemOnly],
  "", [SDNPMemOperand]>;
+def int_wasm_storef16_f32:
+  Intrinsic<[],
+[llvm_float_ty, llvm_ptr_ty],
+[IntrWriteMem, IntrArgMemOnly],
+ "", [SDNPMemOperand]>;
 
 
 
//===--===//
diff --git a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
index d3b496ae5917..d4e9fb057c44 100644
--- a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
+++ b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
@@ -207,6 +207,7 @@ inline unsigned GetDefaultP2AlignAny(unsigned Opc) {
   WASM_LOAD_STORE(LOAD_LANE_I16x8)
   WASM_LOAD_STORE(STORE_LANE_I16x8)
   WASM_LOAD_STORE(LOAD_F16_F32)
+  WASM_LOAD_STORE(STORE_F16_F32)
   return 1;
   WASM_LOAD_STORE(LOAD_I32)
   WASM_LOAD_STORE(LOAD_F32)
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp 
b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Brendan Dahl via cfe-commits 

https://github.com/brendandahl updated 
https://github.com/llvm/llvm-project/pull/91545

>From adcb77e15d09f466f217d754f6f80aeb729aadc4 Mon Sep 17 00:00:00 2001
From: Brendan Dahl 
Date: Wed, 8 May 2024 23:10:07 +
Subject: [PATCH 1/2] [WebAssembly] Implement prototype f32.store_f16
 instruction.

Adds a builtin and intrinsic for the f32.store_f16 instruction.

The instruction stores an f32 value as an f16 memory.
Specified at:
https://github.com/WebAssembly/half-precision/blob/29a9b9462c9285d4ccc1a5dc39214ddfd1892658/proposals/half-precision/Overview.md

Note: the current spec has f32.store_f16 as opcode 0xFD0121, but this is 
incorrect
and will be changed to 0xFC31 soon.
---
 .../clang/Basic/BuiltinsWebAssembly.def   |  1 +
 clang/lib/CodeGen/CGBuiltin.cpp   |  6 +
 clang/test/CodeGen/builtins-wasm.c|  6 +
 llvm/include/llvm/IR/IntrinsicsWebAssembly.td |  5 
 .../MCTargetDesc/WebAssemblyMCTargetDesc.h|  1 +
 .../WebAssembly/WebAssemblyISelLowering.cpp   |  8 ++
 .../WebAssembly/WebAssemblyInstrMemory.td |  4 +++
 .../CodeGen/WebAssembly/half-precision.ll |  9 +++
 llvm/test/CodeGen/WebAssembly/offset.ll   | 27 +++
 llvm/test/MC/WebAssembly/simd-encodings.s |  3 +++
 10 files changed, 70 insertions(+)

diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def 
b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index cf54f8f4422f..41fadd10e943 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")
 
 // Reference Types builtins
 // Some builtins are custom type-checked - see 't' as part of the third 
argument,
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e8a6bd050e17..abb644d8eb50 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -21308,6 +21308,12 @@ Value 
*CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_loadf16_f32);
 return Builder.CreateCall(Callee, {Addr});
   }
+  case WebAssembly::BI__builtin_wasm_storef16_f32: {
+Value *Val = EmitScalarExpr(E->getArg(0));
+Value *Addr = EmitScalarExpr(E->getArg(1));
+Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_storef16_f32);
+return Builder.CreateCall(Callee, {Val, Addr});
+  }
   case WebAssembly::BI__builtin_wasm_table_get: {
 assert(E->getArg(0)->getType()->isArrayType());
 Value *Table = EmitArrayToPointerDecay(E->getArg(0)).emitRawPointer(*this);
diff --git a/clang/test/CodeGen/builtins-wasm.c 
b/clang/test/CodeGen/builtins-wasm.c
index ab1c6cd494ae..bcb15969de1c 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -807,6 +807,12 @@ float load_f16_f32(__fp16 *addr) {
   // WEBASSEMBLY: call float @llvm.wasm.loadf16.f32(ptr %{{.*}})
 }
 
+void store_f16_f32(float val, __fp16 *addr) {
+  return __builtin_wasm_storef16_f32(val, addr);
+  // WEBASSEMBLY: tail call void @llvm.wasm.storef16.f32(float %val, ptr 
%{{.*}})
+  // WEBASSEMBLY-NEXT: ret
+}
+
 __externref_t externref_null() {
   return __builtin_wasm_ref_null_extern();
   // WEBASSEMBLY: tail call ptr addrspace(10) @llvm.wasm.ref.null.extern()
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td 
b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index f8142a8ca9e9..572d334ac955 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -332,6 +332,11 @@ def int_wasm_loadf16_f32:
 [llvm_ptr_ty],
 [IntrReadMem, IntrArgMemOnly],
  "", [SDNPMemOperand]>;
+def int_wasm_storef16_f32:
+  Intrinsic<[],
+[llvm_float_ty, llvm_ptr_ty],
+[IntrWriteMem, IntrArgMemOnly],
+ "", [SDNPMemOperand]>;
 
 
 
//===--===//
diff --git a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
index d3b496ae5917..d4e9fb057c44 100644
--- a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
+++ b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
@@ -207,6 +207,7 @@ inline unsigned GetDefaultP2AlignAny(unsigned Opc) {
   WASM_LOAD_STORE(LOAD_LANE_I16x8)
   WASM_LOAD_STORE(STORE_LANE_I16x8)
   WASM_LOAD_STORE(LOAD_F16_F32)
+  WASM_LOAD_STORE(STORE_F16_F32)
   return 1;
   WASM_LOAD_STORE(LOAD_I32)
   WASM_LOAD_STORE(LOAD_F32)
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp 
b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-09 Thread Brendan Dahl via cfe-commits 


@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")

brendandahl wrote:

What does `pure` mean in this context? The docs in clang/Basic/Builtins.def 
don't have any info on this.

https://github.com/llvm/llvm-project/pull/91545
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-08 Thread Heejin Ahn via cfe-commits 


@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")

aheejin wrote:

Can stores be considered pure?

https://github.com/llvm/llvm-project/pull/91545
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-08 Thread Heejin Ahn via cfe-commits 


@@ -171,12 +171,16 @@ defm STORE8_I64 : WebAssemblyStore;
 defm STORE16_I64 : WebAssemblyStore;
 defm STORE32_I64 : WebAssemblyStore;
 
+defm STORE_F16_F32 : WebAssemblyStore;

aheejin wrote:

```suggestion
// Half-precision store.
defm STORE_F16_F32 :
  WebAssemblyStore;
```
How about adding one-line comment? The same for `LOAD` too.
Also 80-col wrapping.

https://github.com/llvm/llvm-project/pull/91545
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-08 Thread via cfe-commits 

llvmbot wrote:




@llvm/pr-subscribers-llvm-ir

Author: Brendan Dahl (brendandahl)


Changes

Adds a builtin and intrinsic for the f32.store_f16 instruction.

The instruction stores an f32 value as an f16 memory. Specified at:
https://github.com/WebAssembly/half-precision/blob/29a9b9462c9285d4ccc1a5dc39214ddfd1892658/proposals/half-precision/Overview.md

Note: the current spec has f32.store_f16 as opcode 0xFD0121, but this is 
incorrect and will be changed to 0xFC31 soon.

---
Full diff: https://github.com/llvm/llvm-project/pull/91545.diff


10 Files Affected:

- (modified) clang/include/clang/Basic/BuiltinsWebAssembly.def (+1) 
- (modified) clang/lib/CodeGen/CGBuiltin.cpp (+6) 
- (modified) clang/test/CodeGen/builtins-wasm.c (+6) 
- (modified) llvm/include/llvm/IR/IntrinsicsWebAssembly.td (+5) 
- (modified) llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
(+1) 
- (modified) llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp (+8) 
- (modified) llvm/lib/Target/WebAssembly/WebAssemblyInstrMemory.td (+4) 
- (modified) llvm/test/CodeGen/WebAssembly/half-precision.ll (+9) 
- (modified) llvm/test/CodeGen/WebAssembly/offset.ll (+27) 
- (modified) llvm/test/MC/WebAssembly/simd-encodings.s (+3) 


``diff
diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def 
b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index cf54f8f4422f8..41fadd10e9432 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")
 
 // Reference Types builtins
 // Some builtins are custom type-checked - see 't' as part of the third 
argument,
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e8a6bd050e17e..abb644d8eb506 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -21308,6 +21308,12 @@ Value 
*CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_loadf16_f32);
 return Builder.CreateCall(Callee, {Addr});
   }
+  case WebAssembly::BI__builtin_wasm_storef16_f32: {
+Value *Val = EmitScalarExpr(E->getArg(0));
+Value *Addr = EmitScalarExpr(E->getArg(1));
+Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_storef16_f32);
+return Builder.CreateCall(Callee, {Val, Addr});
+  }
   case WebAssembly::BI__builtin_wasm_table_get: {
 assert(E->getArg(0)->getType()->isArrayType());
 Value *Table = EmitArrayToPointerDecay(E->getArg(0)).emitRawPointer(*this);
diff --git a/clang/test/CodeGen/builtins-wasm.c 
b/clang/test/CodeGen/builtins-wasm.c
index ab1c6cd494ae5..bcb15969de1c5 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -807,6 +807,12 @@ float load_f16_f32(__fp16 *addr) {
   // WEBASSEMBLY: call float @llvm.wasm.loadf16.f32(ptr %{{.*}})
 }
 
+void store_f16_f32(float val, __fp16 *addr) {
+  return __builtin_wasm_storef16_f32(val, addr);
+  // WEBASSEMBLY: tail call void @llvm.wasm.storef16.f32(float %val, ptr 
%{{.*}})
+  // WEBASSEMBLY-NEXT: ret
+}
+
 __externref_t externref_null() {
   return __builtin_wasm_ref_null_extern();
   // WEBASSEMBLY: tail call ptr addrspace(10) @llvm.wasm.ref.null.extern()
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td 
b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index f8142a8ca9e93..572d334ac9552 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -332,6 +332,11 @@ def int_wasm_loadf16_f32:
 [llvm_ptr_ty],
 [IntrReadMem, IntrArgMemOnly],
  "", [SDNPMemOperand]>;
+def int_wasm_storef16_f32:
+  Intrinsic<[],
+[llvm_float_ty, llvm_ptr_ty],
+[IntrWriteMem, IntrArgMemOnly],
+ "", [SDNPMemOperand]>;
 
 
 
//===--===//
diff --git a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
index d3b496ae59179..d4e9fb057c44d 100644
--- a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
+++ b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
@@ -207,6 +207,7 @@ inline unsigned GetDefaultP2AlignAny(unsigned Opc) {
   WASM_LOAD_STORE(LOAD_LANE_I16x8)
   WASM_LOAD_STORE(STORE_LANE_I16x8)
   WASM_LOAD_STORE(LOAD_F16_F32)
+  WASM_LOAD_STORE(STORE_F16_F32)
   return 1;
   WASM_LOAD_STORE(LOAD_I32)
   WASM_LOAD_STORE(LOAD_F32)
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp 
b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index ed52fe53bc609..527bb4c9fbea6 100644
---

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-08 Thread via cfe-commits 

llvmbot wrote:



@llvm/pr-subscribers-mc

@llvm/pr-subscribers-backend-webassembly

Author: Brendan Dahl (brendandahl)


Changes

Adds a builtin and intrinsic for the f32.store_f16 instruction.

The instruction stores an f32 value as an f16 memory. Specified at:
https://github.com/WebAssembly/half-precision/blob/29a9b9462c9285d4ccc1a5dc39214ddfd1892658/proposals/half-precision/Overview.md

Note: the current spec has f32.store_f16 as opcode 0xFD0121, but this is 
incorrect and will be changed to 0xFC31 soon.

---
Full diff: https://github.com/llvm/llvm-project/pull/91545.diff


10 Files Affected:

- (modified) clang/include/clang/Basic/BuiltinsWebAssembly.def (+1) 
- (modified) clang/lib/CodeGen/CGBuiltin.cpp (+6) 
- (modified) clang/test/CodeGen/builtins-wasm.c (+6) 
- (modified) llvm/include/llvm/IR/IntrinsicsWebAssembly.td (+5) 
- (modified) llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
(+1) 
- (modified) llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp (+8) 
- (modified) llvm/lib/Target/WebAssembly/WebAssemblyInstrMemory.td (+4) 
- (modified) llvm/test/CodeGen/WebAssembly/half-precision.ll (+9) 
- (modified) llvm/test/CodeGen/WebAssembly/offset.ll (+27) 
- (modified) llvm/test/MC/WebAssembly/simd-encodings.s (+3) 


``diff
diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def 
b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index cf54f8f4422f8..41fadd10e9432 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")
 
 // Reference Types builtins
 // Some builtins are custom type-checked - see 't' as part of the third 
argument,
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e8a6bd050e17e..abb644d8eb506 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -21308,6 +21308,12 @@ Value 
*CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_loadf16_f32);
 return Builder.CreateCall(Callee, {Addr});
   }
+  case WebAssembly::BI__builtin_wasm_storef16_f32: {
+Value *Val = EmitScalarExpr(E->getArg(0));
+Value *Addr = EmitScalarExpr(E->getArg(1));
+Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_storef16_f32);
+return Builder.CreateCall(Callee, {Val, Addr});
+  }
   case WebAssembly::BI__builtin_wasm_table_get: {
 assert(E->getArg(0)->getType()->isArrayType());
 Value *Table = EmitArrayToPointerDecay(E->getArg(0)).emitRawPointer(*this);
diff --git a/clang/test/CodeGen/builtins-wasm.c 
b/clang/test/CodeGen/builtins-wasm.c
index ab1c6cd494ae5..bcb15969de1c5 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -807,6 +807,12 @@ float load_f16_f32(__fp16 *addr) {
   // WEBASSEMBLY: call float @llvm.wasm.loadf16.f32(ptr %{{.*}})
 }
 
+void store_f16_f32(float val, __fp16 *addr) {
+  return __builtin_wasm_storef16_f32(val, addr);
+  // WEBASSEMBLY: tail call void @llvm.wasm.storef16.f32(float %val, ptr 
%{{.*}})
+  // WEBASSEMBLY-NEXT: ret
+}
+
 __externref_t externref_null() {
   return __builtin_wasm_ref_null_extern();
   // WEBASSEMBLY: tail call ptr addrspace(10) @llvm.wasm.ref.null.extern()
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td 
b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index f8142a8ca9e93..572d334ac9552 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -332,6 +332,11 @@ def int_wasm_loadf16_f32:
 [llvm_ptr_ty],
 [IntrReadMem, IntrArgMemOnly],
  "", [SDNPMemOperand]>;
+def int_wasm_storef16_f32:
+  Intrinsic<[],
+[llvm_float_ty, llvm_ptr_ty],
+[IntrWriteMem, IntrArgMemOnly],
+ "", [SDNPMemOperand]>;
 
 
 
//===--===//
diff --git a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
index d3b496ae59179..d4e9fb057c44d 100644
--- a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
+++ b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
@@ -207,6 +207,7 @@ inline unsigned GetDefaultP2AlignAny(unsigned Opc) {
   WASM_LOAD_STORE(LOAD_LANE_I16x8)
   WASM_LOAD_STORE(STORE_LANE_I16x8)
   WASM_LOAD_STORE(LOAD_F16_F32)
+  WASM_LOAD_STORE(STORE_F16_F32)
   return 1;
   WASM_LOAD_STORE(LOAD_I32)
   WASM_LOAD_STORE(LOAD_F32)
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp 
b/llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp
index ed52fe53bc609..527bb4c9fbea6

[clang] [llvm] [WebAssembly] Implement prototype f32.store_f16 instruction. (PR #91545)

2024-05-08 Thread Brendan Dahl via cfe-commits 

https://github.com/brendandahl created 
https://github.com/llvm/llvm-project/pull/91545

Adds a builtin and intrinsic for the f32.store_f16 instruction.

The instruction stores an f32 value as an f16 memory. Specified at:
https://github.com/WebAssembly/half-precision/blob/29a9b9462c9285d4ccc1a5dc39214ddfd1892658/proposals/half-precision/Overview.md

Note: the current spec has f32.store_f16 as opcode 0xFD0121, but this is 
incorrect and will be changed to 0xFC31 soon.

>From adcb77e15d09f466f217d754f6f80aeb729aadc4 Mon Sep 17 00:00:00 2001
From: Brendan Dahl 
Date: Wed, 8 May 2024 23:10:07 +
Subject: [PATCH] [WebAssembly] Implement prototype f32.store_f16 instruction.

Adds a builtin and intrinsic for the f32.store_f16 instruction.

The instruction stores an f32 value as an f16 memory.
Specified at:
https://github.com/WebAssembly/half-precision/blob/29a9b9462c9285d4ccc1a5dc39214ddfd1892658/proposals/half-precision/Overview.md

Note: the current spec has f32.store_f16 as opcode 0xFD0121, but this is 
incorrect
and will be changed to 0xFC31 soon.
---
 .../clang/Basic/BuiltinsWebAssembly.def   |  1 +
 clang/lib/CodeGen/CGBuiltin.cpp   |  6 +
 clang/test/CodeGen/builtins-wasm.c|  6 +
 llvm/include/llvm/IR/IntrinsicsWebAssembly.td |  5 
 .../MCTargetDesc/WebAssemblyMCTargetDesc.h|  1 +
 .../WebAssembly/WebAssemblyISelLowering.cpp   |  8 ++
 .../WebAssembly/WebAssemblyInstrMemory.td |  4 +++
 .../CodeGen/WebAssembly/half-precision.ll |  9 +++
 llvm/test/CodeGen/WebAssembly/offset.ll   | 27 +++
 llvm/test/MC/WebAssembly/simd-encodings.s |  3 +++
 10 files changed, 70 insertions(+)

diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def 
b/clang/include/clang/Basic/BuiltinsWebAssembly.def
index cf54f8f4422f8..41fadd10e9432 100644
--- a/clang/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -192,6 +192,7 @@ 
TARGET_BUILTIN(__builtin_wasm_relaxed_dot_bf16x8_add_f32_f32x4, "V4fV8UsV8UsV4f"
 
 // Half-Precision (fp16)
 TARGET_BUILTIN(__builtin_wasm_loadf16_f32, "fh*", "nU", "half-precision")
+TARGET_BUILTIN(__builtin_wasm_storef16_f32, "vfh*", "nU", "half-precision")
 
 // Reference Types builtins
 // Some builtins are custom type-checked - see 't' as part of the third 
argument,
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index e8a6bd050e17e..abb644d8eb506 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -21308,6 +21308,12 @@ Value 
*CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
 Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_loadf16_f32);
 return Builder.CreateCall(Callee, {Addr});
   }
+  case WebAssembly::BI__builtin_wasm_storef16_f32: {
+Value *Val = EmitScalarExpr(E->getArg(0));
+Value *Addr = EmitScalarExpr(E->getArg(1));
+Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_storef16_f32);
+return Builder.CreateCall(Callee, {Val, Addr});
+  }
   case WebAssembly::BI__builtin_wasm_table_get: {
 assert(E->getArg(0)->getType()->isArrayType());
 Value *Table = EmitArrayToPointerDecay(E->getArg(0)).emitRawPointer(*this);
diff --git a/clang/test/CodeGen/builtins-wasm.c 
b/clang/test/CodeGen/builtins-wasm.c
index ab1c6cd494ae5..bcb15969de1c5 100644
--- a/clang/test/CodeGen/builtins-wasm.c
+++ b/clang/test/CodeGen/builtins-wasm.c
@@ -807,6 +807,12 @@ float load_f16_f32(__fp16 *addr) {
   // WEBASSEMBLY: call float @llvm.wasm.loadf16.f32(ptr %{{.*}})
 }
 
+void store_f16_f32(float val, __fp16 *addr) {
+  return __builtin_wasm_storef16_f32(val, addr);
+  // WEBASSEMBLY: tail call void @llvm.wasm.storef16.f32(float %val, ptr 
%{{.*}})
+  // WEBASSEMBLY-NEXT: ret
+}
+
 __externref_t externref_null() {
   return __builtin_wasm_ref_null_extern();
   // WEBASSEMBLY: tail call ptr addrspace(10) @llvm.wasm.ref.null.extern()
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td 
b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
index f8142a8ca9e93..572d334ac9552 100644
--- a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
+++ b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -332,6 +332,11 @@ def int_wasm_loadf16_f32:
 [llvm_ptr_ty],
 [IntrReadMem, IntrArgMemOnly],
  "", [SDNPMemOperand]>;
+def int_wasm_storef16_f32:
+  Intrinsic<[],
+[llvm_float_ty, llvm_ptr_ty],
+[IntrWriteMem, IntrArgMemOnly],
+ "", [SDNPMemOperand]>;
 
 
 
//===--===//
diff --git a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h 
b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
index d3b496ae59179..d4e9fb057c44d 100644
--- a/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
+++ b/llvm/lib/Target/WebAssembly/MCTargetDesc/WebAssemblyMCTargetDesc.h
@@ -207,6 +207,7 @@ inline unsigned