[libc] Add aarch64 flavors of floor, round, sqrt and trunc.

[sivachandra]

Summary

Only single and double precision flavors have been added.

[sdesmalen]

libc/src/math/aarch64/floor.cpp

16

Hi @sivachandra sorry for the drive-by comment on your patch, I just happened notice the use of ldr/str here. Is there a specific reason for that? I think this can simply use the register values instead, e.g:

double result;
__asm__ __volatile__("frintm %d0, %d1\n"
                     : "=w"(result) : "w"(x) : );
return result;

That way, it will just create a frintm d0, d0 instruction directly.

[sivachandra]

Use %d and %s modifiers instead of the registers directly.

[sivachandra]

libc/src/math/aarch64/floor.cpp

16

Thanks a lot for you comment. I wasn't aware that one could use %d and %s modifiers. I have applied your suggestion everywhere.

[sdesmalen]

Thanks for the changes. I noticed another patch (maybe there are more?) that could probably also be updated now, e.g. ceil/ceilf from D95850.

[sivachandra]

In D95999#2544557, @sdesmalen wrote:

Thanks for the changes. I noticed another patch (maybe there are more?) that could probably also be updated now, e.g. ceil/ceilf from D95850.

I have updated ceil and ceilf also in this change.

libc/src/math/aarch64/CMakeLists.txt

add_entrypoint_object(
   ceilf
   SRCS
     ceilf.cpp
   HDRS
     ../ceilf.h
   COMPILE_OPTIONS
     -O2
 )
+
+add_entrypoint_object(
+  floor
+  SRCS
+    floor.cpp
+  HDRS
+    ../floor.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  floorf
+  SRCS
+    floorf.cpp
+  HDRS
+    ../floorf.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  trunc
+  SRCS
+    trunc.cpp
+  HDRS
+    ../trunc.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  truncf
+  SRCS
+    truncf.cpp
+  HDRS
+    ../truncf.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  round
+  SRCS
+    round.cpp
+  HDRS
+    ../round.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  roundf
+  SRCS
+    roundf.cpp
+  HDRS
+    ../roundf.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  sqrt
+  SRCS
+    sqrt.cpp
+  HDRS
+    ../sqrt.h
+  COMPILE_OPTIONS
+    -O2
+)
+
+add_entrypoint_object(
+  sqrtf
+  SRCS
+    sqrtf.cpp
+  HDRS
+    ../sqrtf.h
+  COMPILE_OPTIONS
+    -O2
+)

libc/src/math/aarch64/ceil.cpp

 //===-- Implementation of the ceil function for aarch64 -------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #include "src/math/ceil.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
 LLVM_LIBC_FUNCTION(double, ceil, (double x)) {
   double y;
-  __asm__ __volatile__("ldr d0, %1\n"
-                       "frintp d0, d0\n"
-                       "str d0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "d0");
+  __asm__ __volatile__("frintp %d0, %d1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/ceilf.cpp

 //===-- Implementation of the ceilf function for aarch64 ------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #include "src/math/ceilf.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
 LLVM_LIBC_FUNCTION(float, ceilf, (float x)) {
   float y;
-  __asm__ __volatile__("ldr s0, %1\n"
-                       "frintp s0, s0\n"
-                       "str s0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "s0");
+  __asm__ __volatile__("frintp %s0, %s1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/floor.cpp

-//===-- Implementation of the ceil function for aarch64 -------------------===//
+//===-- Implementation of the floor function for aarch64 ------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "src/math/ceil.h"
+#include "src/math/floor.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
-LLVM_LIBC_FUNCTION(double, ceil, (double x)) {
+LLVM_LIBC_FUNCTION(double, floor, (double x)) {
   double y;
-  __asm__ __volatile__("ldr d0, %1\n"
-                       "frintp d0, d0\n"
-                       "str d0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "d0");
+  __asm__ __volatile__("frintm %d0, %d1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/floorf.cpp

-//===-- Implementation of the ceilf function for aarch64 ------------------===//
+//===-- Implementation of the floorf function for aarch64 -----------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "src/math/ceilf.h"
+#include "src/math/floorf.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
-LLVM_LIBC_FUNCTION(float, ceilf, (float x)) {
+LLVM_LIBC_FUNCTION(float, floorf, (float x)) {
   float y;
-  __asm__ __volatile__("ldr s0, %1\n"
-                       "frintp s0, s0\n"
-                       "str s0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "s0");
+  __asm__ __volatile__("frintm %s0, %s1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/round.cpp

-//===-- Implementation of the ceil function for aarch64 -------------------===//
+//===-- Implementation of the round function for aarch64 ------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "src/math/ceil.h"
+#include "src/math/round.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
-LLVM_LIBC_FUNCTION(double, ceil, (double x)) {
+LLVM_LIBC_FUNCTION(double, round, (double x)) {
   double y;
-  __asm__ __volatile__("ldr d0, %1\n"
-                       "frintp d0, d0\n"
-                       "str d0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "d0");
+  __asm__ __volatile__("frinta %d0, %d1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/roundf.cpp

-//===-- Implementation of the ceilf function for aarch64 ------------------===//
+//===-- Implementation of the roundf function for aarch64 -----------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "src/math/ceilf.h"
+#include "src/math/roundf.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
-LLVM_LIBC_FUNCTION(float, ceilf, (float x)) {
+LLVM_LIBC_FUNCTION(float, roundf, (float x)) {
   float y;
-  __asm__ __volatile__("ldr s0, %1\n"
-                       "frintp s0, s0\n"
-                       "str s0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "s0");
+  __asm__ __volatile__("frinta %s0, %s1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/sqrt.cpp

-//===-- Implementation of the ceil function for aarch64 -------------------===//
+//===-- Implementation of the sqrt function for aarch64 -------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "src/math/ceil.h"
+#include "src/math/sqrt.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
-LLVM_LIBC_FUNCTION(double, ceil, (double x)) {
+LLVM_LIBC_FUNCTION(double, sqrt, (double x)) {
   double y;
-  __asm__ __volatile__("ldr d0, %1\n"
-                       "frintp d0, d0\n"
-                       "str d0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "d0");
+  __asm__ __volatile__("fsqrt %d0, %d1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/sqrtf.cpp

-//===-- Implementation of the ceilf function for aarch64 ------------------===//
+//===-- Implementation of the sqrtf function for aarch64 ------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "src/math/ceilf.h"
+#include "src/math/sqrtf.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
-LLVM_LIBC_FUNCTION(float, ceilf, (float x)) {
+LLVM_LIBC_FUNCTION(float, sqrtf, (float x)) {
   float y;
-  __asm__ __volatile__("ldr s0, %1\n"
-                       "frintp s0, s0\n"
-                       "str s0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "s0");
+  __asm__ __volatile__("fsqrt %s0, %s1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/trunc.cpp

-//===-- Implementation of the ceil function for aarch64 -------------------===//
+//===-- Implementation of the trunc function for aarch64 ------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "src/math/ceil.h"
+#include "src/math/trunc.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
-LLVM_LIBC_FUNCTION(double, ceil, (double x)) {
+LLVM_LIBC_FUNCTION(double, trunc, (double x)) {
   double y;
-  __asm__ __volatile__("ldr d0, %1\n"
-                       "frintp d0, d0\n"
-                       "str d0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "d0");
+  __asm__ __volatile__("frintz %d0, %d1\n" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc

libc/src/math/aarch64/truncf.cpp

-//===-- Implementation of the ceilf function for aarch64 ------------------===//
+//===-- Implementation of the truncf function for aarch64 -----------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#include "src/math/ceilf.h"
+#include "src/math/truncf.h"
 #include "src/__support/common.h"
 
 namespace __llvm_libc {
 
-LLVM_LIBC_FUNCTION(float, ceilf, (float x)) {
+LLVM_LIBC_FUNCTION(float, truncf, (float x)) {
   float y;
-  __asm__ __volatile__("ldr s0, %1\n"
-                       "frintp s0, s0\n"
-                       "str s0, %0\n"
-                       : "=m"(y)
-                       : "m"(x)
-                       : "s0");
+  __asm__ __volatile__("frintz %s0, %s1\n\t" : "=w"(y) : "w"(x));
   return y;
 }
 
 } // namespace __llvm_libc