Scalar realization of no-vec functions was replaced from builtins to stdlib.ispc.

2014-01-20 16:06:34 +04:00
parent 97cc5b7f48
commit da02236b3a
7 changed files with 94 additions and 88 deletions
--- a/builtins.cpp
+++ b/builtins.cpp
@@ -488,12 +488,8 @@ lSetInternalFunctions(llvm::Module *module) {
        "__num_cores",
        "__packed_load_active",
        "__packed_store_active",
-        "__padds_i8",
-        "__padds_i16",
        "__padds_vi8",
        "__padds_vi16",
-        "__paddus_i8",
-        "__paddus_i16",
        "__paddus_vi8",
        "__paddus_vi16",
        "__popcnt_int32",
@@ -502,12 +498,8 @@ lSetInternalFunctions(llvm::Module *module) {
        "__prefetch_read_uniform_2",
        "__prefetch_read_uniform_3",
        "__prefetch_read_uniform_nt",
-        "__psubs_i8",
-        "__psubs_i16",
        "__psubs_vi8",
        "__psubs_vi16",
-        "__psubus_i8",
-        "__psubus_i16",
        "__psubus_vi8",
        "__psubus_vi16",
        "__rcp_uniform_float",
--- a/builtins/target-avx-common.ll
+++ b/builtins/target-avx-common.ll
@@ -40,7 +40,6 @@ ctlztz()
 define_prefetches()
 define_shuffles()
 aossoa()
-saturation_arithmetic_uniform()

 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; rounding floats
--- a/builtins/target-generic-common.ll
+++ b/builtins/target-generic-common.ll
@@ -41,7 +41,6 @@ stdlib_core()
 scans()
 reduce_equal(WIDTH)
 rdrand_decls()
-saturation_arithmetic_uniform()

 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; broadcast/rotate/shuffle
--- a/builtins/target-sse2-common.ll
+++ b/builtins/target-sse2-common.ll
@@ -34,7 +34,6 @@ define_prefetches()
 define_shuffles()
 aossoa()
 rdrand_decls()
-saturation_arithmetic_uniform()

 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; rcp
--- a/builtins/target-sse4-common.ll
+++ b/builtins/target-sse4-common.ll
@@ -37,7 +37,6 @@ define_prefetches()
 define_shuffles()
 aossoa()
 rdrand_decls()
-saturation_arithmetic_uniform()

 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;; rounding floats
--- a/builtins/util.m4
+++ b/builtins/util.m4
@@ -184,73 +184,7 @@ define(`convert32to16', `
 define(`saturation_arithmetic',
 `ifelse(WIDTH,  `4', `saturation_arithmetic_vec4()', 
        WIDTH,  `8', `saturation_arithmetic_vec8()',
-        WIDTH, `16', `saturation_arithmetic_vec16()',
-                     `saturation_arithmetic_uniform()')')
-
-;; utility function used by saturation_arithmetic_uniform below.  This shouldn't be called by
-;; target .ll files directly.
-;; $1: {add,sub} (used in constructing function names)
-
-define(`saturation_arithmetic_uniform_universal', `
-declare <16 x i8> @llvm.x86.sse2.p$1s.b(<16 x i8>, <16 x i8>) nounwind readnone
-define i8 @__p$1s_i8(i8 %a0, i8 %a1) {
-  %a0_i16 = sext i8 %a0 to i16
-  %a1_i16 = sext i8 %a1 to i16
-  %res = $1 i16 %a0_i16, %a1_i16
-  %over_mask = icmp sgt i16 %res, 127
-  %over_res = select i1 %over_mask, i16 127, i16 %res
-  %under_mask = icmp slt i16 %res, -128
-  %ret_i16 = select i1 %under_mask, i16 -128, i16 %over_res
-  %ret = trunc i16 %ret_i16 to i8
-  ret i8 %ret
-}
-
-declare <8 x i16> @llvm.x86.sse2.p$1s.w(<8 x i16>, <8 x i16>) nounwind readnone
-define i16 @__p$1s_i16(i16 %a0, i16 %a1) {
-  %a0_i32 = sext i16 %a0 to i32
-  %a1_i32 = sext i16 %a1 to i32
-  %res = $1 i32 %a0_i32, %a1_i32
-  %over_mask = icmp sgt i32 %res, 32767
-  %over_res = select i1 %over_mask, i32 32767, i32 %res
-  %under_mask = icmp slt i32 %res, -32768
-  %ret_i32 = select i1 %under_mask, i32 -32768, i32 %over_res
-  %ret = trunc i32 %ret_i32 to i16
-  ret i16 %ret
-}
-
-declare <16 x i8> @llvm.x86.sse2.p$1us.b(<16 x i8>, <16 x i8>) nounwind readnone
-define i8 @__p$1us_i8(i8 %a0, i8 %a1) {
-  %a0_i16 = zext i8 %a0 to i16
-  %a1_i16 = zext i8 %a1 to i16
-  %res = $1 i16 %a0_i16, %a1_i16
-  %over_mask = icmp ugt i16 %res, 255
-  %over_res = select i1 %over_mask, i16 255, i16 %res
-  %under_mask = icmp slt i16 %res, 0
-  %ret_i16 = select i1 %under_mask, i16 0, i16 %over_res
-  %ret = trunc i16 %ret_i16 to i8
-  ret i8 %ret
-}
-
-declare <8 x i16> @llvm.x86.sse2.p$1us.w(<8 x i16>, <8 x i16>) nounwind readnone
-define i16 @__p$1us_i16(i16 %a0, i16 %a1) {
-  %a0_i32 = zext i16 %a0 to i32
-  %a1_i32 = zext i16 %a1 to i32
-  %res = $1 i32 %a0_i32, %a1_i32
-  %over_mask = icmp ugt i32 %res, 65535
-  %over_res = select i1 %over_mask, i32 65535, i32 %res
-  %under_mask = icmp slt i32 %res, 0
-  %ret_i32 = select i1 %under_mask, i32 0, i32 %over_res
-  %ret = trunc i32 %ret_i32 to i16
-  ret i16 %ret
-}
-')
-
-;;uniform saturation arithmetic
-
-define(`saturation_arithmetic_uniform', `
-saturation_arithmetic_uniform_universal(sub)
-saturation_arithmetic_uniform_universal(add)
-')
+        WIDTH, `16', `saturation_arithmetic_vec16()')')

 ;; create vector constant. Used by saturation_arithmetic_novec_universal below.

@@ -278,6 +212,7 @@ ifelse(WIDTH,  `4', `<$1 $2, $1 $2, $1 $2, $1 $2>',
 ;; $1: {add,sub} (used in constructing function names)
                        
 define(`saturation_arithmetic_novec_universal', `
+declare <16 x i8> @llvm.x86.sse2.p$1s.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <WIDTH x i8> @__p$1s_vi8(<WIDTH x i8>, <WIDTH x i8>) {
  %v0_i16 = sext <WIDTH x i8> %0 to <WIDTH x i16>
  %v1_i16 = sext <WIDTH x i8> %1 to <WIDTH x i16>
@@ -290,6 +225,7 @@ define <WIDTH x i8> @__p$1s_vi8(<WIDTH x i8>, <WIDTH x i8>) {
  ret <WIDTH x i8> %ret
 }

+declare <8 x i16> @llvm.x86.sse2.p$1s.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <WIDTH x i16> @__p$1s_vi16(<WIDTH x i16>, <WIDTH x i16>) {
  %v0_i32 = sext <WIDTH x i16> %0 to <WIDTH x i32>
  %v1_i32 = sext <WIDTH x i16> %1 to <WIDTH x i32>
@@ -302,6 +238,7 @@ define <WIDTH x i16> @__p$1s_vi16(<WIDTH x i16>, <WIDTH x i16>) {
  ret <WIDTH x i16> %ret
 }

+declare <16 x i8> @llvm.x86.sse2.p$1us.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <WIDTH x i8> @__p$1us_vi8(<WIDTH x i8>, <WIDTH x i8>) {
  %v0_i16 = zext <WIDTH x i8> %0 to <WIDTH x i16>
  %v1_i16 = zext <WIDTH x i8> %1 to <WIDTH x i16>
@@ -313,7 +250,8 @@ define <WIDTH x i8> @__p$1us_vi8(<WIDTH x i8>, <WIDTH x i8>) {
  %ret = trunc <WIDTH x i16> %ret_i16 to <WIDTH x i8>
  ret <WIDTH x i8> %ret
 }
-  
+
+declare <8 x i16> @llvm.x86.sse2.p$1us.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <WIDTH x i16> @__p$1us_vi16(<WIDTH x i16>, <WIDTH x i16>) {
  %v0_i32 = zext <WIDTH x i16> %0 to <WIDTH x i32>
  %v1_i32 = zext <WIDTH x i16> %1 to <WIDTH x i32>
@@ -337,6 +275,7 @@ saturation_arithmetic_novec_universal(add)
 ;;4-wide vector saturation arithmetic

 define(`saturation_arithmetic_vec4', `
+declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <4 x i8> @__padds_vi8(<4 x i8>, <4 x i8>) {
  convert4to16(i8, %0, %v0)
  convert4to16(i8, %1, %v1)    
@@ -345,6 +284,7 @@ define <4 x i8> @__padds_vi8(<4 x i8>, <4 x i8>) {
  ret <4 x i8> %r
 }

+declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <4 x i16> @__padds_vi16(<4 x i16>, <4 x i16>) {
  convert4to8(i16, %0, %v0)
  convert4to8(i16, %1, %v1)
@@ -353,6 +293,7 @@ define <4 x i16> @__padds_vi16(<4 x i16>, <4 x i16>) {
  ret <4 x i16> %r
 }

+declare <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <4 x i8> @__paddus_vi8(<4 x i8>, <4 x i8>) {
  convert4to16(i8, %0, %v0)
  convert4to16(i8, %1, %v1)
@@ -361,6 +302,7 @@ define <4 x i8> @__paddus_vi8(<4 x i8>, <4 x i8>) {
  ret <4 x i8> %r
 }

+declare <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <4 x i16> @__paddus_vi16(<4 x i16>, <4 x i16>) {
  convert4to8(i16, %0, %v0)
  convert4to8(i16, %1, %v1)
@@ -369,6 +311,7 @@ define <4 x i16> @__paddus_vi16(<4 x i16>, <4 x i16>) {
  ret <4 x i16> %r
 }

+declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <4 x i8> @__psubs_vi8(<4 x i8>, <4 x i8>) {
  convert4to16(i8, %0, %v0)
  convert4to16(i8, %1, %v1)
@@ -377,6 +320,7 @@ define <4 x i8> @__psubs_vi8(<4 x i8>, <4 x i8>) {
  ret <4 x i8> %r
 }

+declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <4 x i16> @__psubs_vi16(<4 x i16>, <4 x i16>) {
  convert4to8(i16, %0, %v0)
  convert4to8(i16, %1, %v1)
@@ -385,6 +329,7 @@ define <4 x i16> @__psubs_vi16(<4 x i16>, <4 x i16>) {
  ret <4 x i16> %r
 }

+declare <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <4 x i8> @__psubus_vi8(<4 x i8>, <4 x i8>) {
  convert4to16(i8, %0, %v0)
  convert4to16(i8, %1, %v1)
@@ -393,6 +338,7 @@ define <4 x i8> @__psubus_vi8(<4 x i8>, <4 x i8>) {
  ret <4 x i8> %r
 }

+declare <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <4 x i16> @__psubus_vi16(<4 x i16>, <4 x i16>) {
  convert4to8(i16, %0, %v0)
  convert4to8(i16, %1, %v1)
@@ -405,6 +351,7 @@ define <4 x i16> @__psubus_vi16(<4 x i16>, <4 x i16>) {
 ;;8-wide vector saturation arithmetic

 define(`saturation_arithmetic_vec8', `
+declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <8 x i8> @__padds_vi8(<8 x i8>, <8 x i8>) {
  convert8to16(i8, %0, %v0)
  convert8to16(i8, %1, %v1)
@@ -413,11 +360,13 @@ define <8 x i8> @__padds_vi8(<8 x i8>, <8 x i8>) {
  ret <8 x i8> %r
 }

+declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <8 x i16> @__padds_vi16(<8 x i16> %a0, <8 x i16> %a1) {
  %res = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %res
 }

+declare <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <8 x i8> @__paddus_vi8(<8 x i8>, <8 x i8>) {
  convert8to16(i8, %0, %v0)
  convert8to16(i8, %1, %v1)
@@ -426,11 +375,13 @@ define <8 x i8> @__paddus_vi8(<8 x i8>, <8 x i8>) {
  ret <8 x i8> %r
 }

+declare <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <8 x i16> @__paddus_vi16(<8 x i16> %a0, <8 x i16> %a1) {
  %res = call <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %res
 }

+declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <8 x i8> @__psubs_vi8(<8 x i8>, <8 x i8>) {
  convert8to16(i8, %0, %v0)
  convert8to16(i8, %1, %v1)
@@ -439,11 +390,13 @@ define <8 x i8> @__psubs_vi8(<8 x i8>, <8 x i8>) {
  ret <8 x i8> %r
 }

+declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <8 x i16> @__psubs_vi16(<8 x i16> %a0, <8 x i16> %a1) {
  %res = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %res
 }

+declare <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <8 x i8> @__psubus_vi8(<8 x i8>, <8 x i8>) {
  convert8to16(i8, %0, %v0)
  convert8to16(i8, %1, %v1)
@@ -452,6 +405,7 @@ define <8 x i8> @__psubus_vi8(<8 x i8>, <8 x i8>) {
  ret <8 x i8> %r    
 }

+declare <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <8 x i16> @__psubus_vi16(<8 x i16> %a0, <8 x i16> %a1) {
  %res = call <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16> %a0, <8 x i16> %a1)
  ret <8 x i16> %res
@@ -461,41 +415,49 @@ define <8 x i16> @__psubus_vi16(<8 x i16> %a0, <8 x i16> %a1) {
 ;;16-wide vector saturation arithmetic

 define(`saturation_arithmetic_vec16', `
+declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <16 x i8> @__padds_vi8(<16 x i8> %a0, <16 x i8> %a1) {
  %res = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a0, <16 x i8> %a1) ; <<16 x i8>> [#uses=1]
  ret <16 x i8> %res
 }

+declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <16 x i16> @__padds_vi16(<16 x i16> %a0, <16 x i16> %a1) {
  binary8to16(ret, i16, @llvm.x86.sse2.padds.w, %a0, %a1)
  ret <16 x i16> %ret
 }

+declare <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <16 x i8> @__paddus_vi8(<16 x i8> %a0, <16 x i8> %a1) {
  %res = call <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8> %a0, <16 x i8> %a1) ; <<16 x i8>> [#uses=1]
  ret <16 x i8> %res
 }

+declare <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <16 x i16> @__paddus_vi16(<16 x i16> %a0, <16 x i16> %a1) {
  binary8to16(ret, i16, @llvm.x86.sse2.paddus.w, %a0, %a1)  
  ret <16 x i16> %ret
 }

+declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <16 x i8> @__psubs_vi8(<16 x i8> %a0, <16 x i8> %a1) {
  %res = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a0, <16 x i8> %a1) ; <<16 x i8>> [#uses=1]
  ret <16 x i8> %res
 }

+declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <16 x i16> @__psubs_vi16(<16 x i16> %a0, <16 x i16> %a1) {
  binary8to16(ret, i16, @llvm.x86.sse2.psubs.w, %a0, %a1)
  ret <16 x i16> %ret
 }

+declare <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8>, <16 x i8>) nounwind readnone
 define <16 x i8> @__psubus_vi8(<16 x i8> %a0, <16 x i8> %a1) {
  %res = call <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8> %a0, <16 x i8> %a1) ; <<16 x i8>> [#uses=1]
  ret <16 x i8> %res
 }

+declare <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16>, <8 x i16>) nounwind readnone
 define <16 x i16> @__psubus_vi16(<16 x i16> %a0, <16 x i16> %a1) {
  binary8to16(ret, i16, @llvm.x86.sse2.psubus.w, %a0, %a1)  
  ret <16 x i16> %ret
--- a/stdlib.ispc
+++ b/stdlib.ispc
@@ -57,6 +57,34 @@
  #error Unknown value of ISPC_MASK_BITS
 #endif

+/* Limits of integral types. */
+#ifndef INT8_MIN
+#define INT8_MIN               (-128)
+#endif
+#ifndef INT16_MIN
+#define INT16_MIN              (-32768)
+#endif
+#ifndef INT32_MIN
+#define INT32_MIN              (-2147483648)
+#endif
+#ifndef INT8_MAX
+#define INT8_MAX               (127)
+#endif
+#ifndef INT16_MAX
+#define INT16_MAX              (32767)
+#endif
+#ifndef INT32_MAX
+#define INT32_MAX              (2147483647)
+#endif
+#ifndef UINT8_MAX
+#define UINT8_MAX              (255)
+#endif
+#ifndef UINT16_MAX
+#define UINT16_MAX             (65535)
+#endif
+#ifndef UINT32_MAX
+#define UINT32_MAX             (4294967295)
+#endif
 ///////////////////////////////////////////////////////////////////////////
 // Low level primitives

@@ -4261,7 +4289,12 @@ static inline void fastmath() {
 // saturation arithmetic

 static inline uniform int8 saturating_add(uniform int8 a, uniform int8 b) {
-    return __padds_i8(a, b);
+    uniform unsigned int8 a_unsig = a, b_unsig = b;
+    uniform unsigned int8 result = a_unsig + b_unsig;
+    a_unsig = (a_unsig >> 7) + INT8_MAX;
+    if ((uniform int8) ((a_unsig ^ b_unsig) | ~(b_unsig ^ result)) >= 0)
+        result = a_unsig;
+    return result;
 }

 static inline varying int8 saturating_add(varying int8 a, varying int8 b) {
@@ -4269,7 +4302,12 @@ static inline varying int8 saturating_add(varying int8 a, varying int8 b) {
 }

 static inline uniform int16 saturating_add(uniform int16 a, uniform int16 b) {
-    return __padds_i16(a, b);
+    uniform unsigned int16 a_unsig = a, b_unsig = b;
+    uniform unsigned int16 result = a_unsig + b_unsig;
+    a_unsig = (a_unsig >> 15) + INT16_MAX;
+    if ((uniform int16) ((a_unsig ^ b_unsig) | ~(b_unsig ^ result)) >= 0)
+        result = a_unsig;
+    return result;
 }

 static inline varying int16 saturating_add(varying int16 a, varying int16 b) {
@@ -4278,7 +4316,9 @@ static inline varying int16 saturating_add(varying int16 a, varying int16 b) {

 static inline uniform unsigned int8 saturating_add(uniform unsigned int8 a, 
                                                   uniform unsigned int8 b) {
-    return __paddus_i8(a, b);
+    uniform unsigned int8 result = a + b;
+    result |= (-(uniform int8)(result < a));
+    return result;
 }

 static inline varying unsigned int8 saturating_add(varying unsigned int8 a, 
@@ -4288,7 +4328,9 @@ static inline varying unsigned int8 saturating_add(varying unsigned int8 a,

 static inline uniform unsigned int16 saturating_add(uniform unsigned int16 a, 
                                                    uniform unsigned int16 b) {
-    return __paddus_i16(a, b);
+    uniform unsigned int16 result = a + b;
+    result |= (-(uniform int16)(result < a));
+    return result;
 }

 static inline varying unsigned int16 saturating_add(varying unsigned int16 a, 
@@ -4297,7 +4339,12 @@ static inline varying unsigned int16 saturating_add(varying unsigned int16 a,
 }

 static inline uniform int8 saturating_sub(uniform int8 a, uniform int8 b) {
-    return __psubs_i8(a, b);
+    uniform unsigned int8 a_unsig = a, b_unsig = b;
+    uniform unsigned int8 result = a_unsig - b_unsig;
+    a_unsig = (a_unsig >> 7) + INT8_MAX;
+    if ((uniform int8) ((a_unsig ^ b_unsig) & (a_unsig ^ result)) < 0)
+        result = a_unsig;
+    return result;
 }

 static inline varying int8 saturating_sub(varying int8 a, varying int8 b) {
@@ -4305,7 +4352,12 @@ static inline varying int8 saturating_sub(varying int8 a, varying int8 b) {
 }

 static inline uniform int16 saturating_sub(uniform int16 a, uniform int16 b) {
-    return __psubs_i16(a, b);
+    uniform unsigned int16 a_unsig = a, b_unsig = b;
+    uniform unsigned int16 result = a_unsig - b_unsig;
+    a_unsig = (a_unsig >> 15) + INT16_MAX;
+    if ((uniform int16) ((a_unsig ^ b_unsig) & (a_unsig ^ result)) < 0)
+        result = a_unsig;
+    return result;
 }

 static inline varying int16 saturating_sub(varying int16 a, varying int16 b) {
@@ -4314,7 +4366,9 @@ static inline varying int16 saturating_sub(varying int16 a, varying int16 b) {

 static inline uniform unsigned int8 saturating_sub(uniform unsigned int8 a, 
                                                   uniform unsigned int8 b) {
-    return __psubus_i8(a, b);
+    uniform unsigned int8 result = a - b;
+    result &= (-(uniform int8)(result <= a));
+    return result;
 }

 static inline varying unsigned int8 saturating_sub(varying unsigned int8 a, 
@@ -4324,7 +4378,9 @@ static inline varying unsigned int8 saturating_sub(varying unsigned int8 a,

 static inline uniform unsigned int16 saturating_sub(uniform unsigned int16 a, 
                                                    uniform unsigned int16 b) {
-    return __psubus_i16(a, b);
+    uniform unsigned int16 result = a - b;
+    result &= (-(uniform int16)(result <= a));
+    return result;
 }

 static inline varying unsigned int16 saturating_sub(varying unsigned int16 a,