Fix invalid const atomic builtin accesses.

Clang fixed a signature issue with builtin atomic functions accidentally
allowing const-qualified arguments. This change removes the invalid
const qualifiers from those calls. The error message looks like:
"address argument to atomic builtin cannot be const-qualified"

Bug: http://b/110779387
Test: Builds with new compiler.
Change-Id: Ib0cb00f17ce8383134a89d4dcce5dcf0e5b738bf
diff --git a/cpu_ref/rsCpuCore.cpp b/cpu_ref/rsCpuCore.cpp
index 14b1e66..7039cec 100644
--- a/cpu_ref/rsCpuCore.cpp
+++ b/cpu_ref/rsCpuCore.cpp
@@ -506,7 +506,7 @@
   return *outBuf;
 }
 
-static void reduce_get_accumulator(uint8_t *&accumPtr, const MTLaunchStructReduce *mtls,
+static void reduce_get_accumulator(uint8_t *&accumPtr, MTLaunchStructReduce *mtls,
                                    const char *walkerName, uint32_t threadIdx) {
   rsAssert(!accumPtr);
 
@@ -531,7 +531,7 @@
 }
 
 static void walk_1d_reduce(void *usr, uint32_t idx) {
-  const MTLaunchStructReduce *mtls = (const MTLaunchStructReduce *)usr;
+  MTLaunchStructReduce *mtls = (MTLaunchStructReduce *)usr;
   RsExpandKernelDriverInfo redp = mtls->redp;
 
   // find accumulator
@@ -570,7 +570,7 @@
 }
 
 static void walk_2d_reduce(void *usr, uint32_t idx) {
-  const MTLaunchStructReduce *mtls = (const MTLaunchStructReduce *)usr;
+  MTLaunchStructReduce *mtls = (MTLaunchStructReduce *)usr;
   RsExpandKernelDriverInfo redp = mtls->redp;
 
   // find accumulator
@@ -609,7 +609,7 @@
 }
 
 static void walk_3d_reduce(void *usr, uint32_t idx) {
-  const MTLaunchStructReduce *mtls = (const MTLaunchStructReduce *)usr;
+  MTLaunchStructReduce *mtls = (MTLaunchStructReduce *)usr;
   RsExpandKernelDriverInfo redp = mtls->redp;
 
   // find accumulator
diff --git a/cpu_ref/rsCpuIntrinsicBLAS.cpp b/cpu_ref/rsCpuIntrinsicBLAS.cpp
index d60a3b9..849693d 100644
--- a/cpu_ref/rsCpuIntrinsicBLAS.cpp
+++ b/cpu_ref/rsCpuIntrinsicBLAS.cpp
@@ -142,7 +142,7 @@
 // Generic GEMM callback routine.
 template <typename T_data, typename T_param, typename Func>
 static void walk_tiled_gemm(Func blasFunc, T_param alpha, T_param beta, int vecSize,
-                            RsBlasCall* call, const MTLaunchStructForEachBlas *mtls) {
+                            RsBlasCall* call, MTLaunchStructForEachBlas *mtls) {
     // setup BLAS enum args
     enum CBLAS_TRANSPOSE TransA = (enum CBLAS_TRANSPOSE)call->transA;
     enum CBLAS_TRANSPOSE TransB = (enum CBLAS_TRANSPOSE)call->transB;
@@ -190,7 +190,7 @@
 
 // SGEMM callback
 static void walk_2d_sgemm(void *usr, uint32_t idx) {
-    const MTLaunchStructForEachBlas *mtls = (const MTLaunchStructForEachBlas *)usr;
+    MTLaunchStructForEachBlas *mtls = (MTLaunchStructForEachBlas *)usr;
     RsBlasCall* call = (RsBlasCall*) mtls->sc;
 
     float alpha = call->alpha.f;
@@ -201,7 +201,7 @@
 
 // DGEMM callback
 static void walk_2d_dgemm(void *usr, uint32_t idx) {
-    const MTLaunchStructForEachBlas *mtls = (const MTLaunchStructForEachBlas *)usr;
+    MTLaunchStructForEachBlas *mtls = (MTLaunchStructForEachBlas *)usr;
     RsBlasCall* call = (RsBlasCall*) mtls->sc;
 
     double alpha = call->alpha.d;
@@ -212,7 +212,7 @@
 
 // CGEMM callback
 static void walk_2d_cgemm(void *usr, uint32_t idx) {
-    const MTLaunchStructForEachBlas *mtls = (const MTLaunchStructForEachBlas *)usr;
+    MTLaunchStructForEachBlas *mtls = (MTLaunchStructForEachBlas *)usr;
     RsBlasCall* call = (RsBlasCall*) mtls->sc;
 
     void * alpha = (void *)&call->alpha.c;
@@ -223,7 +223,7 @@
 
 // ZGEMM callback
 static void walk_2d_zgemm(void *usr, uint32_t idx) {
-    const MTLaunchStructForEachBlas *mtls = (const MTLaunchStructForEachBlas *)usr;
+    MTLaunchStructForEachBlas *mtls = (MTLaunchStructForEachBlas *)usr;
     RsBlasCall* call = (RsBlasCall*) mtls->sc;
 
     void * alpha = (void *)&call->alpha.z;