Dynamic tensor descriptor (#24)

* support dynamic tensor descriptor

* use buffer load OOB feature for padding case

* add navi support

* add int8x4 inference kernel
Co-authored-by: Chao Liu <chao@ixt-rack-81.local.lan>
Co-authored-by: Jing Zhang <jizhan@amd.com>

driver/src/conv_driver.cu

-conv_driver.cpp
\ No newline at end of file

external/half/include/half.hpp

@@ -508,8 +508,8 @@ template <bool B>
 struct bool_type : std::integral_constant<bool, B>
 {
 };
-using std::true_type;
 using std::false_type;
+using std::true_type;
 
 /// Type traits for floating-point types.
 template <typename T>
@@ -854,8 +854,8 @@ inline HALF_CONSTEXPR_NOERR unsigned int signal(unsigned int x, unsigned int y,
           ((x & 0x7FFF) > 0x7C00 && !(x & 0x200)) || ((y & 0x7FFF) > 0x7C00 && !(y & 0x200)) ||
               ((z & 0x7FFF) > 0x7C00 && !(z & 0x200)));
 #endif
-    return ((x & 0x7FFF) > 0x7C00) ? (x | 0x200) : ((y & 0x7FFF) > 0x7C00) ? (y | 0x200)
-                                                                           : (z | 0x200);
+    return ((x & 0x7FFF) > 0x7C00) ? (x | 0x200)
+                                   : ((y & 0x7FFF) > 0x7C00) ? (y | 0x200) : (z | 0x200);
 }
 
 /// Select value or signaling NaN.
@@ -1756,9 +1756,9 @@ uint32 mulhi(uint32 x, uint32 y)
     uint32 xy = (x >> 16) * (y & 0xFFFF), yx = (x & 0xFFFF) * (y >> 16),
            c = (xy & 0xFFFF) + (yx & 0xFFFF) + (((x & 0xFFFF) * (y & 0xFFFF)) >> 16);
     return (x >> 16) * (y >> 16) + (xy >> 16) + (yx >> 16) + (c >> 16) +
-           ((R == std::round_to_nearest) ? ((c >> 15) & 1) : (R == std::round_toward_infinity)
-                                                                 ? ((c & 0xFFFF) != 0)
-                                                                 : 0);
+           ((R == std::round_to_nearest)
+                ? ((c >> 15) & 1)
+                : (R == std::round_toward_infinity) ? ((c & 0xFFFF) != 0) : 0);
 }
 
 /// 64-bit multiplication.
@@ -2379,10 +2379,12 @@ unsigned int erf(unsigned int arg)
             t /
             ((x2.exp < 0) ? f31(exp2((x2.exp > -32) ? (x2.m >> -x2.exp) : 0, 30), 0)
                           : f31(exp2((x2.m << x2.exp) & 0x7FFFFFFF, 22), x2.m >> (31 - x2.exp)));
-    return (!C || sign) ? fixed2half<R, 31, false, true, true>(
+    return (!C || sign)
+               ? fixed2half<R, 31, false, true, true>(
                      0x80000000 - (e.m >> (C - e.exp)), 14 + C, sign & (C - 1U))
-                        : (e.exp < -25) ? underflow<R>() : fixed2half<R, 30, false, false, true>(
-                                                               e.m >> 1, e.exp + 14, 0, e.m & 1);
+               : (e.exp < -25)
+                     ? underflow<R>()
+                     : fixed2half<R, 30, false, false, true>(e.m >> 1, e.exp + 14, 0, e.m & 1);
 }
 
 /// Gamma function and postprocessing.
@@ -2402,8 +2404,7 @@ unsigned int gamma(unsigned int arg)
 			for(unsigned int i=0; i<5; ++i)
 				s += p[i+1] / (arg+i);
 			return std::log(s) + (arg-0.5)*std::log(t) - t;
-*/ static const f31
-        pi(0xC90FDAA2, 1),
+*/ static const f31 pi(0xC90FDAA2, 1),
         lbe(0xB8AA3B29, 0);
     unsigned int abs = arg & 0x7FFF, sign = arg & 0x8000;
     bool bsign = sign != 0;
@@ -2789,7 +2790,7 @@ inline half operator"" _h(long double value)
 {
     return half(detail::binary, detail::float2half<half::round_style>(value));
 }
-}
+} // namespace literal
 #endif
 
 namespace detail {
@@ -2837,8 +2838,8 @@ struct half_caster<half, half, R>
 {
     static half cast(half arg) { return arg; }
 };
-}
-}
+} // namespace detail
+} // namespace half_float
 
 /// Extensions to the C++ standard library.
 namespace std {
@@ -3003,7 +3004,7 @@ struct hash<half_float::half>
     }
 };
 #endif
-}
+} // namespace std
 
 namespace half_float {
 /// \anchor compop
@@ -3122,10 +3123,11 @@ inline half operator+(half x, half y)
         return half(detail::binary,
                     (absx > 0x7C00 || absy > 0x7C00)
                         ? detail::signal(x.data_, y.data_)
-                        : (absy != 0x7C00) ? x.data_ : (sub && absx == 0x7C00) ? detail::invalid()
-                                                                               : y.data_);
+                        : (absy != 0x7C00) ? x.data_
+                                           : (sub && absx == 0x7C00) ? detail::invalid() : y.data_);
     if(!absx)
-        return absy ? y : half(detail::binary,
+        return absy ? y
+                    : half(detail::binary,
                            (half::round_style == std::round_toward_neg_infinity)
                                ? (x.data_ | y.data_)
                                : (x.data_ & y.data_));
@@ -3449,9 +3451,10 @@ inline half fma(half x, half y, half z)
                                                                     : (sign | 0x7C00))
                                                          : z;
     if(!absx || !absy)
-        return absz ? z : half(detail::binary,
-                               (half::round_style == std::round_toward_neg_infinity)
-                                   ? (z.data_ | sign)
+        return absz
+                   ? z
+                   : half(detail::binary,
+                          (half::round_style == std::round_toward_neg_infinity) ? (z.data_ | sign)
                                                                                 : (z.data_ & sign));
     for(; absx < 0x400; absx <<= 1, --exp)
         ;
@@ -3516,8 +3519,7 @@ inline half fma(half x, half y, half z)
 inline HALF_CONSTEXPR_NOERR half fmax(half x, half y)
 {
     return half(detail::binary,
-                (!isnan(y) && (isnan(x) ||
-                               (x.data_ ^ (0x8000 | (0x8000 - (x.data_ >> 15)))) <
+                (!isnan(y) && (isnan(x) || (x.data_ ^ (0x8000 | (0x8000 - (x.data_ >> 15)))) <
                                                (y.data_ ^ (0x8000 | (0x8000 - (y.data_ >> 15))))))
                     ? detail::select(y.data_, x.data_)
                     : detail::select(x.data_, y.data_));
@@ -3533,8 +3535,7 @@ inline HALF_CONSTEXPR_NOERR half fmax(half x, half y)
 inline HALF_CONSTEXPR_NOERR half fmin(half x, half y)
 {
     return half(detail::binary,
-                (!isnan(y) && (isnan(x) ||
-                               (x.data_ ^ (0x8000 | (0x8000 - (x.data_ >> 15)))) >
+                (!isnan(y) && (isnan(x) || (x.data_ ^ (0x8000 | (0x8000 - (x.data_ >> 15)))) >
                                                (y.data_ ^ (0x8000 | (0x8000 - (y.data_ >> 15))))))
                     ? detail::select(y.data_, x.data_)
                     : detail::select(x.data_, y.data_));
@@ -3886,9 +3887,9 @@ inline half log1p(half arg)
 #else
     if(arg.data_ >= 0xBC00)
         return half(detail::binary,
-                    (arg.data_ == 0xBC00) ? detail::pole(0x8000) : (arg.data_ <= 0xFC00)
-                                                                       ? detail::invalid()
-                                                                       : detail::signal(arg.data_));
+                    (arg.data_ == 0xBC00)
+                        ? detail::pole(0x8000)
+                        : (arg.data_ <= 0xFC00) ? detail::invalid() : detail::signal(arg.data_));
     int abs = arg.data_ & 0x7FFF, exp = -15;
     if(!abs || abs >= 0x7C00)
         return (abs > 0x7C00) ? half(detail::binary, detail::signal(arg.data_)) : arg;
@@ -5354,13 +5355,13 @@ inline HALF_CONSTEXPR half copysign(half x, half y)
 /// \retval FP_NORMAL for all other (normal) values
 inline HALF_CONSTEXPR int fpclassify(half arg)
 {
-    return !(arg.data_ & 0x7FFF) ? FP_ZERO : ((arg.data_ & 0x7FFF) < 0x400)
+    return !(arg.data_ & 0x7FFF)
+               ? FP_ZERO
+               : ((arg.data_ & 0x7FFF) < 0x400)
                      ? FP_SUBNORMAL
                      : ((arg.data_ & 0x7FFF) < 0x7C00)
                            ? FP_NORMAL
-                                                       : ((arg.data_ & 0x7FFF) == 0x7C00)
-                                                             ? FP_INFINITE
-                                                             : FP_NAN;
+                           : ((arg.data_ & 0x7FFF) == 0x7C00) ? FP_INFINITE : FP_NAN;
 }
 
 /// Check if finite number.
@@ -5652,7 +5653,7 @@ inline void fethrowexcept(int excepts, const char* msg = "")
         throw std::range_error(msg);
 }
 /// \}
-}
+} // namespace half_float
 
 #undef HALF_UNUSED_NOERR
 #undef HALF_CONSTEXPR

script/cmake-rocm3.5.sh → script/cmake-rocm3.7.sh

@@ -3,19 +3,23 @@ rm -f CMakeCache.txt
 rm -f *.cmake
 rm -rf CMakeFiles
 
-MY_PROJECT_SOURCE=../../../
+MY_PROJECT_SOURCE=../
 MY_PROJECT_INSTALL=../install.dir
 
 cmake                                                                                                                              \
 -D CMAKE_INSTALL_PREFIX=${MY_PROJECT_INSTALL}                                                                                      \
 -D CMAKE_BUILD_TYPE=Release                                                                                                        \
 -D DEVICE_BACKEND="AMD"                                                                                                            \
--D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-enable-global-sgpr-addr -mllvm --amdgpu-spill-vgpr-to-agpr=0"       \
+-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx1030 -gline-tables-only -save-temps=$CWD -ftemplate-backtrace-limit=0"                   \
 -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                                                          \
 -D CMAKE_PREFIX_PATH="/opt/rocm"                                                                                                   \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                                                  \
 ${MY_PROJECT_SOURCE}
 
+#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -save-temps=$CWD"                                                                   \
+#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -gline-tables-only -save-temps=$CWD"                                                \
+#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-spill-vgpr-to-agpr=0"                                               \
+#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-spill-vgpr-to-agpr=0 -save-temps=$CWD"                              \
 #-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-enable-global-sgpr-addr -mllvm --amdgpu-spill-vgpr-to-agpr=0"       \
-#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-enable-global-sgpr-addr -mllvm --amdgpu-spill-vgpr-to-agpr=0 -save-temps"       \
-#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-enable-global-sgpr-addr -mllvm --amdgpu-spill-vgpr-to-agpr=0 -v -gline-tables-only -save-temps"       \
+#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-enable-global-sgpr-addr -mllvm --amdgpu-spill-vgpr-to-agpr=0 -save-temps=$CWD"       \
+#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-enable-global-sgpr-addr -mllvm --amdgpu-spill-vgpr-to-agpr=0 -v -gline-tables-only -save-temps=$CWD"       \

script/count_vgpr.sh

+#!/bin/bash
+FILE=$1
+
+echo	v0	$(	grep -w	v0	$FILE	| wc -l	)
+echo	v1	$(	grep -w	v1	$FILE	| wc -l	)
+echo	v2	$(	grep -w	v2	$FILE	| wc -l	)
+echo	v3	$(	grep -w	v3	$FILE	| wc -l	)
+echo	v4	$(	grep -w	v4	$FILE	| wc -l	)
+echo	v5	$(	grep -w	v5	$FILE	| wc -l	)
+echo	v6	$(	grep -w	v6	$FILE	| wc -l	)
+echo	v7	$(	grep -w	v7	$FILE	| wc -l	)
+echo	v8	$(	grep -w	v8	$FILE	| wc -l	)
+echo	v9	$(	grep -w	v9	$FILE	| wc -l	)
+echo	v10	$(	grep -w	v10	$FILE	| wc -l	)
+echo	v11	$(	grep -w	v11	$FILE	| wc -l	)
+echo	v12	$(	grep -w	v12	$FILE	| wc -l	)
+echo	v13	$(	grep -w	v13	$FILE	| wc -l	)
+echo	v14	$(	grep -w	v14	$FILE	| wc -l	)
+echo	v15	$(	grep -w	v15	$FILE	| wc -l	)
+echo	v16	$(	grep -w	v16	$FILE	| wc -l	)
+echo	v17	$(	grep -w	v17	$FILE	| wc -l	)
+echo	v18	$(	grep -w	v18	$FILE	| wc -l	)
+echo	v19	$(	grep -w	v19	$FILE	| wc -l	)
+echo	v20	$(	grep -w	v20	$FILE	| wc -l	)
+echo	v21	$(	grep -w	v21	$FILE	| wc -l	)
+echo	v22	$(	grep -w	v22	$FILE	| wc -l	)
+echo	v23	$(	grep -w	v23	$FILE	| wc -l	)
+echo	v24	$(	grep -w	v24	$FILE	| wc -l	)
+echo	v25	$(	grep -w	v25	$FILE	| wc -l	)
+echo	v26	$(	grep -w	v26	$FILE	| wc -l	)
+echo	v27	$(	grep -w	v27	$FILE	| wc -l	)
+echo	v28	$(	grep -w	v28	$FILE	| wc -l	)
+echo	v29	$(	grep -w	v29	$FILE	| wc -l	)
+echo	v30	$(	grep -w	v30	$FILE	| wc -l	)
+echo	v31	$(	grep -w	v31	$FILE	| wc -l	)
+echo	v32	$(	grep -w	v32	$FILE	| wc -l	)
+echo	v33	$(	grep -w	v33	$FILE	| wc -l	)
+echo	v34	$(	grep -w	v34	$FILE	| wc -l	)
+echo	v35	$(	grep -w	v35	$FILE	| wc -l	)
+echo	v36	$(	grep -w	v36	$FILE	| wc -l	)
+echo	v37	$(	grep -w	v37	$FILE	| wc -l	)
+echo	v38	$(	grep -w	v38	$FILE	| wc -l	)
+echo	v39	$(	grep -w	v39	$FILE	| wc -l	)
+echo	v40	$(	grep -w	v40	$FILE	| wc -l	)
+echo	v41	$(	grep -w	v41	$FILE	| wc -l	)
+echo	v42	$(	grep -w	v42	$FILE	| wc -l	)
+echo	v43	$(	grep -w	v43	$FILE	| wc -l	)
+echo	v44	$(	grep -w	v44	$FILE	| wc -l	)
+echo	v45	$(	grep -w	v45	$FILE	| wc -l	)
+echo	v46	$(	grep -w	v46	$FILE	| wc -l	)
+echo	v47	$(	grep -w	v47	$FILE	| wc -l	)
+echo	v48	$(	grep -w	v48	$FILE	| wc -l	)
+echo	v49	$(	grep -w	v49	$FILE	| wc -l	)
+echo	v50	$(	grep -w	v50	$FILE	| wc -l	)
+echo	v51	$(	grep -w	v51	$FILE	| wc -l	)
+echo	v52	$(	grep -w	v52	$FILE	| wc -l	)
+echo	v53	$(	grep -w	v53	$FILE	| wc -l	)
+echo	v54	$(	grep -w	v54	$FILE	| wc -l	)
+echo	v55	$(	grep -w	v55	$FILE	| wc -l	)
+echo	v56	$(	grep -w	v56	$FILE	| wc -l	)
+echo	v57	$(	grep -w	v57	$FILE	| wc -l	)
+echo	v58	$(	grep -w	v58	$FILE	| wc -l	)
+echo	v59	$(	grep -w	v59	$FILE	| wc -l	)
+echo	v60	$(	grep -w	v60	$FILE	| wc -l	)
+echo	v61	$(	grep -w	v61	$FILE	| wc -l	)
+echo	v62	$(	grep -w	v62	$FILE	| wc -l	)
+echo	v63	$(	grep -w	v63	$FILE	| wc -l	)
+echo	v64	$(	grep -w	v64	$FILE	| wc -l	)
+echo	v65	$(	grep -w	v65	$FILE	| wc -l	)
+echo	v66	$(	grep -w	v66	$FILE	| wc -l	)
+echo	v67	$(	grep -w	v67	$FILE	| wc -l	)
+echo	v68	$(	grep -w	v68	$FILE	| wc -l	)
+echo	v69	$(	grep -w	v69	$FILE	| wc -l	)
+echo	v70	$(	grep -w	v70	$FILE	| wc -l	)
+echo	v71	$(	grep -w	v71	$FILE	| wc -l	)
+echo	v72	$(	grep -w	v72	$FILE	| wc -l	)
+echo	v73	$(	grep -w	v73	$FILE	| wc -l	)
+echo	v74	$(	grep -w	v74	$FILE	| wc -l	)
+echo	v75	$(	grep -w	v75	$FILE	| wc -l	)
+echo	v76	$(	grep -w	v76	$FILE	| wc -l	)
+echo	v77	$(	grep -w	v77	$FILE	| wc -l	)
+echo	v78	$(	grep -w	v78	$FILE	| wc -l	)
+echo	v79	$(	grep -w	v79	$FILE	| wc -l	)
+echo	v80	$(	grep -w	v80	$FILE	| wc -l	)
+echo	v81	$(	grep -w	v81	$FILE	| wc -l	)
+echo	v82	$(	grep -w	v82	$FILE	| wc -l	)
+echo	v83	$(	grep -w	v83	$FILE	| wc -l	)
+echo	v84	$(	grep -w	v84	$FILE	| wc -l	)
+echo	v85	$(	grep -w	v85	$FILE	| wc -l	)
+echo	v86	$(	grep -w	v86	$FILE	| wc -l	)
+echo	v87	$(	grep -w	v87	$FILE	| wc -l	)
+echo	v88	$(	grep -w	v88	$FILE	| wc -l	)
+echo	v89	$(	grep -w	v89	$FILE	| wc -l	)
+echo	v90	$(	grep -w	v90	$FILE	| wc -l	)
+echo	v91	$(	grep -w	v91	$FILE	| wc -l	)
+echo	v92	$(	grep -w	v92	$FILE	| wc -l	)
+echo	v93	$(	grep -w	v93	$FILE	| wc -l	)
+echo	v94	$(	grep -w	v94	$FILE	| wc -l	)
+echo	v95	$(	grep -w	v95	$FILE	| wc -l	)
+echo	v96	$(	grep -w	v96	$FILE	| wc -l	)
+echo	v97	$(	grep -w	v97	$FILE	| wc -l	)
+echo	v98	$(	grep -w	v98	$FILE	| wc -l	)
+echo	v99	$(	grep -w	v99	$FILE	| wc -l	)
+echo	v100	$(	grep -w	v100	$FILE	| wc -l	)
+echo	v101	$(	grep -w	v101	$FILE	| wc -l	)
+echo	v102	$(	grep -w	v102	$FILE	| wc -l	)
+echo	v103	$(	grep -w	v103	$FILE	| wc -l	)
+echo	v104	$(	grep -w	v104	$FILE	| wc -l	)
+echo	v105	$(	grep -w	v105	$FILE	| wc -l	)
+echo	v106	$(	grep -w	v106	$FILE	| wc -l	)
+echo	v107	$(	grep -w	v107	$FILE	| wc -l	)
+echo	v108	$(	grep -w	v108	$FILE	| wc -l	)
+echo	v109	$(	grep -w	v109	$FILE	| wc -l	)
+echo	v110	$(	grep -w	v110	$FILE	| wc -l	)
+echo	v111	$(	grep -w	v111	$FILE	| wc -l	)
+echo	v112	$(	grep -w	v112	$FILE	| wc -l	)
+echo	v113	$(	grep -w	v113	$FILE	| wc -l	)
+echo	v114	$(	grep -w	v114	$FILE	| wc -l	)
+echo	v115	$(	grep -w	v115	$FILE	| wc -l	)
+echo	v116	$(	grep -w	v116	$FILE	| wc -l	)
+echo	v117	$(	grep -w	v117	$FILE	| wc -l	)
+echo	v118	$(	grep -w	v118	$FILE	| wc -l	)
+echo	v119	$(	grep -w	v119	$FILE	| wc -l	)
+echo	v120	$(	grep -w	v120	$FILE	| wc -l	)
+echo	v121	$(	grep -w	v121	$FILE	| wc -l	)
+echo	v122	$(	grep -w	v122	$FILE	| wc -l	)
+echo	v123	$(	grep -w	v123	$FILE	| wc -l	)
+echo	v124	$(	grep -w	v124	$FILE	| wc -l	)
+echo	v125	$(	grep -w	v125	$FILE	| wc -l	)
+echo	v126	$(	grep -w	v126	$FILE	| wc -l	)
+echo	v127	$(	grep -w	v127	$FILE	| wc -l	)
+echo	v128	$(	grep -w	v128	$FILE	| wc -l	)
+echo	v129	$(	grep -w	v129	$FILE	| wc -l	)
+echo	v130	$(	grep -w	v130	$FILE	| wc -l	)
+echo	v131	$(	grep -w	v131	$FILE	| wc -l	)
+echo	v132	$(	grep -w	v132	$FILE	| wc -l	)
+echo	v133	$(	grep -w	v133	$FILE	| wc -l	)
+echo	v134	$(	grep -w	v134	$FILE	| wc -l	)
+echo	v135	$(	grep -w	v135	$FILE	| wc -l	)
+echo	v136	$(	grep -w	v136	$FILE	| wc -l	)
+echo	v137	$(	grep -w	v137	$FILE	| wc -l	)
+echo	v138	$(	grep -w	v138	$FILE	| wc -l	)
+echo	v139	$(	grep -w	v139	$FILE	| wc -l	)
+echo	v140	$(	grep -w	v140	$FILE	| wc -l	)
+echo	v141	$(	grep -w	v141	$FILE	| wc -l	)
+echo	v142	$(	grep -w	v142	$FILE	| wc -l	)
+echo	v143	$(	grep -w	v143	$FILE	| wc -l	)
+echo	v144	$(	grep -w	v144	$FILE	| wc -l	)
+echo	v145	$(	grep -w	v145	$FILE	| wc -l	)
+echo	v146	$(	grep -w	v146	$FILE	| wc -l	)
+echo	v147	$(	grep -w	v147	$FILE	| wc -l	)
+echo	v148	$(	grep -w	v148	$FILE	| wc -l	)
+echo	v149	$(	grep -w	v149	$FILE	| wc -l	)
+echo	v150	$(	grep -w	v150	$FILE	| wc -l	)
+echo	v151	$(	grep -w	v151	$FILE	| wc -l	)
+echo	v152	$(	grep -w	v152	$FILE	| wc -l	)
+echo	v153	$(	grep -w	v153	$FILE	| wc -l	)
+echo	v154	$(	grep -w	v154	$FILE	| wc -l	)
+echo	v155	$(	grep -w	v155	$FILE	| wc -l	)
+echo	v156	$(	grep -w	v156	$FILE	| wc -l	)
+echo	v157	$(	grep -w	v157	$FILE	| wc -l	)
+echo	v158	$(	grep -w	v158	$FILE	| wc -l	)
+echo	v159	$(	grep -w	v159	$FILE	| wc -l	)
+echo	v160	$(	grep -w	v160	$FILE	| wc -l	)
+echo	v161	$(	grep -w	v161	$FILE	| wc -l	)
+echo	v162	$(	grep -w	v162	$FILE	| wc -l	)
+echo	v163	$(	grep -w	v163	$FILE	| wc -l	)
+echo	v164	$(	grep -w	v164	$FILE	| wc -l	)
+echo	v165	$(	grep -w	v165	$FILE	| wc -l	)
+echo	v166	$(	grep -w	v166	$FILE	| wc -l	)
+echo	v167	$(	grep -w	v167	$FILE	| wc -l	)
+echo	v168	$(	grep -w	v168	$FILE	| wc -l	)
+echo	v169	$(	grep -w	v169	$FILE	| wc -l	)
+echo	v170	$(	grep -w	v170	$FILE	| wc -l	)
+echo	v171	$(	grep -w	v171	$FILE	| wc -l	)
+echo	v172	$(	grep -w	v172	$FILE	| wc -l	)
+echo	v173	$(	grep -w	v173	$FILE	| wc -l	)
+echo	v174	$(	grep -w	v174	$FILE	| wc -l	)
+echo	v175	$(	grep -w	v175	$FILE	| wc -l	)
+echo	v176	$(	grep -w	v176	$FILE	| wc -l	)
+echo	v177	$(	grep -w	v177	$FILE	| wc -l	)
+echo	v178	$(	grep -w	v178	$FILE	| wc -l	)
+echo	v179	$(	grep -w	v179	$FILE	| wc -l	)
+echo	v180	$(	grep -w	v180	$FILE	| wc -l	)
+echo	v181	$(	grep -w	v181	$FILE	| wc -l	)
+echo	v182	$(	grep -w	v182	$FILE	| wc -l	)
+echo	v183	$(	grep -w	v183	$FILE	| wc -l	)
+echo	v184	$(	grep -w	v184	$FILE	| wc -l	)
+echo	v185	$(	grep -w	v185	$FILE	| wc -l	)
+echo	v186	$(	grep -w	v186	$FILE	| wc -l	)
+echo	v187	$(	grep -w	v187	$FILE	| wc -l	)
+echo	v188	$(	grep -w	v188	$FILE	| wc -l	)
+echo	v189	$(	grep -w	v189	$FILE	| wc -l	)
+echo	v190	$(	grep -w	v190	$FILE	| wc -l	)
+echo	v191	$(	grep -w	v191	$FILE	| wc -l	)
+echo	v192	$(	grep -w	v192	$FILE	| wc -l	)
+echo	v193	$(	grep -w	v193	$FILE	| wc -l	)
+echo	v194	$(	grep -w	v194	$FILE	| wc -l	)
+echo	v195	$(	grep -w	v195	$FILE	| wc -l	)
+echo	v196	$(	grep -w	v196	$FILE	| wc -l	)
+echo	v197	$(	grep -w	v197	$FILE	| wc -l	)
+echo	v198	$(	grep -w	v198	$FILE	| wc -l	)
+echo	v199	$(	grep -w	v199	$FILE	| wc -l	)
+echo	v200	$(	grep -w	v200	$FILE	| wc -l	)
+echo	v201	$(	grep -w	v201	$FILE	| wc -l	)
+echo	v202	$(	grep -w	v202	$FILE	| wc -l	)
+echo	v203	$(	grep -w	v203	$FILE	| wc -l	)
+echo	v204	$(	grep -w	v204	$FILE	| wc -l	)
+echo	v205	$(	grep -w	v205	$FILE	| wc -l	)
+echo	v206	$(	grep -w	v206	$FILE	| wc -l	)
+echo	v207	$(	grep -w	v207	$FILE	| wc -l	)
+echo	v208	$(	grep -w	v208	$FILE	| wc -l	)
+echo	v209	$(	grep -w	v209	$FILE	| wc -l	)
+echo	v210	$(	grep -w	v210	$FILE	| wc -l	)
+echo	v211	$(	grep -w	v211	$FILE	| wc -l	)
+echo	v212	$(	grep -w	v212	$FILE	| wc -l	)
+echo	v213	$(	grep -w	v213	$FILE	| wc -l	)
+echo	v214	$(	grep -w	v214	$FILE	| wc -l	)
+echo	v215	$(	grep -w	v215	$FILE	| wc -l	)
+echo	v216	$(	grep -w	v216	$FILE	| wc -l	)
+echo	v217	$(	grep -w	v217	$FILE	| wc -l	)
+echo	v218	$(	grep -w	v218	$FILE	| wc -l	)
+echo	v219	$(	grep -w	v219	$FILE	| wc -l	)
+echo	v220	$(	grep -w	v220	$FILE	| wc -l	)
+echo	v221	$(	grep -w	v221	$FILE	| wc -l	)
+echo	v222	$(	grep -w	v222	$FILE	| wc -l	)
+echo	v223	$(	grep -w	v223	$FILE	| wc -l	)
+echo	v224	$(	grep -w	v224	$FILE	| wc -l	)
+echo	v225	$(	grep -w	v225	$FILE	| wc -l	)
+echo	v226	$(	grep -w	v226	$FILE	| wc -l	)
+echo	v227	$(	grep -w	v227	$FILE	| wc -l	)
+echo	v228	$(	grep -w	v228	$FILE	| wc -l	)
+echo	v229	$(	grep -w	v229	$FILE	| wc -l	)
+echo	v230	$(	grep -w	v230	$FILE	| wc -l	)
+echo	v231	$(	grep -w	v231	$FILE	| wc -l	)
+echo	v232	$(	grep -w	v232	$FILE	| wc -l	)
+echo	v233	$(	grep -w	v233	$FILE	| wc -l	)
+echo	v234	$(	grep -w	v234	$FILE	| wc -l	)
+echo	v235	$(	grep -w	v235	$FILE	| wc -l	)
+echo	v236	$(	grep -w	v236	$FILE	| wc -l	)
+echo	v237	$(	grep -w	v237	$FILE	| wc -l	)
+echo	v238	$(	grep -w	v238	$FILE	| wc -l	)
+echo	v239	$(	grep -w	v239	$FILE	| wc -l	)
+echo	v240	$(	grep -w	v240	$FILE	| wc -l	)
+echo	v241	$(	grep -w	v241	$FILE	| wc -l	)
+echo	v242	$(	grep -w	v242	$FILE	| wc -l	)
+echo	v243	$(	grep -w	v243	$FILE	| wc -l	)
+echo	v244	$(	grep -w	v244	$FILE	| wc -l	)
+echo	v245	$(	grep -w	v245	$FILE	| wc -l	)
+echo	v246	$(	grep -w	v246	$FILE	| wc -l	)
+echo	v247	$(	grep -w	v247	$FILE	| wc -l	)
+echo	v248	$(	grep -w	v248	$FILE	| wc -l	)
+echo	v249	$(	grep -w	v249	$FILE	| wc -l	)
+echo	v250	$(	grep -w	v250	$FILE	| wc -l	)
+echo	v251	$(	grep -w	v251	$FILE	| wc -l	)
+echo	v252	$(	grep -w	v252	$FILE	| wc -l	)
+echo	v253	$(	grep -w	v253	$FILE	| wc -l	)
+echo	v254	$(	grep -w	v254	$FILE	| wc -l	)
+echo	v255	$(	grep -w	v255	$FILE	| wc -l	)

script/hipclang_opt.sh

+rm *.ll *.s
+
+BC_FILE=$1
+
+/opt/rocm/llvm/bin/llvm-dis $BC_FILE -o original.ll
+/opt/rocm/llvm/bin/opt -S -inline -inline-threshold=104857 original.ll > inline.ll
+/opt/rocm/llvm/bin/opt -S -sroa inline.ll > sroa.ll
+/opt/rocm/llvm/bin/opt -S -O3 sroa.ll > o3.ll
+
+/opt/rocm/llvm/bin/llc -mcpu=gfx906 original.ll
+/opt/rocm/llvm/bin/llc -mcpu=gfx906 inline.ll
+/opt/rocm/llvm/bin/llc -mcpu=gfx906 sroa.ll
+/opt/rocm/llvm/bin/llc -mcpu=gfx906 o3.ll
+
+#/opt/rocm/llvm/bin/opt -S -O3 -sroa inline.ll > o3.ll
+#/opt/rocm/llvm/bin/opt -S -O3 -sroa o3.ll > o3_2.ll
+#/opt/rocm/llvm/bin/opt -S -O3 -sroa o3_2.ll > o3_3.ll
+#/opt/rocm/llvm/bin/opt -S -O3 -sroa o3_3.ll > o3_4.ll
+
+#/opt/rocm/llvm/bin/llc -mcpu=gfx908 opt.ll
+#/opt/rocm/llvm/bin/llc -mcpu=gfx908 inline.ll
+#/opt/rocm/llvm/bin/llc -mcpu=gfx908 o3.ll
+#/opt/rocm/llvm/bin/llc -mcpu=gfx908 o3_2.ll
+#/opt/rocm/llvm/bin/llc -mcpu=gfx908 o3_3.ll
+#/opt/rocm/llvm/bin/llc -mcpu=gfx908 o3_4.ll