From 93fbbfc8adab9a2fba148eddf311b80d7b07ad30 Mon Sep 17 00:00:00 2001 From: "Yaxun (Sam) Liu" Date: Sat, 2 Jun 2018 18:01:09 -0400 Subject: [PATCH 01/13] Add more function declarations for hip-clang --- hipamd/include/hip/hcc_detail/hip_runtime.h | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/hipamd/include/hip/hcc_detail/hip_runtime.h b/hipamd/include/hip/hcc_detail/hip_runtime.h index 1a6b0f7dda..7c1f2a4d16 100644 --- a/hipamd/include/hip/hcc_detail/hip_runtime.h +++ b/hipamd/include/hip/hcc_detail/hip_runtime.h @@ -98,6 +98,9 @@ struct Empty_launch_parm {}; #include "grid_launch_GGL.hpp" #endif // GENERIC_GRID_LAUNCH +#endif // HCC + +#if __HCC_OR_HIP_CLANG__ extern int HIP_TRACE_API; #ifdef __cplusplus @@ -106,15 +109,16 @@ extern int HIP_TRACE_API; #include #include #include +#if __HCC___ #include #include +#endif // __HCC__ // TODO-HCC remove old definitions ; ~1602 hcc supports __HCC_ACCELERATOR__ define. #if defined(__KALMAR_ACCELERATOR__) && !defined(__HCC_ACCELERATOR__) #define __HCC_ACCELERATOR__ __KALMAR_ACCELERATOR__ #endif - // TODO-HCC add a dummy implementation of assert, need to replace with a proper kernel exit call. #if __HIP_DEVICE_COMPILE__ == 1 #undef assert @@ -179,10 +183,6 @@ extern int HIP_TRACE_API; #define __HCC_C__ #endif -#endif // defined __HCC__ - -#if __HCC_OR_HIP_CLANG__ - // TODO - hipify-clang - change to use the function call. //#define warpSize hc::__wavesize() static constexpr int warpSize = 64; From 8ddadf835d7e727d7429ba216db4e5410a2f1161 Mon Sep 17 00:00:00 2001 From: "Yaxun (Sam) Liu" Date: Tue, 5 Jun 2018 08:46:20 -0400 Subject: [PATCH 02/13] Add missing macro MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT to hip_complex.h --- hipamd/include/hip/hcc_detail/hip_complex.h | 3 +++ 1 file changed, 3 insertions(+) diff --git a/hipamd/include/hip/hcc_detail/hip_complex.h b/hipamd/include/hip/hcc_detail/hip_complex.h index 973d5f564b..22808ed576 100644 --- a/hipamd/include/hip/hcc_detail/hip_complex.h +++ b/hipamd/include/hip/hcc_detail/hip_complex.h @@ -94,6 +94,9 @@ THE SOFTWARE. ret.y = lhs.y * rhs; \ return ret; \ } +#define MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(ComplexT, T) \ + ComplexT(T val) : x(val), y(val) {} \ + ComplexT(T val1, T val2) : x(val1), y(val2) {} #endif From cc44a3d234bf7f047cf1b116486162c6133560be Mon Sep 17 00:00:00 2001 From: "Yaxun (Sam) Liu" Date: Thu, 31 May 2018 13:58:44 -0400 Subject: [PATCH 03/13] Include cmath instead of math.h in hip_complex.h --- hipamd/include/hip/hcc_detail/hip_complex.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/hipamd/include/hip/hcc_detail/hip_complex.h b/hipamd/include/hip/hcc_detail/hip_complex.h index 22808ed576..3dc7a2ed56 100644 --- a/hipamd/include/hip/hcc_detail/hip_complex.h +++ b/hipamd/include/hip/hcc_detail/hip_complex.h @@ -24,7 +24,7 @@ THE SOFTWARE. #define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COMPLEX_H #include "hip/hcc_detail/hip_vector_types.h" -#include +#include #if __cplusplus #define COMPLEX_ADD_OP_OVERLOAD(type) \ From 5d0d2ea4559861aa051d61da1ad30a339510543f Mon Sep 17 00:00:00 2001 From: "Yaxun (Sam) Liu" Date: Wed, 6 Jun 2018 11:15:54 -0400 Subject: [PATCH 04/13] Includes or by __cplusplus in hip_complex.h --- hipamd/include/hip/hcc_detail/hip_complex.h | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/hipamd/include/hip/hcc_detail/hip_complex.h b/hipamd/include/hip/hcc_detail/hip_complex.h index 3dc7a2ed56..72f139ba84 100644 --- a/hipamd/include/hip/hcc_detail/hip_complex.h +++ b/hipamd/include/hip/hcc_detail/hip_complex.h @@ -24,7 +24,15 @@ THE SOFTWARE. #define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COMPLEX_H #include "hip/hcc_detail/hip_vector_types.h" + +// TODO: Clang has a bug which allows device functions to call std functions +// when std functions are introduced into default namespace by using statement. +// math.h may be included after this bug is fixed. +#if __cplusplus #include +#else +#include "math.h" +#endif #if __cplusplus #define COMPLEX_ADD_OP_OVERLOAD(type) \ From 40d2d8a0bd1bce882da4686a152f409117362c75 Mon Sep 17 00:00:00 2001 From: "Yaxun (Sam) Liu" Date: Sun, 17 Jun 2018 12:18:05 -0400 Subject: [PATCH 05/13] Add missing __device__ __host__ to complex constructor Also add missing typedef value_type --- hipamd/include/hip/hcc_detail/hip_complex.h | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/hipamd/include/hip/hcc_detail/hip_complex.h b/hipamd/include/hip/hcc_detail/hip_complex.h index 72f139ba84..783833dc7e 100644 --- a/hipamd/include/hip/hcc_detail/hip_complex.h +++ b/hipamd/include/hip/hcc_detail/hip_complex.h @@ -103,14 +103,15 @@ THE SOFTWARE. return ret; \ } #define MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(ComplexT, T) \ - ComplexT(T val) : x(val), y(val) {} \ - ComplexT(T val1, T val2) : x(val1), y(val2) {} + __device__ __host__ ComplexT(T val) : x(val), y(val) {} \ + __device__ __host__ ComplexT(T val1, T val2) : x(val1), y(val2) {} #endif struct hipFloatComplex { #ifdef __cplusplus public: + typedef float value_type; __device__ __host__ hipFloatComplex() : x(0.0f), y(0.0f) {} __device__ __host__ hipFloatComplex(float x) : x(x), y(0.0f) {} __device__ __host__ hipFloatComplex(float x, float y) : x(x), y(y) {} @@ -130,6 +131,7 @@ struct hipFloatComplex { struct hipDoubleComplex { #ifdef __cplusplus public: + typedef double value_type; __device__ __host__ hipDoubleComplex() : x(0.0f), y(0.0f) {} __device__ __host__ hipDoubleComplex(double x) : x(x), y(0.0f) {} __device__ __host__ hipDoubleComplex(double x, double y) : x(x), y(y) {} From 0de3ef7b32aa95f1f4816b25423f6df46f7d5ea2 Mon Sep 17 00:00:00 2001 From: "Yaxun (Sam) Liu" Date: Mon, 18 Jun 2018 11:57:57 -0400 Subject: [PATCH 06/13] Add abs/real/imag functions for hipFloatComplex/hipDoubleComplex --- hipamd/include/hip/hcc_detail/hip_complex.h | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/hipamd/include/hip/hcc_detail/hip_complex.h b/hipamd/include/hip/hcc_detail/hip_complex.h index 783833dc7e..7f4dd840b7 100644 --- a/hipamd/include/hip/hcc_detail/hip_complex.h +++ b/hipamd/include/hip/hcc_detail/hip_complex.h @@ -234,7 +234,6 @@ __device__ __host__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hi __device__ __host__ static inline float hipCabsf(hipFloatComplex z) { return sqrtf(hipCsqabsf(z)); } - __device__ __host__ static inline double hipCreal(hipDoubleComplex z) { return z.x; } __device__ __host__ static inline double hipCimag(hipDoubleComplex z) { return z.y; } @@ -314,4 +313,12 @@ __device__ __host__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, h return make_hipDoubleComplex(real, imag); } +#define __DEFINE_HIP_COMPLEX_FUN(func) \ +__device__ __host__ inline float func(const hipFloatComplex& z) { return hipC##func##f(z); } \ +__device__ __host__ inline double func(const hipDoubleComplex& z) { return hipC##func(z); } + +__DEFINE_HIP_COMPLEX_FUN(abs) +__DEFINE_HIP_COMPLEX_FUN(real) +__DEFINE_HIP_COMPLEX_FUN(imag) + #endif From 6bdb5f35417771d29e539e3d36ded5c205060998 Mon Sep 17 00:00:00 2001 From: pradeepisro Date: Tue, 19 Jun 2018 12:26:18 +0530 Subject: [PATCH 07/13] added changes to FindHIP.cmake which would allow us to build using ninja #467 --- hipamd/cmake/FindHIP.cmake | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/hipamd/cmake/FindHIP.cmake b/hipamd/cmake/FindHIP.cmake index a45fb1b44d..e540ac35a6 100644 --- a/hipamd/cmake/FindHIP.cmake +++ b/hipamd/cmake/FindHIP.cmake @@ -471,7 +471,13 @@ macro(HIP_PREPARE_TARGET_COMMANDS _target _format _generated_files _source_files INPUT "${custom_target_script_pregen}" ) set(main_dep DEPENDS ${source_file}) - set(verbose_output "$(VERBOSE)") + if(CMAKE_GENERATOR MATCHES "Makefiles") + set(verbose_output "$(VERBOSE)") + elseif(HIP_VERBOSE_BUILD) + set(verbose_output ON) + else() + set(verbose_output OFF) + endif() # Create up the comment string file(RELATIVE_PATH generated_file_relative_path "${CMAKE_BINARY_DIR}" "${generated_file}") From bdfd6957aa9da6cb81563436b56a2b72a46a7822 Mon Sep 17 00:00:00 2001 From: pradeepisro Date: Tue, 19 Jun 2018 13:50:24 +0530 Subject: [PATCH 08/13] added option HIP_BUILD_VERBOSE to enable verbosity in HIP build --- hipamd/cmake/FindHIP.cmake | 1 + 1 file changed, 1 insertion(+) diff --git a/hipamd/cmake/FindHIP.cmake b/hipamd/cmake/FindHIP.cmake index e540ac35a6..b761356c9a 100644 --- a/hipamd/cmake/FindHIP.cmake +++ b/hipamd/cmake/FindHIP.cmake @@ -20,6 +20,7 @@ foreach(config ${_hip_configuration_types}) mark_as_advanced(HIP_HIPCC_FLAGS_${config_upper} HIP_HCC_FLAGS_${config_upper} HIP_NVCC_FLAGS_${config_upper}) endforeach() option(HIP_HOST_COMPILATION_CPP "Host code compilation mode" ON) +option(HIP_VERBOSE_BUILD "Print out the commands run while compiling the HIP source file. With the Makefile generator this defaults to VERBOSE variable specified on the command line, but can be forced on with this option." OFF) mark_as_advanced(HIP_HOST_COMPILATION_CPP) ############################################################################### From 2b5242b2c36ecc3f84d8ca027908bcb04a7f3b40 Mon Sep 17 00:00:00 2001 From: "Yaxun (Sam) Liu" Date: Mon, 18 Jun 2018 14:35:48 -0400 Subject: [PATCH 09/13] Add conj, operator-,==,!= for hipFloatComplex/hipDoubleComplex --- hipamd/include/hip/hcc_detail/hip_complex.h | 44 ++++++++++++++++++--- 1 file changed, 38 insertions(+), 6 deletions(-) diff --git a/hipamd/include/hip/hcc_detail/hip_complex.h b/hipamd/include/hip/hcc_detail/hip_complex.h index 7f4dd840b7..48a2852a5a 100644 --- a/hipamd/include/hip/hcc_detail/hip_complex.h +++ b/hipamd/include/hip/hcc_detail/hip_complex.h @@ -35,6 +35,24 @@ THE SOFTWARE. #endif #if __cplusplus +#define COMPLEX_NEG_OP_OVERLOAD(type) \ + __device__ __host__ static inline type operator-(const type& op) { \ + type ret; \ + ret.x = -op.x; \ + ret.y = -op.y; \ + return ret; \ + } + +#define COMPLEX_EQ_OP_OVERLOAD(type) \ + __device__ __host__ static inline bool operator==(const type& lhs, const type& rhs) { \ + return lhs.x == rhs.x && lhs.y == rhs.y; \ + } + +#define COMPLEX_NE_OP_OVERLOAD(type) \ + __device__ __host__ static inline bool operator!=(const type& lhs, const type& rhs) { \ + return !(lhs == rhs); \ + } + #define COMPLEX_ADD_OP_OVERLOAD(type) \ __device__ __host__ static inline type operator+(const type& lhs, const type& rhs) { \ type ret; \ @@ -150,6 +168,9 @@ struct hipDoubleComplex { #if __cplusplus +COMPLEX_NEG_OP_OVERLOAD(hipFloatComplex) +COMPLEX_EQ_OP_OVERLOAD(hipFloatComplex) +COMPLEX_NE_OP_OVERLOAD(hipFloatComplex) COMPLEX_ADD_OP_OVERLOAD(hipFloatComplex) COMPLEX_SUB_OP_OVERLOAD(hipFloatComplex) COMPLEX_MUL_OP_OVERLOAD(hipFloatComplex) @@ -169,6 +190,9 @@ COMPLEX_SCALAR_PRODUCT(hipFloatComplex, double) COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed long long) COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned long long) +COMPLEX_NEG_OP_OVERLOAD(hipDoubleComplex) +COMPLEX_EQ_OP_OVERLOAD(hipDoubleComplex) +COMPLEX_NE_OP_OVERLOAD(hipDoubleComplex) COMPLEX_ADD_OP_OVERLOAD(hipDoubleComplex) COMPLEX_SUB_OP_OVERLOAD(hipDoubleComplex) COMPLEX_MUL_OP_OVERLOAD(hipDoubleComplex) @@ -313,12 +337,20 @@ __device__ __host__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, h return make_hipDoubleComplex(real, imag); } -#define __DEFINE_HIP_COMPLEX_FUN(func) \ -__device__ __host__ inline float func(const hipFloatComplex& z) { return hipC##func##f(z); } \ -__device__ __host__ inline double func(const hipDoubleComplex& z) { return hipC##func(z); } +// Complex functions returning real numbers. +#define __DEFINE_HIP_COMPLEX_REAL_FUN(func, hipFun) \ +__device__ __host__ inline float func(const hipFloatComplex& z) { return hipFun##f(z); } \ +__device__ __host__ inline double func(const hipDoubleComplex& z) { return hipFun(z); } -__DEFINE_HIP_COMPLEX_FUN(abs) -__DEFINE_HIP_COMPLEX_FUN(real) -__DEFINE_HIP_COMPLEX_FUN(imag) +__DEFINE_HIP_COMPLEX_REAL_FUN(abs, hipCabs) +__DEFINE_HIP_COMPLEX_REAL_FUN(real, hipCreal) +__DEFINE_HIP_COMPLEX_REAL_FUN(imag, hipCimag) + +// Complex functions returning complex numbers. +#define __DEFINE_HIP_COMPLEX_FUN(func, hipFun) \ +__device__ __host__ inline hipFloatComplex func(const hipFloatComplex& z) { return hipFun##f(z); } \ +__device__ __host__ inline hipDoubleComplex func(const hipDoubleComplex& z) { return hipFun(z); } + +__DEFINE_HIP_COMPLEX_FUN(conj, hipConj) #endif From dc850cc3e77169efbc4f11285dca6f5a522e4c14 Mon Sep 17 00:00:00 2001 From: Aaron Enye Shi Date: Thu, 21 Jun 2018 18:12:55 +0000 Subject: [PATCH 10/13] HIPCC - Use clang if clang++ is not compiled --- hipamd/bin/hipcc | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/hipamd/bin/hipcc b/hipamd/bin/hipcc index 03f35b27fc..1be33d55e1 100755 --- a/hipamd/bin/hipcc +++ b/hipamd/bin/hipcc @@ -115,8 +115,13 @@ if ($HIP_PLATFORM eq "clang") { $ROCM_PATH=$ENV{'ROCM_PATH'} // "/opt/rocm"; $HIPCC="$HIP_CLANG_PATH/clang++"; $HIPCXXFLAGS .= "-std=c++11 -I$HIP_PATH/include"; - $HIPLDFLAGS = "--hip-link --hip-device-lib-path=$DEVICE_LIB_PATH -L$HIP_PATH/lib -lhip_hcc"; + + # If $HIPCC clang++ is not compiled, use clang instead + if ( ! -e $HIPCC ) { + $HIPCC="$HIP_CLANG_PATH/clang"; + $HIPLDFLAGS = "--driver-mode=g++ " . $HIPLDFLAGS; + } } elsif ($HIP_PLATFORM eq "hcc") { $HSA_PATH=$ENV{'HSA_PATH'} // "/opt/rocm/hsa"; From 9452298fbe29bf53cad18aa1d368d3345d5eaa9f Mon Sep 17 00:00:00 2001 From: Alex Voicu Date: Mon, 25 Jun 2018 19:16:27 +0100 Subject: [PATCH 11/13] Removes use of unimplemented OCML functionality. --- .../include/hip/hcc_detail/math_functions.h | 18 ++++++------- hipamd/include/hip/hcc_detail/math_fwd.h | 26 ------------------- 2 files changed, 9 insertions(+), 35 deletions(-) diff --git a/hipamd/include/hip/hcc_detail/math_functions.h b/hipamd/include/hip/hcc_detail/math_functions.h index a9288dc204..8f7c885fc7 100644 --- a/hipamd/include/hip/hcc_detail/math_functions.h +++ b/hipamd/include/hip/hcc_detail/math_functions.h @@ -508,19 +508,19 @@ inline float __fadd_rz(float x, float y) { return __ocml_add_rtz_f32(x, y); } __device__ inline -float __fdiv_rd(float x, float y) { return __ocml_div_rtp_f32(x, y); } +float __fdiv_rd(float x, float y) { return x / y; } __device__ inline -float __fdiv_rn(float x, float y) { return __ocml_div_rte_f32(x, y); } +float __fdiv_rn(float x, float y) { return x / y; } __device__ inline -float __fdiv_ru(float x, float y) { return __ocml_div_rtn_f32(x, y); } +float __fdiv_ru(float x, float y) { return x / y; } __device__ inline -float __fdiv_rz(float x, float y) { return __ocml_div_rtz_f32(x, y); } +float __fdiv_rz(float x, float y) { return x / y; } __device__ inline -float __fdividef(float x, float y) { return __ocml_div_rte_f32(x, y); } +float __fdividef(float x, float y) { return x / y; } __device__ inline float __fmaf_rd(float x, float y, float z) @@ -1028,16 +1028,16 @@ inline double __dadd_rz(double x, double y) { return __ocml_add_rtz_f64(x, y); } __device__ inline -double __ddiv_rd(double x, double y) { return __ocml_div_rtp_f64(x, y); } +double __ddiv_rd(double x, double y) { return x / y; } __device__ inline -double __ddiv_rn(double x, double y) { return __ocml_div_rte_f64(x, y); } +double __ddiv_rn(double x, double y) { return x / y; } __device__ inline -double __ddiv_ru(double x, double y) { return __ocml_div_rtn_f64(x, y); } +double __ddiv_ru(double x, double y) { return x / y; } __device__ inline -double __ddiv_rz(double x, double y) { return __ocml_div_rtz_f64(x, y); } +double __ddiv_rz(double x, double y) { return x / y; } __device__ inline double __dmul_rd(double x, double y) { return __ocml_mul_rtp_f64(x, y); } diff --git a/hipamd/include/hip/hcc_detail/math_fwd.h b/hipamd/include/hip/hcc_detail/math_fwd.h index ac5d6e7dc6..102714e117 100644 --- a/hipamd/include/hip/hcc_detail/math_fwd.h +++ b/hipamd/include/hip/hcc_detail/math_fwd.h @@ -69,8 +69,6 @@ float __ocml_cosh_f32(float); __device__ float __ocml_cospi_f32(float); __device__ -float __ocml_div_rtz_f32(float, float); -__device__ float __ocml_i0_f32(float); __device__ float __ocml_i1_f32(float); @@ -290,18 +288,6 @@ __attribute__((const)) float __ocml_mul_rtz_f32(float, float); __device__ __attribute__((const)) -float __ocml_div_rte_f32(float, float); -__device__ -__attribute__((const)) -float __ocml_div_rtn_f32(float, float); -__device__ -__attribute__((const)) -float __ocml_div_rtp_f32(float, float); -__device__ -__attribute__((const)) -float __ocml_div_rtz_f32(float, float); -__device__ -__attribute__((const)) float __ocml_sqrt_rte_f32(float); __device__ __attribute__((const)) @@ -598,18 +584,6 @@ __attribute__((const)) double __ocml_mul_rtz_f64(double, double); __device__ __attribute__((const)) -double __ocml_div_rte_f64(double, double); -__device__ -__attribute__((const)) -double __ocml_div_rtn_f64(double, double); -__device__ -__attribute__((const)) -double __ocml_div_rtp_f64(double, double); -__device__ -__attribute__((const)) -double __ocml_div_rtz_f64(double, double); -__device__ -__attribute__((const)) double __ocml_sqrt_rte_f64(double); __device__ __attribute__((const)) From c74682ef1b647d39f5ea9b82f28e327e5df8e882 Mon Sep 17 00:00:00 2001 From: "Yaxun (Sam) Liu" Date: Mon, 25 Jun 2018 15:33:41 -0400 Subject: [PATCH 12/13] Include host_defines.h in hip_fp16.h since it uses __host__ __device__ attributes --- hipamd/include/hip/hcc_detail/hip_fp16.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/hipamd/include/hip/hcc_detail/hip_fp16.h b/hipamd/include/hip/hcc_detail/hip_fp16.h index 8657bc30a3..777113a9de 100644 --- a/hipamd/include/hip/hcc_detail/hip_fp16.h +++ b/hipamd/include/hip/hcc_detail/hip_fp16.h @@ -21,7 +21,7 @@ THE SOFTWARE. */ #pragma once - +#include "hip/hcc_detail/host_defines.h" #include #if defined(__cplusplus) #include From f426491f17f044b97014a61f59216bb6d7a35e66 Mon Sep 17 00:00:00 2001 From: Maneesh Gupta Date: Thu, 28 Jun 2018 11:19:22 +0530 Subject: [PATCH 13/13] Fix typo --- hipamd/include/hip/hcc_detail/hip_runtime.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/hipamd/include/hip/hcc_detail/hip_runtime.h b/hipamd/include/hip/hcc_detail/hip_runtime.h index 7c1f2a4d16..ca4e73eee0 100644 --- a/hipamd/include/hip/hcc_detail/hip_runtime.h +++ b/hipamd/include/hip/hcc_detail/hip_runtime.h @@ -109,7 +109,7 @@ extern int HIP_TRACE_API; #include #include #include -#if __HCC___ +#if __HCC__ #include #include #endif // __HCC__