derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
753 lines
24 KiB
ReStructuredText
753 lines
24 KiB
ReStructuredText
..
|
|
This file was automatically generated. Do not edit.
|
|
|
|
st.global.b8
|
|
^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.b8 [addr], src; // PTX ISA 10, SM_50
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src);
|
|
|
|
st.global.b16
|
|
^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.b16 [addr], src; // PTX ISA 10, SM_50
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src);
|
|
|
|
st.global.b32
|
|
^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.b32 [addr], src; // PTX ISA 10, SM_50
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src);
|
|
|
|
st.global.b64
|
|
^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.b64 [addr], src; // PTX ISA 10, SM_50
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src);
|
|
|
|
st.global.b128
|
|
^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.b128 [addr], src; // PTX ISA 83, SM_70
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src);
|
|
|
|
st.global.L2::cache_hint.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L2::cache_hint.b8 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L2::cache_hint.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L2::cache_hint.b16 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L2::cache_hint.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L2::cache_hint.b32 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L2::cache_hint.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L2::cache_hint.b64 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L2::cache_hint.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L2::cache_hint.b128 [addr], src, cache_policy; // PTX ISA 83, SM_80
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_normal.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.b8 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src);
|
|
|
|
st.global.L1::evict_normal.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.b16 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src);
|
|
|
|
st.global.L1::evict_normal.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.b32 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src);
|
|
|
|
st.global.L1::evict_normal.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.b64 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src);
|
|
|
|
st.global.L1::evict_normal.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.b128 [addr], src; // PTX ISA 83, SM_70
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src);
|
|
|
|
st.global.L1::evict_normal.L2::cache_hint.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.L2::cache_hint.b8 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_normal.L2::cache_hint.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.L2::cache_hint.b16 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_normal.L2::cache_hint.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.L2::cache_hint.b32 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_normal.L2::cache_hint.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.L2::cache_hint.b64 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_normal.L2::cache_hint.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_normal.L2::cache_hint.b128 [addr], src, cache_policy; // PTX ISA 83, SM_80
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_evict_normal_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_unchanged.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.b8 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src);
|
|
|
|
st.global.L1::evict_unchanged.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.b16 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src);
|
|
|
|
st.global.L1::evict_unchanged.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.b32 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src);
|
|
|
|
st.global.L1::evict_unchanged.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.b64 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src);
|
|
|
|
st.global.L1::evict_unchanged.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.b128 [addr], src; // PTX ISA 83, SM_70
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src);
|
|
|
|
st.global.L1::evict_unchanged.L2::cache_hint.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.L2::cache_hint.b8 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_unchanged.L2::cache_hint.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.L2::cache_hint.b16 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_unchanged.L2::cache_hint.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.L2::cache_hint.b32 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_unchanged.L2::cache_hint.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.L2::cache_hint.b64 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_unchanged.L2::cache_hint.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_unchanged.L2::cache_hint.b128 [addr], src, cache_policy; // PTX ISA 83, SM_80
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_evict_unchanged_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_first.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.b8 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_evict_first(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src);
|
|
|
|
st.global.L1::evict_first.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.b16 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_evict_first(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src);
|
|
|
|
st.global.L1::evict_first.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.b32 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_evict_first(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src);
|
|
|
|
st.global.L1::evict_first.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.b64 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_evict_first(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src);
|
|
|
|
st.global.L1::evict_first.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.b128 [addr], src; // PTX ISA 83, SM_70
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_evict_first(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src);
|
|
|
|
st.global.L1::evict_first.L2::cache_hint.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.L2::cache_hint.b8 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_evict_first_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_first.L2::cache_hint.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.L2::cache_hint.b16 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_evict_first_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_first.L2::cache_hint.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.L2::cache_hint.b32 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_evict_first_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_first.L2::cache_hint.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.L2::cache_hint.b64 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_evict_first_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_first.L2::cache_hint.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_first.L2::cache_hint.b128 [addr], src, cache_policy; // PTX ISA 83, SM_80
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_evict_first_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_last.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.b8 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_evict_last(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src);
|
|
|
|
st.global.L1::evict_last.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.b16 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_evict_last(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src);
|
|
|
|
st.global.L1::evict_last.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.b32 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_evict_last(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src);
|
|
|
|
st.global.L1::evict_last.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.b64 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_evict_last(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src);
|
|
|
|
st.global.L1::evict_last.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.b128 [addr], src; // PTX ISA 83, SM_70
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_evict_last(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src);
|
|
|
|
st.global.L1::evict_last.L2::cache_hint.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.L2::cache_hint.b8 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_evict_last_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_last.L2::cache_hint.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.L2::cache_hint.b16 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_evict_last_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_last.L2::cache_hint.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.L2::cache_hint.b32 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_evict_last_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_last.L2::cache_hint.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.L2::cache_hint.b64 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_evict_last_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::evict_last.L2::cache_hint.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::evict_last.L2::cache_hint.b128 [addr], src, cache_policy; // PTX ISA 83, SM_80
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_evict_last_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::no_allocate.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.b8 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src);
|
|
|
|
st.global.L1::no_allocate.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.b16 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src);
|
|
|
|
st.global.L1::no_allocate.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.b32 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src);
|
|
|
|
st.global.L1::no_allocate.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.b64 [addr], src; // PTX ISA 74, SM_70
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src);
|
|
|
|
st.global.L1::no_allocate.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.b128 [addr], src; // PTX ISA 83, SM_70
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src);
|
|
|
|
st.global.L1::no_allocate.L2::cache_hint.b8
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.L2::cache_hint.b8 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B8, enable_if_t<sizeof(B8) == 1, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B8* addr,
|
|
B8 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::no_allocate.L2::cache_hint.b16
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.L2::cache_hint.b16 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B16, enable_if_t<sizeof(B16) == 2, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B16* addr,
|
|
B16 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::no_allocate.L2::cache_hint.b32
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.L2::cache_hint.b32 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B32, enable_if_t<sizeof(B32) == 4, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B32* addr,
|
|
B32 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::no_allocate.L2::cache_hint.b64
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.L2::cache_hint.b64 [addr], src, cache_policy; // PTX ISA 74, SM_80
|
|
// .space = { .global }
|
|
template <typename B64, enable_if_t<sizeof(B64) == 8, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B64* addr,
|
|
B64 src,
|
|
uint64_t cache_policy);
|
|
|
|
st.global.L1::no_allocate.L2::cache_hint.b128
|
|
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
.. code:: cuda
|
|
|
|
// st.space.L1::no_allocate.L2::cache_hint.b128 [addr], src, cache_policy; // PTX ISA 83, SM_80
|
|
// .space = { .global }
|
|
template <typename B128, enable_if_t<sizeof(B128) == 16, bool> = true>
|
|
__device__ static inline void st_L1_no_allocate_L2_cache_hint(
|
|
cuda::ptx::space_global_t,
|
|
B128* addr,
|
|
B128 src,
|
|
uint64_t cache_policy);
|