Address review comments

-Fix LD and ST naming for the quadword -Remove bf16 support for the QV instructions
ARM-software · Jun 23, 2023 · 5508e93 · 5508e93
1 parent 530c532
commit 5508e93
Showing 1 changed file with 50 additions and 58 deletions.
diff --git a/main/acle.md b/main/acle.md
@@ -9544,11 +9544,17 @@ Extract vector segment from each pair of quadword segments.
 Contiguous zero-extending load to quadword (single vector).
 
  ``` c
-   svuint32_t svld1uw_u128[_u32](svbool_t, const uint32_t *ptr);
-   svuint32_t svld1uw_u128[_u32](svbool_t, const uint32_t *ptr, int64_t vnum);
+   // Variants are also available for:
+   // _s8, _u16, _f16, _u32, _s32
+   // _bf16, _f16, _f32
+   svuint8_t svld1quw[_u8](svbool_t, const uint8_t *ptr);
+   svuint8_t svld1quw_vnum[_u8](svbool_t, const uint8_t *ptr, int64_t vnum);
 
-   svuint64_t svld1ud_u128[_u64](svbool_t, const uint64_t *ptr);
-   svuint64_t svld1ud_u128[_u64](svbool_t, const uint64_t *ptr, int64_t vnum);
+   // Variants are also available for:
+   // _s8, _u16, _f16, _u32, _s32, _u64, _s64
+   // _bf16, _f16, _f32, _f64
+   svuint8_t svld1qud[_u8](svbool_t, const uint8_t *ptr);
+   svuint8_t svld1qud_vnum[_u8](svbool_t, const uint8_t *ptr, int64_t vnum);
    ```
 
 #### LD1B, LD1D, LD1H, LD1W
@@ -9599,10 +9605,17 @@ Gather Load Quadword.
 
  ``` c
    // Variants are also available for:
-   // _u64base_u8, _u64base_u16, _u64base_s16, _u64base_u32, _u64base_s32,
-   // _u64base _u64, _u64base_s64
-   // _u64base_bf16, _u64base_f16, _u64base_f32, _u64base_f64
-   svint8_t svld1q_gather[_u64base_s8](svbool_t pg, svint64_t zn, const void *rm);
+   // _u8, _u16, _s16, _u32, _s32, _u64, _s64
+   // _bf16, _f16, _f32, _f64
+   svint8_t svld1q_gather[_u64base]_s8(svbool_t pg, svuint64_t zn);
+   svint8_t svld1q_gather[_u64base]_offset_s8(svbool_t pg, svuint64_t zn, int64_t offset);
+   svint8_t svld1q_gather[_u64base]_index_s8(svbool_t pg, svuint64_t zn, int64_t index);
+   svint8_t svld1q_gather_[u64]offset[_s8](svbool_t pg, const int8_t *base, svuint64_t offset);
+
+   // Variants are also available for:
+   // _u16, _u32, _s32, _u64, _s64
+   // _bf16, _f16, _f32, _f64
+   svint16_t svld1q_gather_[u64]index[_s16](svbool_t pg, const int16_t *base, svuint64_t index);
    ```
 
 #### LD2Q, LD3Q, LD4Q
@@ -9670,14 +9683,9 @@ Max/Min reduction of quadword vector segments.
  ``` c
    // Variants are also available for:
    // _s8, _u16, _s16, _u32, _s32, _u64, _s64
-   // _bf16, _f16, _f32, _f64
-   uint8x16_t svmaxqv[_u8]_m(svbool_t pg, svuint8_t zn);
-   uint8x16_t svmaxqv[_u8]_z(svbool_t pg, svuint8_t zn);
-   uint8x16_t svmaxqv[_u8]_x(svbool_t pg, svuint8_t zn);
-
-   uint8x16_t svminqv[_u8]_m(svbool_t pg, svuint8_t zn);
-   uint8x16_t svminqv[_u8]_z(svbool_t pg, svuint8_t zn);
-   uint8x16_t svminqv[_u8]_x(svbool_t pg, svuint8_t zn);
+   // _f16, _f32, _f64
+   uint8x16_t svmaxqv[_u8](svbool_t pg, svuint8_t zn);
+   uint8x16_t svminqv[_u8](svbool_t pg, svuint8_t zn);
    ```
 
 #### FMAXNMQV, FMINNMQV
@@ -9686,13 +9694,8 @@ Max/Min recursive reduction of quadword vector segments.
 
  ``` c
    // Variants are also available for  _f32, _f64
-   float16x8_t svmaxnmqv[_f16]_m(svbool_t pg, svfloat16_t zn);
-   float16x8_t svmaxnmqv[_f16]_z(svbool_t pg, svfloat16_t zn);
-   float16x8_t svmaxnmqv[_f16]_x(svbool_t pg, svfloat16_t zn);
-
-   float16x8_t svminnmqv[_f16]_m(svbool_t pg, svfloat16_t zn);
-   float16x8_t svminnmqv[_f16]_z(svbool_t pg, svfloat16_t zn);
-   float16x8_t svminnmqv[_f16]_x(svbool_t pg, svfloat16_t zn);
+   float16x8_t svmaxnmqv[_f16](svbool_t pg, svfloat16_t zn);
+   float16x8_t svminnmqv[_f16](svbool_t pg, svfloat16_t zn);
    ```
 
 #### BFMLSLB, BFMLSLT
@@ -9772,6 +9775,7 @@ Reverse doublewords in elements.
    // All the intrinsics below are [SVE2.1 or SME]
    // Variants are available for:
    // _s8, _s16, _u16, _s32, _u32, _s64, _u64
+   // _bf16, _f16, _f32, _f64
    svuint8_t svrevd[_u8]_m(svuint8_t zd, svbool_t pg, svuint8_t zn);
    svuint8_t svrevd[_u8]_z(svbool_t pg, svuint8_t zn);
    svuint8_t svrevd[_u8]_x(svbool_t pg, svuint8_t zn);
@@ -9803,23 +9807,21 @@ Contiguous store of single vector operand. It is truncating store from quadword.
  ``` c
 
    // Variants are also available for:
-   //  s128_vnum[_s32]
-   void svst1w_u128_vnum[_u32](svbool_t pg, uint32_t *ptr, uint64_t vnum,
-                               svuint32_t zt);
+   // _s8, _u16, _f16, _u32, _s32
+   // _bf16, _f16, _f32
+   void svst1quw[_u8](svbool_t pg, uint8_t *rn, svuint8_t zt);
+   void svst1quw_vnum[_u8](svbool_t pg, uint8_t *ptr, int64_t vnum,
+                               svuint8_t zt);
 
-   // Variants are also available for:
-   //  s128[_s32]
-   void svst1w_u128[_u32](svbool_t pg, uint32_t *rn, svuint32_t zt);
 
    // Variants are also available for:
-   //  s128_vnum[_s64]
-   void svst1d_u128_vnum[_u64](svbool_t pg, uint64_t *ptr, int64_t vnum,
-                               svuint64_t zt);
-
-   // Variants are also available for:
-   //  s128[_s64]
-   void svst1d_u128[_u64](svbool_t pg, uint64_t *rn, svuint64_t zt);
+   // _s8, _u16, _f16, _u32, _s32, _u64, _s64
+   // _bf16, _f16, _f32, _f64
+   void svst1qud[_u8](svbool_t pg, uint8_t *rn, svuint8_t zt);
+   void svst1qud_vnum[_u8](svbool_t pg, uint8_t *ptr, int64_t vnum,
+                               svuint8_t zt);
    ```
+
 #### ST1B, ST1D, ST1H, ST1W
 
 Contiguous store of multi-vector operand.
@@ -9926,27 +9928,17 @@ Scatter store quadwords.
 
  ``` c
    // Variants are also available for:
-   // _u64base_u8, _u64base_u16, _u64base_s16, _u64base_u32, _u64base_s32,
-   // _u64base_u64, _u64base_s64
-   // _u64base_bf16, _u64base_f16, _u64base_f32, _u64base_f64
-   void svst1q_scatter[_u64base_s8](svbool_t pg, svint64_t zn, const void *rm,
-                                    svint8_t zt);
-   ```
-
-#### ST2Q, ST3Q, ST4Q
-
-Contiguous store.
+   // _u8, _u16, _s16, _u32, _s32, _u64, _s64
+   // _bf16, _f16, _f32, _f64
+   void svst1q_scatter[_u64base]_s8(svbool_t pg, svuint64_t zn, svint8_t data);
+   void svst1q_scatter[_u64base]_offset_s8(svbool_t pg, svuint64_t zn, int64_t offset, svint8_t data);
+   void svst1q_scatter[_u64base]_index_s8(svbool_t pg, svuint64_t zn, int64_t index, svint8_t data);
+   void svst1q_scatter_[u64]offset[_s8](svbool_t pg, const uint *base, svuint64_t offset, svint8_t data);
 
- ``` c
    // Variants are also available for:
-   // _s8 _u16, _s16, _u32, _s32, _u64, _s64
+   // _u16, _u32, _s32, _u64, _s64
    // _bf16, _f16, _f32, _f64
-   void svst2q[_u8](svbool_t pg, uint8_t *rn, svuint8x2_t zt);
-   void svst2q_vnum[_u8](svbool_t pg, uint8_t *rn, int64_t vnum, svuint8x2_t zt);
-   void svst3q[_u8](svbool_t pg, uint8_t *rn, svuint8x3_t zt);
-   void svst3q_vnum[_u8](svbool_t pg, uint8_t *rn, int64_t vnum, svuint8x3_t zt);
-   void svst4q[_u8](svbool_t pg, uint8_t *rn, svuint8x4_t zt);
-   void svst4q_vnum[_u8](svbool_t pg, uint8_t *rn, int64_t vnum, svuint8x4_t zt);
+   void svst1q_scatter_[u64]index_s16(svbool_t pg, const int16_t *base, svuint64_t index, svint16_t data);
    ```
 
 #### ST2Q, ST3Q, ST4Q
@@ -9971,9 +9963,9 @@ Programmable table lookup within each quadword vector segment (zeroing).
 
  ``` c
    // Variants are also available for:
-   // _s8, _u16, _s16, _u32, _s32, _u64, _s64
+   // _u8, _u16, _s16, _u32, _s32, _u64, _s64
    // _bf16, _f16, _f32, _f64
-   svuint8_t svtblq[_u8](svuint8_t zn, svuint8_t zm);
+   svint8_t svtblq[_s8](svint8_t zn, svuint8_t zm);
    ```
 
 #### TBXQ
@@ -9982,9 +9974,9 @@ Programmable table lookup within each quadword vector segment (merging).
 
  ``` c
    // Variants are also available for:
-   // _s8, _u16, _s16, _u32, _s32, _u64, _s64
+   // _u8, _u16, _s16, _u32, _s32, _u64, _s64
    // _bf16, _f16, _f32, _f64
-   svuint8_t svtbxq[_u8](svuint8_t zn, svuint8_t zm);
+   svint8_t svtbxq[_s8](svint8_t fallback, svint8_t zn, svuint8_t zm);
    ```
 
 #### UZPQ1, UZPQ2