Change kernel function names to use ISA name directly (pytorch#156)

Summary: Pull Request resolved: pytorch#156 Use instruction set name from the description structure while naming the kernel functions Reviewed By: jspark1105 Differential Revision: D18239352 fbshipit-source-id: 0678e4ad01e19465e8ea7cdf6563feaf2ce7e501
clin99 · Oct 31, 2019 · 0bf264f · 0bf264f
1 parent b884868
commit 0bf264f
Show file tree

Hide file tree

Showing 6 changed files with 86 additions and 77 deletions.
diff --git a/src/FbgemmFP16.cc b/src/FbgemmFP16.cc
@@ -45,34 +45,34 @@ inline void PackA(int nrow, int ncol, const float* from, int ldim, float* to) {
 // }
 
 struct KernelInfo {
-  using knl_ptr = funcptr_fp16;
-  // optimized kernels to cover all cases
-  // 2 in ?x2 should be the same as kernel_ncol_blocks.
-  // Here with kernel_ncol_blocks = 2, we can provide up to 6x2 kernels, due to
-  // the restrictions of ymm register numbers (16).
-  static constexpr knl_ptr kernel_avx2[] = {nullptr,
-                                            gemmkernel_1x2_AVX2_fA0fB0fC0,
-                                            gemmkernel_2x2_AVX2_fA0fB0fC0,
-                                            gemmkernel_3x2_AVX2_fA0fB0fC0,
-                                            gemmkernel_4x2_AVX2_fA0fB0fC0,
-                                            gemmkernel_5x2_AVX2_fA0fB0fC0,
-                                            gemmkernel_6x2_AVX2_fA0fB0fC0};
+using knl_ptr = funcptr_fp16;
+// optimized kernels to cover all cases
+// 2 in ?x2 should be the same as kernel_ncol_blocks.
+// Here with kernel_ncol_blocks = 2, we can provide up to 6x2 kernels, due to
+// the restrictions of ymm register numbers (16).
+static constexpr knl_ptr kernel_avx2[] = {nullptr,
+                                          gemmkernel_1x2_Avx2_fA0fB0fC0,
+                                          gemmkernel_2x2_Avx2_fA0fB0fC0,
+                                          gemmkernel_3x2_Avx2_fA0fB0fC0,
+                                          gemmkernel_4x2_Avx2_fA0fB0fC0,
+                                          gemmkernel_5x2_Avx2_fA0fB0fC0,
+                                          gemmkernel_6x2_Avx2_fA0fB0fC0};
 
-  static constexpr knl_ptr kernel_avx512[] = {nullptr,
-                                              gemmkernel_1x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_2x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_3x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_4x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_5x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_6x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_7x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_8x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_9x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_10x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_11x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_12x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_13x2_AVX512_fA0fB0fC0,
-                                              gemmkernel_14x2_AVX512_fA0fB0fC0};
+static constexpr knl_ptr kernel_avx512[] = {nullptr,
+                                            gemmkernel_1x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_2x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_3x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_4x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_5x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_6x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_7x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_8x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_9x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_10x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_11x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_12x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_13x2_Avx512_fA0fB0fC0,
+                                            gemmkernel_14x2_Avx512_fA0fB0fC0};
 
   // autotuned kernel splits for various cases m = 1:mb_max
   // may need re-autotuning for new uarch

diff --git a/src/FbgemmFP16UKernelsAvx2.cc b/src/FbgemmFP16UKernelsAvx2.cc
@@ -8,7 +8,7 @@
 
 namespace fbgemm {
 
-void __attribute__((noinline)) gemmkernel_1x2_AVX2_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_1x2_Avx2_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -106,7 +106,7 @@ void __attribute__((noinline)) gemmkernel_1x2_AVX2_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_2x2_AVX2_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_2x2_Avx2_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -217,7 +217,7 @@ void __attribute__((noinline)) gemmkernel_2x2_AVX2_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_3x2_AVX2_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_3x2_Avx2_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -341,7 +341,7 @@ void __attribute__((noinline)) gemmkernel_3x2_AVX2_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_4x2_AVX2_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_4x2_Avx2_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -478,7 +478,7 @@ void __attribute__((noinline)) gemmkernel_4x2_AVX2_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_5x2_AVX2_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_5x2_Avx2_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -628,7 +628,7 @@ void __attribute__((noinline)) gemmkernel_5x2_AVX2_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_6x2_AVX2_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_6x2_Avx2_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"

diff --git a/src/FbgemmFP16UKernelsAvx2.h b/src/FbgemmFP16UKernelsAvx2.h
@@ -23,12 +23,12 @@ struct GemmParams {
   uint64_t b_block_cols;
   uint64_t b_block_size;
 };
-void __attribute__((noinline)) gemmkernel_1x2_AVX2_fA0fB0fC0(GemmParams* gp);
-void __attribute__((noinline)) gemmkernel_2x2_AVX2_fA0fB0fC0(GemmParams* gp);
-void __attribute__((noinline)) gemmkernel_3x2_AVX2_fA0fB0fC0(GemmParams* gp);
-void __attribute__((noinline)) gemmkernel_4x2_AVX2_fA0fB0fC0(GemmParams* gp);
-void __attribute__((noinline)) gemmkernel_5x2_AVX2_fA0fB0fC0(GemmParams* gp);
-void __attribute__((noinline)) gemmkernel_6x2_AVX2_fA0fB0fC0(GemmParams* gp);
+void __attribute__((noinline)) gemmkernel_1x2_Avx2_fA0fB0fC0(GemmParams* gp);
+void __attribute__((noinline)) gemmkernel_2x2_Avx2_fA0fB0fC0(GemmParams* gp);
+void __attribute__((noinline)) gemmkernel_3x2_Avx2_fA0fB0fC0(GemmParams* gp);
+void __attribute__((noinline)) gemmkernel_4x2_Avx2_fA0fB0fC0(GemmParams* gp);
+void __attribute__((noinline)) gemmkernel_5x2_Avx2_fA0fB0fC0(GemmParams* gp);
+void __attribute__((noinline)) gemmkernel_6x2_Avx2_fA0fB0fC0(GemmParams* gp);
 typedef void (*funcptr_fp16)(GemmParams* gp);
 ;
 

diff --git a/src/FbgemmFP16UKernelsAvx512.cc b/src/FbgemmFP16UKernelsAvx512.cc
@@ -8,7 +8,7 @@
 
 namespace fbgemm {
 
-void __attribute__((noinline)) gemmkernel_1x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_1x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -46,6 +46,7 @@ void __attribute__((noinline)) gemmkernel_1x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm0,zmm0,zmm0\t\n"
       "vxorps zmm1,zmm1,zmm1\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm3,YMMWORD PTR [r10 + 0]\t\n"
@@ -105,7 +106,7 @@ void __attribute__((noinline)) gemmkernel_1x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_2x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_2x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -145,6 +146,7 @@ void __attribute__((noinline)) gemmkernel_2x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm2,zmm2,zmm2\t\n"
       "vxorps zmm3,zmm3,zmm3\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm5,YMMWORD PTR [r10 + 0]\t\n"
@@ -215,7 +217,7 @@ void __attribute__((noinline)) gemmkernel_2x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_3x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_3x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -257,6 +259,7 @@ void __attribute__((noinline)) gemmkernel_3x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm4,zmm4,zmm4\t\n"
       "vxorps zmm5,zmm5,zmm5\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm7,YMMWORD PTR [r10 + 0]\t\n"
@@ -338,7 +341,7 @@ void __attribute__((noinline)) gemmkernel_3x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_4x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_4x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -382,6 +385,7 @@ void __attribute__((noinline)) gemmkernel_4x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm6,zmm6,zmm6\t\n"
       "vxorps zmm7,zmm7,zmm7\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm9,YMMWORD PTR [r10 + 0]\t\n"
@@ -474,7 +478,7 @@ void __attribute__((noinline)) gemmkernel_4x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_5x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_5x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -520,6 +524,7 @@ void __attribute__((noinline)) gemmkernel_5x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm8,zmm8,zmm8\t\n"
       "vxorps zmm9,zmm9,zmm9\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm11,YMMWORD PTR [r10 + 0]\t\n"
@@ -623,7 +628,7 @@ void __attribute__((noinline)) gemmkernel_5x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_6x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_6x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -671,6 +676,7 @@ void __attribute__((noinline)) gemmkernel_6x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm10,zmm10,zmm10\t\n"
       "vxorps zmm11,zmm11,zmm11\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm13,YMMWORD PTR [r10 + 0]\t\n"
@@ -785,7 +791,7 @@ void __attribute__((noinline)) gemmkernel_6x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_7x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_7x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -835,6 +841,7 @@ void __attribute__((noinline)) gemmkernel_7x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm12,zmm12,zmm12\t\n"
       "vxorps zmm13,zmm13,zmm13\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm15,YMMWORD PTR [r10 + 0]\t\n"
@@ -960,7 +967,7 @@ void __attribute__((noinline)) gemmkernel_7x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_8x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_8x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -1012,6 +1019,7 @@ void __attribute__((noinline)) gemmkernel_8x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm14,zmm14,zmm14\t\n"
       "vxorps zmm15,zmm15,zmm15\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm17,YMMWORD PTR [r10 + 0]\t\n"
@@ -1148,7 +1156,7 @@ void __attribute__((noinline)) gemmkernel_8x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline)) gemmkernel_9x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_9x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -1202,6 +1210,7 @@ void __attribute__((noinline)) gemmkernel_9x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm16,zmm16,zmm16\t\n"
       "vxorps zmm17,zmm17,zmm17\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm19,YMMWORD PTR [r10 + 0]\t\n"
@@ -1349,8 +1358,7 @@ void __attribute__((noinline)) gemmkernel_9x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline))
-gemmkernel_10x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_10x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -1406,6 +1414,7 @@ gemmkernel_10x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm18,zmm18,zmm18\t\n"
       "vxorps zmm19,zmm19,zmm19\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm21,YMMWORD PTR [r10 + 0]\t\n"
@@ -1564,8 +1573,7 @@ gemmkernel_10x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline))
-gemmkernel_11x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_11x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -1623,6 +1631,7 @@ gemmkernel_11x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm20,zmm20,zmm20\t\n"
       "vxorps zmm21,zmm21,zmm21\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm23,YMMWORD PTR [r10 + 0]\t\n"
@@ -1792,8 +1801,7 @@ gemmkernel_11x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline))
-gemmkernel_12x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_12x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -1853,6 +1861,7 @@ gemmkernel_12x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm22,zmm22,zmm22\t\n"
       "vxorps zmm23,zmm23,zmm23\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm25,YMMWORD PTR [r10 + 0]\t\n"
@@ -2033,8 +2042,7 @@ gemmkernel_12x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline))
-gemmkernel_13x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_13x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -2096,6 +2104,7 @@ gemmkernel_13x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm24,zmm24,zmm24\t\n"
       "vxorps zmm25,zmm25,zmm25\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm27,YMMWORD PTR [r10 + 0]\t\n"
@@ -2287,8 +2296,7 @@ gemmkernel_13x2_AVX512_fA0fB0fC0(GemmParams* gp) {
         "r12",
         "memory");
 }
-void __attribute__((noinline))
-gemmkernel_14x2_AVX512_fA0fB0fC0(GemmParams* gp) {
+void __attribute__((noinline)) gemmkernel_14x2_Avx512_fA0fB0fC0(GemmParams* gp) {
   asm volatile(
 #if !defined(__clang__)
       "mov r14, %[gp]\t\n"
@@ -2352,6 +2360,7 @@ gemmkernel_14x2_AVX512_fA0fB0fC0(GemmParams* gp) {
       "vxorps zmm26,zmm26,zmm26\t\n"
       "vxorps zmm27,zmm27,zmm27\t\n"
 
+
       "loop_inner%=:\t\n"
 
       "vcvtph2ps zmm29,YMMWORD PTR [r10 + 0]\t\n"