floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'upstream/develop' into dyn

Browse Source
This commit is contained in:
Isuru Fernando
2017-08-06 19:07:00 +05:30
parent 1d1854032b 719fcc56b0
commit 505b218829
32 changed files with 527 additions and 353 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
kernel/CMakeLists.txt
View File

@@ -1,7 +1,6 @@
include_directories(${PROJECT_SOURCE_DIR})
# Makefile
function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
set (OPENBLAS_SRC "")
@@ -21,7 +20,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
endif ()
if (${ARCH} STREQUAL "x86")
if (NOT MSVC)
if (NOT "${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
GenerateNamedObjects("${KERNELDIR}/cpuid.S" "" "" false "" "" true)
else()
GenerateNamedObjects("${KERNELDIR}/cpuid_win.c" "" "" false "" "" true)

42
kernel/arm64/casum_thunderx2t99.c
View File

@@ -147,57 +147,57 @@ static FLOAT casum_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
" fmov s6, "REG0" \n"
" fmov s7, "REG0" \n"
" cmp "N", xzr \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
" cmp "INC_X", xzr \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
" cmp "INC_X", #1 \n"
" bne .Lasum_kernel_S_BEGIN \n"
" bne 5f //asum_kernel_S_BEGIN \n"
".Lasum_kernel_F_BEGIN: \n"
"1: //asum_kernel_F_BEGIN: \n"
" asr "J", "N", #5 \n"
" cmp "J", xzr \n"
" beq .Lasum_kernel_F1 \n"
" beq 3f //asum_kernel_F1 \n"
".Lasum_kernel_F32: \n"
"2: //asum_kernel_F32: \n"
" "KERNEL_F32" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_F32 \n"
" bne 2b //asum_kernel_F32 \n"
" "KERNEL_F32_FINALIZE" \n"
".Lasum_kernel_F1: \n"
"3: //asum_kernel_F1: \n"
" ands "J", "N", #31 \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
".Lasum_kernel_F10: \n"
"4: //asum_kernel_F10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_F10 \n"
" b .Lasum_kernel_L999 \n"
" bne 4b //asum_kernel_F10 \n"
" b 9f //asum_kernel_L999 \n"
".Lasum_kernel_S_BEGIN: \n"
"5: //asum_kernel_S_BEGIN: \n"
" "INIT_S" \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Lasum_kernel_S1 \n"
" ble 7f //asum_kernel_S1 \n"
".Lasum_kernel_S4: \n"
"6: //asum_kernel_S4: \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_S4 \n"
" bne 6b //asum_kernel_S4 \n"
".Lasum_kernel_S1: \n"
"7: //asum_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
".Lasum_kernel_S10: \n"
"8: //asum_kernel_S10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_S10 \n"
" bne 8b //asum_kernel_S10 \n"
".Lasum_kernel_L999: \n"
"9: //asum_kernel_L999: \n"
" fmov %[ASUM_], "SUMFD" \n"
: [ASUM_] "=r" (asum) //%0

44
kernel/arm64/copy_thunderx2t99.c
View File

@@ -90,62 +90,62 @@ static int do_copy(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_
" mov "Y", %[Y_] \n"
" mov "INC_Y", %[INCY_] \n"
" cmp "N", xzr \n"
" ble .Lcopy_kernel_L999 \n"
" ble 8f //copy_kernel_L999 \n"
" cmp "INC_X", #1 \n"
" bne .Lcopy_kernel_S_BEGIN \n"
" bne 4f //copy_kernel_S_BEGIN \n"
" cmp "INC_Y", #1 \n"
" bne .Lcopy_kernel_S_BEGIN \n"
" bne 4f //copy_kernel_S_BEGIN \n"
".Lcopy_kernel_F_BEGIN: \n"
"// .Lcopy_kernel_F_BEGIN: \n"
" "INIT" \n"
" asr "J", "N", #"N_DIV_SHIFT" \n"
" cmp "J", xzr \n"
" beq .Lcopy_kernel_F1 \n"
" beq 2f //copy_kernel_F1 \n"
" .align 5 \n"
".Lcopy_kernel_F: \n"
"1: //copy_kernel_F: \n"
" "KERNEL_F" \n"
" subs "J", "J", #1 \n"
" bne .Lcopy_kernel_F \n"
" bne 1b //copy_kernel_F \n"
".Lcopy_kernel_F1: \n"
"2: //copy_kernel_F1: \n"
#if defined(COMPLEX) && defined(DOUBLE)
" b .Lcopy_kernel_L999 \n"
" b 8f //copy_kernel_L999 \n"
#else
" ands "J", "N", #"N_REM_MASK" \n"
" ble .Lcopy_kernel_L999 \n"
" ble 8f //copy_kernel_L999 \n"
#endif
".Lcopy_kernel_F10: \n"
"3: //copy_kernel_F10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Lcopy_kernel_F10 \n"
" b .Lcopy_kernel_L999 \n"
" bne 3b //copy_kernel_F10 \n"
" b 8f //copy_kernel_L999 \n"
".Lcopy_kernel_S_BEGIN: \n"
"4: //copy_kernel_S_BEGIN: \n"
" "INIT" \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Lcopy_kernel_S1 \n"
" ble 6f //copy_kernel_S1 \n"
".Lcopy_kernel_S4: \n"
"5: //copy_kernel_S4: \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Lcopy_kernel_S4 \n"
" bne 5b //copy_kernel_S4 \n"
".Lcopy_kernel_S1: \n"
"6: //copy_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Lcopy_kernel_L999 \n"
" ble 8f //copy_kernel_L999 \n"
".Lcopy_kernel_S10: \n"
"7: //copy_kernel_S10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Lcopy_kernel_S10 \n"
" bne 7b //copy_kernel_S10 \n"
".Lcopy_kernel_L999: \n"
"8: //copy_kernel_L999: \n"
:
: [N_] "r" (n), //%1

42
kernel/arm64/dasum_thunderx2t99.c
View File

@@ -141,58 +141,58 @@ static FLOAT dasum_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
" fmov d6, "REG0" \n"
" fmov d7, "REG0" \n"
" cmp "N", xzr \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
" cmp "INC_X", xzr \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
" cmp "INC_X", #1 \n"
" bne .Lasum_kernel_S_BEGIN \n"
" bne 5f //asum_kernel_S_BEGIN \n"
".Lasum_kernel_F_BEGIN: \n"
"1: //asum_kernel_F_BEGIN: \n"
" asr "J", "N", #5 \n"
" cmp "J", xzr \n"
" beq .Lasum_kernel_F1 \n"
" beq 3f //asum_kernel_F1 \n"
".align 5 \n"
".Lasum_kernel_F32: \n"
"2: //asum_kernel_F32: \n"
" "KERNEL_F32" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_F32 \n"
" bne 2b //asum_kernel_F32 \n"
" "KERNEL_F32_FINALIZE" \n"
".Lasum_kernel_F1: \n"
"3: //asum_kernel_F1: \n"
" ands "J", "N", #31 \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
".Lasum_kernel_F10: \n"
"4: //asum_kernel_F10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_F10 \n"
" b .Lasum_kernel_L999 \n"
" bne 4b //asum_kernel_F10 \n"
" b 9f //asum_kernel_L999 \n"
".Lasum_kernel_S_BEGIN: \n"
"5: //asum_kernel_S_BEGIN: \n"
" "INIT_S" \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Lasum_kernel_S1 \n"
" ble 7f //asum_kernel_S1 \n"
".Lasum_kernel_S4: \n"
"6: //asum_kernel_S4: \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_S4 \n"
" bne 6b //asum_kernel_S4 \n"
".Lasum_kernel_S1: \n"
"7: //asum_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
".Lasum_kernel_S10: \n"
"8: //asum_kernel_S10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_S10 \n"
" bne 8b //asum_kernel_S10 \n"
".Lasum_kernel_L999: \n"
"9: //asum_kernel_L999: \n"
" fmov %[ASUM_], "SUMF" \n"
: [ASUM_] "=r" (asum) //%0

42
kernel/arm64/dot_thunderx2t99.c
View File

@@ -291,61 +291,61 @@ static RETURN_TYPE dot_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, B
" fmov d6, xzr \n"
" fmov d7, xzr \n"
" cmp "N", xzr \n"
" ble .Ldot_kernel_L999 \n"
" ble 9f //dot_kernel_L999 \n"
" cmp "INC_X", #1 \n"
" bne .Ldot_kernel_S_BEGIN \n"
" bne 5f //dot_kernel_S_BEGIN \n"
" cmp "INC_Y", #1 \n"
" bne .Ldot_kernel_S_BEGIN \n"
" bne 5f //dot_kernel_S_BEGIN \n"
".Ldot_kernel_F_BEGIN: \n"
"1: //dot_kernel_F_BEGIN: \n"
" lsl "INC_X", "INC_X", "INC_SHIFT" \n"
" lsl "INC_Y", "INC_Y", "INC_SHIFT" \n"
" asr "J", "N", #"N_DIV_SHIFT" \n"
" cmp "J", xzr \n"
" beq .Ldot_kernel_F1 \n"
" beq 3f //dot_kernel_F1 \n"
" .align 5 \n"
".Ldot_kernel_F: \n"
"2: //dot_kernel_F: \n"
" "KERNEL_F" \n"
" subs "J", "J", #1 \n"
" bne .Ldot_kernel_F \n"
" bne 2b //dot_kernel_F \n"
" "KERNEL_F_FINALIZE" \n"
".Ldot_kernel_F1: \n"
"3: //dot_kernel_F1: \n"
" ands "J", "N", #"N_REM_MASK" \n"
" ble .Ldot_kernel_L999 \n"
" ble 9f //dot_kernel_L999 \n"
".Ldot_kernel_F10: \n"
"4: //dot_kernel_F10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Ldot_kernel_F10 \n"
" b .Ldot_kernel_L999 \n"
" bne 4b //dot_kernel_F10 \n"
" b 9f //dot_kernel_L999 \n"
".Ldot_kernel_S_BEGIN: \n"
"5: //dot_kernel_S_BEGIN: \n"
" lsl "INC_X", "INC_X", "INC_SHIFT" \n"
" lsl "INC_Y", "INC_Y", "INC_SHIFT" \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Ldot_kernel_S1 \n"
" ble 7f //dot_kernel_S1 \n"
".Ldot_kernel_S4: \n"
"6: //dot_kernel_S4: \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Ldot_kernel_S4 \n"
" bne 6b //dot_kernel_S4 \n"
".Ldot_kernel_S1: \n"
"7: //dot_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Ldot_kernel_L999 \n"
" ble 9f //dot_kernel_L999 \n"
".Ldot_kernel_S10: \n"
"8: //dot_kernel_S10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Ldot_kernel_S10 \n"
" bne 8b //dot_kernel_S10 \n"
".Ldot_kernel_L999: \n"
"9: //dot_kernel_L999: \n"
" str "DOTF", [%[DOT_]] \n"
:

44
kernel/arm64/dznrm2_thunderx2t99.c
View File

@@ -74,33 +74,33 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
" fmov "SCALE", xzr \n"
" fmov "SSQ", #1.0 \n"
" cmp "N", xzr \n"
" ble .Lnrm2_kernel_L999 \n"
" ble 9f //nrm2_kernel_L999 \n"
" cmp "INC_X", xzr \n"
" ble .Lnrm2_kernel_L999 \n"
" ble 9f //nrm2_kernel_L999 \n"
".Lnrm2_kernel_F_BEGIN: \n"
"1: //nrm2_kernel_F_BEGIN: \n"
" fmov "REGZERO", xzr \n"
" fmov "REGONE", #1.0 \n"
" lsl "INC_X", "INC_X", #"INC_SHIFT" \n"
" mov "J", "N" \n"
" cmp "J", xzr \n"
" beq .Lnrm2_kernel_L999 \n"
" beq 9f //nrm2_kernel_L999 \n"
".Lnrm2_kernel_F_ZERO_SKIP: \n"
"2: //nrm2_kernel_F_ZERO_SKIP: \n"
" ldr d4, ["X"] \n"
" fcmp d4, "REGZERO" \n"
" bne .Lnrm2_kernel_F_INIT \n"
" bne 3f //nrm2_kernel_F_INIT \n"
#if defined(COMPLEX)
" ldr d4, ["X", #8] \n"
" fcmp d4, "REGZERO" \n"
" bne .Lnrm2_kernel_F_INIT_I \n"
" bne 4f //nrm2_kernel_F_INIT_I \n"
#endif
" add "X", "X", "INC_X" \n"
" subs "J", "J", #1 \n"
" beq .Lnrm2_kernel_L999 \n"
" b .Lnrm2_kernel_F_ZERO_SKIP \n"
" beq 9f //nrm2_kernel_L999 \n"
" b 2b //nrm2_kernel_F_ZERO_SKIP \n"
".Lnrm2_kernel_F_INIT: \n"
"3: //nrm2_kernel_F_INIT: \n"
" ldr d4, ["X"] \n"
" fabs d4, d4 \n"
" fmax "CUR_MAX", "SCALE", d4 \n"
@@ -112,7 +112,7 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
" fadd "SSQ", "SSQ", d4 \n"
" fmov "SCALE", "CUR_MAX" \n"
#if defined(COMPLEX)
".Lnrm2_kernel_F_INIT_I: \n"
"4: //nrm2_kernel_F_INIT_I: \n"
" ldr d3, ["X", #8] \n"
" fabs d3, d3 \n"
" fmax "CUR_MAX", "SCALE", d3 \n"
@@ -126,16 +126,16 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
#endif
" add "X", "X", "INC_X" \n"
" subs "J", "J", #1 \n"
" beq .Lnrm2_kernel_L999 \n"
" beq 9f //nrm2_kernel_L999 \n"
".Lnrm2_kernel_F_START: \n"
"5: //nrm2_kernel_F_START: \n"
" cmp "INC_X", #"SZ" \n"
" bne .Lnrm2_kernel_F1 \n"
" bne 8f //nrm2_kernel_F1 \n"
" asr "K", "J", #4 \n"
" cmp "K", xzr \n"
" beq .Lnrm2_kernel_F1 \n"
" beq 8f //nrm2_kernel_F1 \n"
".Lnrm2_kernel_F: \n"
"6: //nrm2_kernel_F: \n"
" ldp q16, q17, ["X"] \n"
" ldp q18, q19, ["X", #32] \n"
" ldp q20, q21, ["X", #64] \n"
@@ -255,13 +255,13 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
" fmov "SCALE", "CUR_MAX" \n"
#endif
" subs "K", "K", #1 \n"
" bne .Lnrm2_kernel_F \n"
" bne 6b //nrm2_kernel_F \n"
".Lnrm2_kernel_F_DONE: \n"
"7: //nrm2_kernel_F_DONE: \n"
" ands "J", "J", #15 \n"
" beq .Lnrm2_kernel_L999 \n"
" beq 9f //nrm2_kernel_L999 \n"
".Lnrm2_kernel_F1: \n"
"8: //nrm2_kernel_F1: \n"
" ldr d4, ["X"] \n"
" fabs d4, d4 \n"
" fmax "CUR_MAX", "SCALE", d4 \n"
@@ -286,9 +286,9 @@ static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
#endif
" add "X", "X", "INC_X" \n"
" subs "J", "J", #1 \n"
" bne .Lnrm2_kernel_F1 \n"
" bne 8b //nrm2_kernel_F1 \n"
".Lnrm2_kernel_L999: \n"
"9: //nrm2_kernel_L999: \n"
" str "SSQ", [%[SSQ_]] \n"
" str "SCALE", [%[SCALE_]] \n"

54
kernel/arm64/iamax_thunderx2t99.c
View File

@@ -208,7 +208,7 @@ extern int blas_level1_thread_with_return_value(int mode, BLASLONG m, BLASLONG n
#endif
static BLASLONG iamax_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
static BLASLONG __attribute__((noinline)) iamax_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
{
BLASLONG index = 0;
@@ -220,72 +220,72 @@ static BLASLONG iamax_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
" mov "INC_X", %[INCX_] \n"
" cmp "N", xzr \n"
" ble .Liamax_kernel_zero \n"
" ble 10f //iamax_kernel_zero \n"
" cmp "INC_X", xzr \n"
" ble .Liamax_kernel_zero \n"
" ble 10f //iamax_kernel_zero \n"
" cmp "INC_X", #1 \n"
" bne .Liamax_kernel_S_BEGIN \n"
" bne 5f //iamax_kernel_S_BEGIN \n"
" mov x7, "X" \n"
".Liamax_kernel_F_BEGIN: \n"
"1: //iamax_kernel_F_BEGIN: \n"
" "INIT" \n"
" subs "N", "N", #1 \n"
" ble .Liamax_kernel_L999 \n"
" ble 9f //iamax_kernel_L999 \n"
" asr "J", "N", #"N_DIV_SHIFT" \n"
" cmp "J", xzr \n"
" beq .Liamax_kernel_F1 \n"
" beq 3f //iamax_kernel_F1 \n"
" add "Z", "Z", #1 \n"
".Liamax_kernel_F: \n"
"2: //iamax_kernel_F: \n"
" "KERNEL_F" \n"
" subs "J", "J", #1 \n"
" bne .Liamax_kernel_F \n"
" bne 2b //iamax_kernel_F \n"
" "KERNEL_F_FINALIZE" \n"
" sub "Z", "Z", #1 \n"
".Liamax_kernel_F1: \n"
"3: //iamax_kernel_F1: \n"
" ands "J", "N", #"N_REM_MASK" \n"
" ble .Liamax_kernel_L999 \n"
" ble 9f //iamax_kernel_L999 \n"
".Liamax_kernel_F10: \n"
"4: //iamax_kernel_F10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Liamax_kernel_F10 \n"
" b .Liamax_kernel_L999 \n"
" bne 4b //iamax_kernel_F10 \n"
" b 9f //iamax_kernel_L999 \n"
".Liamax_kernel_S_BEGIN: \n"
"5: //iamax_kernel_S_BEGIN: \n"
" "INIT" \n"
" subs "N", "N", #1 \n"
" ble .Liamax_kernel_L999 \n"
" ble 9f //iamax_kernel_L999 \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Liamax_kernel_S1 \n"
" ble 7f //iamax_kernel_S1 \n"
".Liamax_kernel_S4: \n"
"6: //iamax_kernel_S4: \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Liamax_kernel_S4 \n"
" bne 6b //iamax_kernel_S4 \n"
".Liamax_kernel_S1: \n"
"7: //iamax_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Liamax_kernel_L999 \n"
" ble 9f //iamax_kernel_L999 \n"
".Liamax_kernel_S10: \n"
"8: //iamax_kernel_S10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Liamax_kernel_S10 \n"
" bne 8b //iamax_kernel_S10 \n"
".Liamax_kernel_L999: \n"
"9: //iamax_kernel_L999: \n"
" mov x0, "INDEX" \n"
" b .Liamax_kernel_DONE \n"
" b 11f //iamax_kernel_DONE \n"
".Liamax_kernel_zero: \n"
"10: //iamax_kernel_zero: \n"
" mov x0, xzr \n"
".Liamax_kernel_DONE: \n"
"11: //iamax_kernel_DONE: \n"
" mov %[INDEX_], "INDEX" \n"
: [INDEX_] "=r" (index) //%0

52
kernel/arm64/izamax_thunderx2t99.c
View File

@@ -229,72 +229,72 @@ static BLASLONG izamax_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
" mov "INC_X", %[INCX_] \n"
" cmp "N", xzr \n"
" ble .Lizamax_kernel_zero \n"
" ble 10f //izamax_kernel_zero \n"
" cmp "INC_X", xzr \n"
" ble .Lizamax_kernel_zero \n"
" ble 10f //izamax_kernel_zero \n"
" cmp "INC_X", #1 \n"
" bne .Lizamax_kernel_S_BEGIN \n"
" bne 5f //izamax_kernel_S_BEGIN \n"
" mov x7, "X" \n"
".Lizamax_kernel_F_BEGIN: \n"
"1: //izamax_kernel_F_BEGIN: \n"
" "INIT" \n"
" subs "N", "N", #1 \n"
" ble .Lizamax_kernel_L999 \n"
" ble 9f //izamax_kernel_L999 \n"
" asr "J", "N", #"N_DIV_SHIFT" \n"
" cmp "J", xzr \n"
" beq .Lizamax_kernel_F1 \n"
" beq 3f //izamax_kernel_F1 \n"
" add "Z", "Z", #1 \n"
".Lizamax_kernel_F: \n"
"2: //izamax_kernel_F: \n"
" "KERNEL_F" \n"
" subs "J", "J", #1 \n"
" bne .Lizamax_kernel_F \n"
" bne 2b //izamax_kernel_F \n"
" "KERNEL_F_FINALIZE" \n"
" sub "Z", "Z", #1 \n"
".Lizamax_kernel_F1: \n"
"3: //izamax_kernel_F1: \n"
" ands "J", "N", #"N_REM_MASK" \n"
" ble .Lizamax_kernel_L999 \n"
" ble 9f //izamax_kernel_L999 \n"
".Lizamax_kernel_F10: \n"
"4: //izamax_kernel_F10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lizamax_kernel_F10 \n"
" b .Lizamax_kernel_L999 \n"
" bne 4b //izamax_kernel_F10 \n"
" b 9f //izamax_kernel_L999 \n"
".Lizamax_kernel_S_BEGIN: \n"
"5: //izamax_kernel_S_BEGIN: \n"
" "INIT" \n"
" subs "N", "N", #1 \n"
" ble .Lizamax_kernel_L999 \n"
" ble 9f //izamax_kernel_L999 \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Lizamax_kernel_S1 \n"
" ble 7f //izamax_kernel_S1 \n"
".Lizamax_kernel_S4: \n"
"6: //izamax_kernel_S4: \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lizamax_kernel_S4 \n"
" bne 6b //izamax_kernel_S4 \n"
".Lizamax_kernel_S1: \n"
"7: //izamax_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Lizamax_kernel_L999 \n"
" ble 9f //izamax_kernel_L999 \n"
".Lizamax_kernel_S10: \n"
"8: //izamax_kernel_S10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lizamax_kernel_S10 \n"
" bne 8b //izamax_kernel_S10 \n"
".Lizamax_kernel_L999: \n"
"9: //izamax_kernel_L999: \n"
" mov x0, "INDEX" \n"
" b .Lizamax_kernel_DONE \n"
" b 11f //izamax_kernel_DONE \n"
".Lizamax_kernel_zero: \n"
"10: //izamax_kernel_zero: \n"
" mov x0, xzr \n"
".Lizamax_kernel_DONE: \n"
"11: //izamax_kernel_DONE: \n"
" mov %[INDEX_], "INDEX" \n"
: [INDEX_] "=r" (index) //%0

42
kernel/arm64/sasum_thunderx2t99.c
View File

@@ -143,58 +143,58 @@ static FLOAT sasum_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
" fmov s6, "REG0" \n"
" fmov s7, "REG0" \n"
" cmp "N", xzr \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
" cmp "INC_X", xzr \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
" cmp "INC_X", #1 \n"
" bne .Lasum_kernel_S_BEGIN \n"
" bne 5f //asum_kernel_S_BEGIN \n"
".Lasum_kernel_F_BEGIN: \n"
"1: //asum_kernel_F_BEGIN: \n"
" asr "J", "N", #6 \n"
" cmp "J", xzr \n"
" beq .Lasum_kernel_F1 \n"
" beq 3f //asum_kernel_F1 \n"
".align 5 \n"
".Lasum_kernel_F64: \n"
"2: //asum_kernel_F64: \n"
" "KERNEL_F64" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_F64 \n"
" bne 2b //asum_kernel_F64 \n"
" "KERNEL_F64_FINALIZE" \n"
".Lasum_kernel_F1: \n"
"3: //asum_kernel_F1: \n"
" ands "J", "N", #63 \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
".Lasum_kernel_F10: \n"
"4: //asum_kernel_F10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_F10 \n"
" b .Lasum_kernel_L999 \n"
" bne 4b //asum_kernel_F10 \n"
" b 9f //asum_kernel_L999 \n"
".Lasum_kernel_S_BEGIN: \n"
"5: //asum_kernel_S_BEGIN: \n"
" "INIT_S" \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Lasum_kernel_S1 \n"
" ble 7f //asum_kernel_S1 \n"
".Lasum_kernel_S4: \n"
"6: //asum_kernel_S4: \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_S4 \n"
" bne 6b //asum_kernel_S4 \n"
".Lasum_kernel_S1: \n"
"7: //asum_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
".Lasum_kernel_S10: \n"
"8: //asum_kernel_S10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_S10 \n"
" bne 8b //asum_kernel_S10 \n"
".Lasum_kernel_L999: \n"
"9: //asum_kernel_L999: \n"
" fmov %[ASUM_], "SUMFD" \n"
: [ASUM_] "=r" (asum) //%0

42
kernel/arm64/scnrm2_thunderx2t99.c
View File

@@ -227,58 +227,58 @@ static double nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
" fmov d6, xzr \n"
" fmov d7, xzr \n"
" cmp "N", xzr \n"
" ble .Lnrm2_kernel_L999 \n"
" ble 9f //nrm2_kernel_L999 \n"
" cmp "INC_X", xzr \n"
" ble .Lnrm2_kernel_L999 \n"
" ble 9f //nrm2_kernel_L999 \n"
" cmp "INC_X", #1 \n"
" bne .Lnrm2_kernel_S_BEGIN \n"
" bne 5f //nrm2_kernel_S_BEGIN \n"
".Lnrm2_kernel_F_BEGIN: \n"
"1: //nrm2_kernel_F_BEGIN: \n"
" asr "J", "N", #"N_DIV_SHIFT" \n"
" cmp "J", xzr \n"
" beq .Lnrm2_kernel_S_BEGIN \n"
" beq 5f //nrm2_kernel_S_BEGIN \n"
" .align 5 \n"
".Lnrm2_kernel_F: \n"
"2: //nrm2_kernel_F: \n"
" "KERNEL_F" \n"
" subs "J", "J", #1 \n"
" bne .Lnrm2_kernel_F \n"
" bne 2b //nrm2_kernel_F \n"
" "KERNEL_F_FINALIZE" \n"
".Lnrm2_kernel_F1: \n"
"3: //nrm2_kernel_F1: \n"
" ands "J", "N", #"N_REM_MASK" \n"
" ble .Lnrm2_kernel_L999 \n"
" ble 9f //nrm2_kernel_L999 \n"
".Lnrm2_kernel_F10: \n"
"4: //nrm2_kernel_F10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Lnrm2_kernel_F10 \n"
" b .Lnrm2_kernel_L999 \n"
" bne 4b //nrm2_kernel_F10 \n"
" b 9f //nrm2_kernel_L999 \n"
".Lnrm2_kernel_S_BEGIN: \n"
"5: //nrm2_kernel_S_BEGIN: \n"
" lsl "INC_X", "INC_X", #"INC_SHIFT" \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Lnrm2_kernel_S1 \n"
" ble 7f //nrm2_kernel_S1 \n"
".Lnrm2_kernel_S4: \n"
"6: //nrm2_kernel_S4: \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lnrm2_kernel_S4 \n"
" bne 6b //nrm2_kernel_S4 \n"
".Lnrm2_kernel_S1: \n"
"7: //nrm2_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Lnrm2_kernel_L999 \n"
" ble 9f //nrm2_kernel_L999 \n"
".Lnrm2_kernel_S10: \n"
"8: //nrm2_kernel_S10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lnrm2_kernel_S10 \n"
" bne 8b //nrm2_kernel_S10 \n"
".Lnrm2_kernel_L999: \n"
"9: //nrm2_kernel_L999: \n"
" "KERNEL_FINALIZE" \n"
" fmov %[RET_], "SSQD" \n"

42
kernel/arm64/zasum_thunderx2t99.c
View File

@@ -143,58 +143,58 @@ static FLOAT zasum_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x)
" fmov d6, "REG0" \n"
" fmov d7, "REG0" \n"
" cmp "N", xzr \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
" cmp "INC_X", xzr \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
" cmp "INC_X", #1 \n"
" bne .Lasum_kernel_S_BEGIN \n"
" bne 5f //asum_kernel_S_BEGIN \n"
".Lasum_kernel_F_BEGIN: \n"
"1: //asum_kernel_F_BEGIN: \n"
" asr "J", "N", #4 \n"
" cmp "J", xzr \n"
" beq .Lasum_kernel_F1 \n"
" beq 3f //asum_kernel_F1 \n"
".align 5 \n"
".Lasum_kernel_F16: \n"
"2: //asum_kernel_F16: \n"
" "KERNEL_F16" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_F16 \n"
" bne 2b //asum_kernel_F16 \n"
" "KERNEL_F16_FINALIZE" \n"
".Lasum_kernel_F1: \n"
"3: //asum_kernel_F1: \n"
" ands "J", "N", #15 \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
".Lasum_kernel_F10: \n"
"4: //asum_kernel_F10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_F10 \n"
" b .Lasum_kernel_L999 \n"
" bne 4b //asum_kernel_F10 \n"
" b 9f //asum_kernel_L999 \n"
".Lasum_kernel_S_BEGIN: \n"
"5: //asum_kernel_S_BEGIN: \n"
" "INIT_S" \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Lasum_kernel_S1 \n"
" ble 7f //asum_kernel_S1 \n"
".Lasum_kernel_S4: \n"
"6: //asum_kernel_S4: \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_S4 \n"
" bne 6b //asum_kernel_S4 \n"
".Lasum_kernel_S1: \n"
"7: //asum_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Lasum_kernel_L999 \n"
" ble 9f //asum_kernel_L999 \n"
".Lasum_kernel_S10: \n"
"8: //asum_kernel_S10: \n"
" "KERNEL_S1" \n"
" subs "J", "J", #1 \n"
" bne .Lasum_kernel_S10 \n"
" bne 8b //asum_kernel_S10 \n"
".Lasum_kernel_L999: \n"
"9: //asum_kernel_L999: \n"
" fmov %[ASUM_], "SUMF" \n"
: [ASUM_] "=r" (asum) //%0

42
kernel/arm64/zdot_thunderx2t99.c
View File

@@ -218,61 +218,61 @@ static void zdot_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLON
" fmov d6, xzr \n"
" fmov d7, xzr \n"
" cmp "N", xzr \n"
" ble .Ldot_kernel_L999 \n"
" ble 9f //dot_kernel_L999 \n"
" cmp "INC_X", #1 \n"
" bne .Ldot_kernel_S_BEGIN \n"
" bne 5f //dot_kernel_S_BEGIN \n"
" cmp "INC_Y", #1 \n"
" bne .Ldot_kernel_S_BEGIN \n"
" bne 5f //dot_kernel_S_BEGIN \n"
".Ldot_kernel_F_BEGIN: \n"
"1: //dot_kernel_F_BEGIN: \n"
" lsl "INC_X", "INC_X", "INC_SHIFT" \n"
" lsl "INC_Y", "INC_Y", "INC_SHIFT" \n"
" asr "J", "N", #"N_DIV_SHIFT" \n"
" cmp "J", xzr \n"
" beq .Ldot_kernel_F1 \n"
" beq 3f //dot_kernel_F1 \n"
" .align 5 \n"
".Ldot_kernel_F: \n"
"2: //dot_kernel_F: \n"
" "KERNEL_F" \n"
" subs "J", "J", #1 \n"
" bne .Ldot_kernel_F \n"
" bne 2b //dot_kernel_F \n"
" "KERNEL_F_FINALIZE" \n"
".Ldot_kernel_F1: \n"
"3: //dot_kernel_F1: \n"
" ands "J", "N", #"N_REM_MASK" \n"
" ble .Ldot_kernel_L999 \n"
" ble 9f //dot_kernel_L999 \n"
".Ldot_kernel_F10: \n"
"4: //dot_kernel_F10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Ldot_kernel_F10 \n"
" b .Ldot_kernel_L999 \n"
" bne 4b //dot_kernel_F10 \n"
" b 9f //dot_kernel_L999 \n"
".Ldot_kernel_S_BEGIN: \n"
"5: //dot_kernel_S_BEGIN: \n"
" lsl "INC_X", "INC_X", "INC_SHIFT" \n"
" lsl "INC_Y", "INC_Y", "INC_SHIFT" \n"
" asr "J", "N", #2 \n"
" cmp "J", xzr \n"
" ble .Ldot_kernel_S1 \n"
" ble 7f //dot_kernel_S1 \n"
".Ldot_kernel_S4: \n"
"6: //dot_kernel_S4: \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Ldot_kernel_S4 \n"
" bne 6b //dot_kernel_S4 \n"
".Ldot_kernel_S1: \n"
"7: //dot_kernel_S1: \n"
" ands "J", "N", #3 \n"
" ble .Ldot_kernel_L999 \n"
" ble 9f //dot_kernel_L999 \n"
".Ldot_kernel_S10: \n"
"8: //dot_kernel_S10: \n"
" "KERNEL_F1" \n"
" subs "J", "J", #1 \n"
" bne .Ldot_kernel_S10 \n"
" bne 8b //dot_kernel_S10 \n"
".Ldot_kernel_L999: \n"
"9: //dot_kernel_L999: \n"
" str "DOTF", [%[DOTR_]] \n"
" str "DOTI", [%[DOTI_]] \n"

9
kernel/x86_64/cdot.c
View File

@@ -91,16 +91,15 @@ static void cdot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
#endif
FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
{
BLASLONG i;
BLASLONG ix,iy;
FLOAT _Complex result;
FLOAT dot[8] = { 0.0, 0.0, 0.0 , 0.0, 0.0, 0.0, 0.0, 0.0 } ;
if ( n <= 0 )
{
result = OPENBLAS_MAKE_COMPLEX_FLOAT (0.0, 0.0) ;
OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (0.0, 0.0) ;
return(result);
}
@@ -160,11 +159,11 @@ FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG in
}
#if !defined(CONJ)
result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]-dot[1], dot[4]+dot[5]) ;
OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]-dot[1], dot[4]+dot[5]) ;
// CREAL(result) = dot[0] - dot[1];
// CIMAG(result) = dot[4] + dot[5];
#else
result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]+dot[1], dot[4]-dot[5]) ;
OPENBLAS_COMPLEX_FLOAT result = OPENBLAS_MAKE_COMPLEX_FLOAT (dot[0]+dot[1], dot[4]-dot[5]) ;
// CREAL(result) = dot[0] + dot[1];
// CIMAG(result) = dot[4] - dot[5];

9
kernel/x86_64/zdot.c
View File

@@ -86,18 +86,17 @@ static void zdot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d)
#endif
FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
{
BLASLONG i;
BLASLONG ix,iy;
FLOAT _Complex result;
FLOAT dot[4] = { 0.0, 0.0, 0.0 , 0.0 } ;
if ( n <= 0 )
{
// CREAL(result) = 0.0 ;
// CIMAG(result) = 0.0 ;
result=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
return(result);
}
@@ -151,11 +150,11 @@ FLOAT _Complex CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG in
}
#if !defined(CONJ)
result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]);
OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]-dot[1],dot[2]+dot[3]);
// CREAL(result) = dot[0] - dot[1];
// CIMAG(result) = dot[2] + dot[3];
#else
result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]);
OPENBLAS_COMPLEX_FLOAT result=OPENBLAS_MAKE_COMPLEX_FLOAT(dot[0]+dot[1],dot[2]-dot[3]);
// CREAL(result) = dot[0] + dot[1];
// CIMAG(result) = dot[2] - dot[3];