* [Patch 0/4] math-emu @ 2008-10-28 3:50 Liu Yu 2008-10-28 3:50 ` [PATCH 1/4] math-emu: Fix single float point division bug Liu Yu 0 siblings, 1 reply; 9+ messages in thread From: Liu Yu @ 2008-10-28 3:50 UTC (permalink / raw) To: linuxppc-dev; +Cc: kumar.gala These patches are related to math-emu. [Patch 1/4] Fix a fdiv bug of FP math-emu [Patch 2/4] Update a macro to new version [Patch 3/4] Remove redundant code [Patch 4/4] Add SPE/EFP handlers for E500v1/v2 processors ^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH 1/4] math-emu: Fix single float point division bug 2008-10-28 3:50 [Patch 0/4] math-emu Liu Yu @ 2008-10-28 3:50 ` Liu Yu 2008-10-28 3:50 ` [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN Liu Yu 2008-11-06 4:03 ` [PATCH 1/4] math-emu: Fix single float point division bug Kumar Gala 0 siblings, 2 replies; 9+ messages in thread From: Liu Yu @ 2008-10-28 3:50 UTC (permalink / raw) To: linuxppc-dev; +Cc: kumar.gala, Liu Yu PowerPC float point division emulation is derived from gcc. I reported this problem on gcc maillist and got this reply: http://gcc.gnu.org/ml/gcc/2008-03/msg00543.html Since UDIV_NEEDS_NORMALIZATION is not used by kernel, we should use _FP_DIV_MEAT_1_udiv_norm to make sure the single float point is normalized before udiv_qrnnd. Signed-off-by: Liu Yu <yu.liu@freescale.com> --- arch/powerpc/include/asm/sfp-machine.h | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/arch/powerpc/include/asm/sfp-machine.h b/arch/powerpc/include/asm/sfp-machine.h index ced34f1..da12ea7 100644 --- a/arch/powerpc/include/asm/sfp-machine.h +++ b/arch/powerpc/include/asm/sfp-machine.h @@ -82,7 +82,7 @@ #define _FP_MUL_MEAT_S(R,X,Y) _FP_MUL_MEAT_1_wide(_FP_WFRACBITS_S,R,X,Y,umul_ppmm) #define _FP_MUL_MEAT_D(R,X,Y) _FP_MUL_MEAT_2_wide(_FP_WFRACBITS_D,R,X,Y,umul_ppmm) -#define _FP_DIV_MEAT_S(R,X,Y) _FP_DIV_MEAT_1_udiv(S,R,X,Y) +#define _FP_DIV_MEAT_S(R,X,Y) _FP_DIV_MEAT_1_udiv_norm(S,R,X,Y) #define _FP_DIV_MEAT_D(R,X,Y) _FP_DIV_MEAT_2_udiv(D,R,X,Y) /* These macros define what NaN looks like. They're supposed to expand to -- 1.5.4 ^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN 2008-10-28 3:50 ` [PATCH 1/4] math-emu: Fix single float point division bug Liu Yu @ 2008-10-28 3:50 ` Liu Yu 2008-10-28 3:50 ` [PATCH 3/4] math-emu: Remove redundant 'ret' Liu Yu 2008-11-06 4:05 ` [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN Kumar Gala 2008-11-06 4:03 ` [PATCH 1/4] math-emu: Fix single float point division bug Kumar Gala 1 sibling, 2 replies; 9+ messages in thread From: Liu Yu @ 2008-10-28 3:50 UTC (permalink / raw) To: linuxppc-dev; +Cc: kumar.gala, Liu Yu Move to using the same macro definition for _FP_CHOOSENAN as s390, sh, sparc32/64. The original author didn't understand this and matched what sparc64 was doing and they have updated to this definition. Signed-off-by: Liu Yu <yu.liu@freescale.com> --- arch/powerpc/include/asm/sfp-machine.h | 26 +++++++++++++++++--------- 1 files changed, 17 insertions(+), 9 deletions(-) diff --git a/arch/powerpc/include/asm/sfp-machine.h b/arch/powerpc/include/asm/sfp-machine.h index da12ea7..88af036 100644 --- a/arch/powerpc/include/asm/sfp-machine.h +++ b/arch/powerpc/include/asm/sfp-machine.h @@ -111,16 +111,24 @@ #define FP_EX_DIVZERO (1 << (31 - 5)) #define FP_EX_INEXACT (1 << (31 - 6)) -/* This macro appears to be called when both X and Y are NaNs, and - * has to choose one and copy it to R. i386 goes for the larger of the - * two, sparc64 just picks Y. I don't understand this at all so I'll - * go with sparc64 because it's shorter :-> -- PMM +/* + * If one NaN is signaling and the other is not, + * we choose that one, otherwise we choose X. */ -#define _FP_CHOOSENAN(fs, wc, R, X, Y, OP) \ - do { \ - R##_s = Y##_s; \ - _FP_FRAC_COPY_##wc(R,Y); \ - R##_c = FP_CLS_NAN; \ +#define _FP_CHOOSENAN(fs, wc, R, X, Y, OP) \ + do { \ + if ((_FP_FRAC_HIGH_RAW_##fs(Y) & _FP_QNANBIT_##fs) \ + && !(_FP_FRAC_HIGH_RAW_##fs(X) & _FP_QNANBIT_##fs)) \ + { \ + R##_s = X##_s; \ + _FP_FRAC_COPY_##wc(R,X); \ + } \ + else \ + { \ + R##_s = Y##_s; \ + _FP_FRAC_COPY_##wc(R,Y); \ + } \ + R##_c = FP_CLS_NAN; \ } while (0) -- 1.5.4 ^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCH 3/4] math-emu: Remove redundant 'ret' 2008-10-28 3:50 ` [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN Liu Yu @ 2008-10-28 3:50 ` Liu Yu 2008-10-28 3:50 ` [PATCH 4/4] Add SPE/EFP math emulation for E500v1/v2 processors Liu Yu 2008-11-06 4:05 ` [PATCH 3/4] math-emu: Remove redundant 'ret' Kumar Gala 2008-11-06 4:05 ` [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN Kumar Gala 1 sibling, 2 replies; 9+ messages in thread From: Liu Yu @ 2008-10-28 3:50 UTC (permalink / raw) To: linuxppc-dev; +Cc: kumar.gala, Liu Yu FP_DECL_EX is already used, so ret is redundant. And FP_SET_EXCEPTION will add status into return value. Signed-off-by: Liu Yu <yu.liu@freescale.com> --- arch/powerpc/math-emu/fadd.c | 1 - arch/powerpc/math-emu/fcmpo.c | 5 ++--- arch/powerpc/math-emu/fdiv.c | 9 ++++----- arch/powerpc/math-emu/fdivs.c | 9 ++++----- arch/powerpc/math-emu/fmadd.c | 5 ++--- arch/powerpc/math-emu/fmadds.c | 5 ++--- arch/powerpc/math-emu/fmsub.c | 5 ++--- arch/powerpc/math-emu/fmsubs.c | 5 ++--- arch/powerpc/math-emu/fmul.c | 3 +-- arch/powerpc/math-emu/fmuls.c | 3 +-- arch/powerpc/math-emu/fnmadd.c | 5 ++--- arch/powerpc/math-emu/fnmadds.c | 5 ++--- arch/powerpc/math-emu/fnmsub.c | 5 ++--- arch/powerpc/math-emu/fnmsubs.c | 5 ++--- arch/powerpc/math-emu/fsqrt.c | 5 ++--- arch/powerpc/math-emu/fsqrts.c | 5 ++--- arch/powerpc/math-emu/fsub.c | 3 +-- arch/powerpc/math-emu/fsubs.c | 3 +-- 18 files changed, 34 insertions(+), 52 deletions(-) diff --git a/arch/powerpc/math-emu/fadd.c b/arch/powerpc/math-emu/fadd.c index 04d3b4a..0158a16 100644 --- a/arch/powerpc/math-emu/fadd.c +++ b/arch/powerpc/math-emu/fadd.c @@ -13,7 +13,6 @@ fadd(void *frD, void *frA, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p\n", __func__, frD, frA, frB); diff --git a/arch/powerpc/math-emu/fcmpo.c b/arch/powerpc/math-emu/fcmpo.c index b5dc449..5bce011 100644 --- a/arch/powerpc/math-emu/fcmpo.c +++ b/arch/powerpc/math-emu/fcmpo.c @@ -14,7 +14,6 @@ fcmpo(u32 *ccr, int crfD, void *frA, void *frB) FP_DECL_EX; int code[4] = { (1 << 3), (1 << 1), (1 << 2), (1 << 0) }; long cmp; - int ret = 0; #ifdef DEBUG printk("%s: %p (%08x) %d %p %p\n", __func__, ccr, *ccr, crfD, frA, frB); @@ -29,7 +28,7 @@ fcmpo(u32 *ccr, int crfD, void *frA, void *frB) #endif if (A_c == FP_CLS_NAN || B_c == FP_CLS_NAN) - ret |= EFLAG_VXVC; + FP_SET_EXCEPTION(EFLAG_VXVC); FP_CMP_D(cmp, A, B, 2); cmp = code[(cmp + 1) & 3]; @@ -44,5 +43,5 @@ fcmpo(u32 *ccr, int crfD, void *frA, void *frB) printk("CR: %08x\n", *ccr); #endif - return ret; + return FP_CUR_EXCEPTIONS; } diff --git a/arch/powerpc/math-emu/fdiv.c b/arch/powerpc/math-emu/fdiv.c index 2db1509..a29239c 100644 --- a/arch/powerpc/math-emu/fdiv.c +++ b/arch/powerpc/math-emu/fdiv.c @@ -13,7 +13,6 @@ fdiv(void *frD, void *frA, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p\n", __func__, frD, frA, frB); @@ -28,22 +27,22 @@ fdiv(void *frD, void *frA, void *frB) #endif if (A_c == FP_CLS_ZERO && B_c == FP_CLS_ZERO) { - ret |= EFLAG_VXZDZ; + FP_SET_EXCEPTION(EFLAG_VXZDZ); #ifdef DEBUG printk("%s: FPSCR_VXZDZ raised\n", __func__); #endif } if (A_c == FP_CLS_INF && B_c == FP_CLS_INF) { - ret |= EFLAG_VXIDI; + FP_SET_EXCEPTION(EFLAG_VXIDI); #ifdef DEBUG printk("%s: FPSCR_VXIDI raised\n", __func__); #endif } if (B_c == FP_CLS_ZERO && A_c != FP_CLS_ZERO) { - ret |= EFLAG_DIVZERO; + FP_SET_EXCEPTION(EFLAG_DIVZERO); if (__FPU_TRAP_P(EFLAG_DIVZERO)) - return ret; + return FP_CUR_EXCEPTIONS; } FP_DIV_D(R, A, B); diff --git a/arch/powerpc/math-emu/fdivs.c b/arch/powerpc/math-emu/fdivs.c index 797f6a9..526bc26 100644 --- a/arch/powerpc/math-emu/fdivs.c +++ b/arch/powerpc/math-emu/fdivs.c @@ -14,7 +14,6 @@ fdivs(void *frD, void *frA, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p\n", __func__, frD, frA, frB); @@ -29,22 +28,22 @@ fdivs(void *frD, void *frA, void *frB) #endif if (A_c == FP_CLS_ZERO && B_c == FP_CLS_ZERO) { - ret |= EFLAG_VXZDZ; + FP_SET_EXCEPTION(EFLAG_VXZDZ); #ifdef DEBUG printk("%s: FPSCR_VXZDZ raised\n", __func__); #endif } if (A_c == FP_CLS_INF && B_c == FP_CLS_INF) { - ret |= EFLAG_VXIDI; + FP_SET_EXCEPTION(EFLAG_VXIDI); #ifdef DEBUG printk("%s: FPSCR_VXIDI raised\n", __func__); #endif } if (B_c == FP_CLS_ZERO && A_c != FP_CLS_ZERO) { - ret |= EFLAG_DIVZERO; + FP_SET_EXCEPTION(EFLAG_DIVZERO); if (__FPU_TRAP_P(EFLAG_DIVZERO)) - return ret; + return FP_CUR_EXCEPTIONS; } FP_DIV_D(R, A, B); diff --git a/arch/powerpc/math-emu/fmadd.c b/arch/powerpc/math-emu/fmadd.c index 925313a..8c3f20a 100644 --- a/arch/powerpc/math-emu/fmadd.c +++ b/arch/powerpc/math-emu/fmadd.c @@ -15,7 +15,6 @@ fmadd(void *frD, void *frA, void *frB, void *frC) FP_DECL_D(C); FP_DECL_D(T); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frA, frB, frC); @@ -33,12 +32,12 @@ fmadd(void *frD, void *frA, void *frB, void *frC) if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(T, A, C); if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, T, B); diff --git a/arch/powerpc/math-emu/fmadds.c b/arch/powerpc/math-emu/fmadds.c index aea80ef..794fb31 100644 --- a/arch/powerpc/math-emu/fmadds.c +++ b/arch/powerpc/math-emu/fmadds.c @@ -16,7 +16,6 @@ fmadds(void *frD, void *frA, void *frB, void *frC) FP_DECL_D(C); FP_DECL_D(T); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frA, frB, frC); @@ -34,12 +33,12 @@ fmadds(void *frD, void *frA, void *frB, void *frC) if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(T, A, C); if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, T, B); diff --git a/arch/powerpc/math-emu/fmsub.c b/arch/powerpc/math-emu/fmsub.c index a644d52..626f6fe 100644 --- a/arch/powerpc/math-emu/fmsub.c +++ b/arch/powerpc/math-emu/fmsub.c @@ -15,7 +15,6 @@ fmsub(void *frD, void *frA, void *frB, void *frC) FP_DECL_D(C); FP_DECL_D(T); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frA, frB, frC); @@ -33,7 +32,7 @@ fmsub(void *frD, void *frA, void *frB, void *frC) if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(T, A, C); @@ -41,7 +40,7 @@ fmsub(void *frD, void *frA, void *frB, void *frC) B_s ^= 1; if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, T, B); diff --git a/arch/powerpc/math-emu/fmsubs.c b/arch/powerpc/math-emu/fmsubs.c index 2fdeeb9..3425bc8 100644 --- a/arch/powerpc/math-emu/fmsubs.c +++ b/arch/powerpc/math-emu/fmsubs.c @@ -16,7 +16,6 @@ fmsubs(void *frD, void *frA, void *frB, void *frC) FP_DECL_D(C); FP_DECL_D(T); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frA, frB, frC); @@ -34,7 +33,7 @@ fmsubs(void *frD, void *frA, void *frB, void *frC) if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(T, A, C); @@ -42,7 +41,7 @@ fmsubs(void *frD, void *frA, void *frB, void *frC) B_s ^= 1; if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, T, B); diff --git a/arch/powerpc/math-emu/fmul.c b/arch/powerpc/math-emu/fmul.c index 391fd17..2c19297 100644 --- a/arch/powerpc/math-emu/fmul.c +++ b/arch/powerpc/math-emu/fmul.c @@ -13,7 +13,6 @@ fmul(void *frD, void *frA, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p\n", __func__, frD, frA, frB); @@ -31,7 +30,7 @@ fmul(void *frD, void *frA, void *frB) if ((A_c == FP_CLS_INF && B_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && B_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(R, A, B); diff --git a/arch/powerpc/math-emu/fmuls.c b/arch/powerpc/math-emu/fmuls.c index 2d3ec5f..f5ad5c9 100644 --- a/arch/powerpc/math-emu/fmuls.c +++ b/arch/powerpc/math-emu/fmuls.c @@ -14,7 +14,6 @@ fmuls(void *frD, void *frA, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p\n", __func__, frD, frA, frB); @@ -32,7 +31,7 @@ fmuls(void *frD, void *frA, void *frB) if ((A_c == FP_CLS_INF && B_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && B_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(R, A, B); diff --git a/arch/powerpc/math-emu/fnmadd.c b/arch/powerpc/math-emu/fnmadd.c index 2497b86..e817bc5 100644 --- a/arch/powerpc/math-emu/fnmadd.c +++ b/arch/powerpc/math-emu/fnmadd.c @@ -15,7 +15,6 @@ fnmadd(void *frD, void *frA, void *frB, void *frC) FP_DECL_D(C); FP_DECL_D(T); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frA, frB, frC); @@ -33,12 +32,12 @@ fnmadd(void *frD, void *frA, void *frB, void *frC) if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(T, A, C); if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, T, B); diff --git a/arch/powerpc/math-emu/fnmadds.c b/arch/powerpc/math-emu/fnmadds.c index ee9d71e..4db4b7d 100644 --- a/arch/powerpc/math-emu/fnmadds.c +++ b/arch/powerpc/math-emu/fnmadds.c @@ -16,7 +16,6 @@ fnmadds(void *frD, void *frA, void *frB, void *frC) FP_DECL_D(C); FP_DECL_D(T); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frA, frB, frC); @@ -34,12 +33,12 @@ fnmadds(void *frD, void *frA, void *frB, void *frC) if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(T, A, C); if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, T, B); diff --git a/arch/powerpc/math-emu/fnmsub.c b/arch/powerpc/math-emu/fnmsub.c index 3885a77..f65979f 100644 --- a/arch/powerpc/math-emu/fnmsub.c +++ b/arch/powerpc/math-emu/fnmsub.c @@ -15,7 +15,6 @@ fnmsub(void *frD, void *frA, void *frB, void *frC) FP_DECL_D(C); FP_DECL_D(T); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frA, frB, frC); @@ -33,7 +32,7 @@ fnmsub(void *frD, void *frA, void *frB, void *frC) if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(T, A, C); @@ -41,7 +40,7 @@ fnmsub(void *frD, void *frA, void *frB, void *frC) B_s ^= 1; if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, T, B); diff --git a/arch/powerpc/math-emu/fnmsubs.c b/arch/powerpc/math-emu/fnmsubs.c index f835dfe..9021dac 100644 --- a/arch/powerpc/math-emu/fnmsubs.c +++ b/arch/powerpc/math-emu/fnmsubs.c @@ -16,7 +16,6 @@ fnmsubs(void *frD, void *frA, void *frB, void *frC) FP_DECL_D(C); FP_DECL_D(T); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frA, frB, frC); @@ -34,7 +33,7 @@ fnmsubs(void *frD, void *frA, void *frB, void *frC) if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) || (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF)) - ret |= EFLAG_VXIMZ; + FP_SET_EXCEPTION(EFLAG_VXIMZ); FP_MUL_D(T, A, C); @@ -42,7 +41,7 @@ fnmsubs(void *frD, void *frA, void *frB, void *frC) B_s ^= 1; if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, T, B); diff --git a/arch/powerpc/math-emu/fsqrt.c b/arch/powerpc/math-emu/fsqrt.c index 3e90072..a55fc7d 100644 --- a/arch/powerpc/math-emu/fsqrt.c +++ b/arch/powerpc/math-emu/fsqrt.c @@ -12,7 +12,6 @@ fsqrt(void *frD, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frB); @@ -25,9 +24,9 @@ fsqrt(void *frD, void *frB) #endif if (B_s && B_c != FP_CLS_ZERO) - ret |= EFLAG_VXSQRT; + FP_SET_EXCEPTION(EFLAG_VXSQRT); if (B_c == FP_CLS_NAN) - ret |= EFLAG_VXSNAN; + FP_SET_EXCEPTION(EFLAG_VXSNAN); FP_SQRT_D(R, B); diff --git a/arch/powerpc/math-emu/fsqrts.c b/arch/powerpc/math-emu/fsqrts.c index 2843be9..31dccbf 100644 --- a/arch/powerpc/math-emu/fsqrts.c +++ b/arch/powerpc/math-emu/fsqrts.c @@ -13,7 +13,6 @@ fsqrts(void *frD, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p %p\n", __func__, frD, frB); @@ -26,9 +25,9 @@ fsqrts(void *frD, void *frB) #endif if (B_s && B_c != FP_CLS_ZERO) - ret |= EFLAG_VXSQRT; + FP_SET_EXCEPTION(EFLAG_VXSQRT); if (B_c == FP_CLS_NAN) - ret |= EFLAG_VXSNAN; + FP_SET_EXCEPTION(EFLAG_VXSNAN); FP_SQRT_D(R, B); diff --git a/arch/powerpc/math-emu/fsub.c b/arch/powerpc/math-emu/fsub.c index 78b0944..02c5dff 100644 --- a/arch/powerpc/math-emu/fsub.c +++ b/arch/powerpc/math-emu/fsub.c @@ -13,7 +13,6 @@ fsub(void *frD, void *frA, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p\n", __func__, frD, frA, frB); @@ -31,7 +30,7 @@ fsub(void *frD, void *frA, void *frB) B_s ^= 1; if (A_s != B_s && A_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, A, B); diff --git a/arch/powerpc/math-emu/fsubs.c b/arch/powerpc/math-emu/fsubs.c index d3bf908..5d9b18c 100644 --- a/arch/powerpc/math-emu/fsubs.c +++ b/arch/powerpc/math-emu/fsubs.c @@ -14,7 +14,6 @@ fsubs(void *frD, void *frA, void *frB) FP_DECL_D(B); FP_DECL_D(R); FP_DECL_EX; - int ret = 0; #ifdef DEBUG printk("%s: %p %p %p\n", __func__, frD, frA, frB); @@ -32,7 +31,7 @@ fsubs(void *frD, void *frA, void *frB) B_s ^= 1; if (A_s != B_s && A_c == FP_CLS_INF && B_c == FP_CLS_INF) - ret |= EFLAG_VXISI; + FP_SET_EXCEPTION(EFLAG_VXISI); FP_ADD_D(R, A, B); -- 1.5.4 ^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCH 4/4] Add SPE/EFP math emulation for E500v1/v2 processors. 2008-10-28 3:50 ` [PATCH 3/4] math-emu: Remove redundant 'ret' Liu Yu @ 2008-10-28 3:50 ` Liu Yu 2008-11-06 4:05 ` Kumar Gala 2008-11-06 4:05 ` [PATCH 3/4] math-emu: Remove redundant 'ret' Kumar Gala 1 sibling, 1 reply; 9+ messages in thread From: Liu Yu @ 2008-10-28 3:50 UTC (permalink / raw) To: linuxppc-dev; +Cc: kumar.gala, Liu Yu This patch add the handlers of SPE/EFP exceptions. The code is used to emulate float point arithmetic, when MSR(SPE) is enabled and receive EFP data interrupt or EFP round interrupt. This patch has no conflict with or dependence on FP math-emu. The code has been tested by TestFloat. Now the code doesn't support SPE/EFP instructions emulation (it won't be called when receive program interrupt), but it could be easily added. Signed-off-by: Liu Yu <yu.liu@freescale.com> --- arch/powerpc/include/asm/processor.h | 6 + arch/powerpc/include/asm/sfp-machine.h | 36 ++- arch/powerpc/kernel/head_fsl_booke.S | 7 +- arch/powerpc/kernel/traps.c | 62 +++- arch/powerpc/math-emu/Makefile | 2 + arch/powerpc/math-emu/math_efp.c | 720 ++++++++++++++++++++++++++++++++ 6 files changed, 813 insertions(+), 20 deletions(-) create mode 100644 arch/powerpc/math-emu/math_efp.c diff --git a/arch/powerpc/include/asm/processor.h b/arch/powerpc/include/asm/processor.h index 101ed87..cd7a478 100644 --- a/arch/powerpc/include/asm/processor.h +++ b/arch/powerpc/include/asm/processor.h @@ -207,6 +207,11 @@ struct thread_struct { #define INIT_SP_LIMIT \ (_ALIGN_UP(sizeof(init_thread_info), 16) + (unsigned long) &init_stack) +#ifdef CONFIG_SPE +#define SPEFSCR_INIT .spefscr = SPEFSCR_FINVE | SPEFSCR_FDBZE | SPEFSCR_FUNFE | SPEFSCR_FOVFE, +#else +#define SPEFSCR_INIT +#endif #ifdef CONFIG_PPC32 #define INIT_THREAD { \ @@ -215,6 +220,7 @@ struct thread_struct { .fs = KERNEL_DS, \ .pgdir = swapper_pg_dir, \ .fpexc_mode = MSR_FE0 | MSR_FE1, \ + SPEFSCR_INIT \ } #else #define INIT_THREAD { \ diff --git a/arch/powerpc/include/asm/sfp-machine.h b/arch/powerpc/include/asm/sfp-machine.h index 88af036..3d9f831 100644 --- a/arch/powerpc/include/asm/sfp-machine.h +++ b/arch/powerpc/include/asm/sfp-machine.h @@ -97,6 +97,20 @@ #define _FP_KEEPNANFRACP 1 +#ifdef FP_EX_BOOKE_E500_SPE +#define FP_EX_INEXACT (1 << 21) +#define FP_EX_INVALID (1 << 20) +#define FP_EX_DIVZERO (1 << 19) +#define FP_EX_UNDERFLOW (1 << 18) +#define FP_EX_OVERFLOW (1 << 17) +#define FP_INHIBIT_RESULTS 0 + +#define __FPU_FPSCR (current->thread.spefscr) +#define __FPU_ENABLED_EXC \ +({ \ + (__FPU_FPSCR >> 2) & 0x1f; \ +}) +#else /* Exception flags. We use the bit positions of the appropriate bits in the FPSCR, which also correspond to the FE_* bits. This makes everything easier ;-). */ @@ -111,6 +125,18 @@ #define FP_EX_DIVZERO (1 << (31 - 5)) #define FP_EX_INEXACT (1 << (31 - 6)) +#define __FPU_FPSCR (current->thread.fpscr.val) + +/* We only actually write to the destination register + * if exceptions signalled (if any) will not trap. + */ +#define __FPU_ENABLED_EXC \ +({ \ + (__FPU_FPSCR >> 3) & 0x1f; \ +}) + +#endif + /* * If one NaN is signaling and the other is not, * we choose that one, otherwise we choose X. @@ -135,16 +161,6 @@ #include <linux/kernel.h> #include <linux/sched.h> -#define __FPU_FPSCR (current->thread.fpscr.val) - -/* We only actually write to the destination register - * if exceptions signalled (if any) will not trap. - */ -#define __FPU_ENABLED_EXC \ -({ \ - (__FPU_FPSCR >> 3) & 0x1f; \ -}) - #define __FPU_TRAP_P(bits) \ ((__FPU_ENABLED_EXC & (bits)) != 0) diff --git a/arch/powerpc/kernel/head_fsl_booke.S b/arch/powerpc/kernel/head_fsl_booke.S index 1751882..8075afb 100644 --- a/arch/powerpc/kernel/head_fsl_booke.S +++ b/arch/powerpc/kernel/head_fsl_booke.S @@ -1027,12 +1027,13 @@ END_FTR_SECTION_IFSET(CPU_FTR_BIG_PHYS) /* SPE Floating Point Data */ #ifdef CONFIG_SPE EXCEPTION(0x2030, SPEFloatingPointData, SPEFloatingPointException, EXC_XFER_EE); -#else - EXCEPTION(0x2040, SPEFloatingPointData, unknown_exception, EXC_XFER_EE) -#endif /* CONFIG_SPE */ /* SPE Floating Point Round */ + EXCEPTION(0x2050, SPEFloatingPointRound, SPEFloatingPointRoundException, EXC_XFER_EE) +#else + EXCEPTION(0x2040, SPEFloatingPointData, unknown_exception, EXC_XFER_EE) EXCEPTION(0x2050, SPEFloatingPointRound, unknown_exception, EXC_XFER_EE) +#endif /* CONFIG_SPE */ /* Performance Monitor */ EXCEPTION(0x2060, PerformanceMonitor, performance_monitor_exception, EXC_XFER_STD) diff --git a/arch/powerpc/kernel/traps.c b/arch/powerpc/kernel/traps.c index 81ccb8d..cab88aa 100644 --- a/arch/powerpc/kernel/traps.c +++ b/arch/powerpc/kernel/traps.c @@ -1161,37 +1161,85 @@ void CacheLockingException(struct pt_regs *regs, unsigned long address, #ifdef CONFIG_SPE void SPEFloatingPointException(struct pt_regs *regs) { + extern int do_spe_mathemu(struct pt_regs *regs); unsigned long spefscr; int fpexc_mode; int code = 0; + int err; + + preempt_disable(); + if (regs->msr & MSR_SPE) + giveup_spe(current); + preempt_enable(); spefscr = current->thread.spefscr; fpexc_mode = current->thread.fpexc_mode; - /* Hardware does not neccessarily set sticky - * underflow/overflow/invalid flags */ if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) { code = FPE_FLTOVF; - spefscr |= SPEFSCR_FOVFS; } else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) { code = FPE_FLTUND; - spefscr |= SPEFSCR_FUNFS; } else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV)) code = FPE_FLTDIV; else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) { code = FPE_FLTINV; - spefscr |= SPEFSCR_FINVS; } else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES)) code = FPE_FLTRES; - current->thread.spefscr = spefscr; + err = do_spe_mathemu(regs); + if (err == 0) { + regs->nip += 4; /* skip emulated instruction */ + emulate_single_step(regs); + return; + } + + if (err == -EFAULT) { + /* got an error reading the instruction */ + _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip); + } else if (err == -EINVAL) { + /* didn't recognize the instruction */ + printk(KERN_ERR "unrecognized spe instruction " + "in %s at %lx\n", current->comm, regs->nip); + } else { + _exception(SIGFPE, regs, code, regs->nip); + } - _exception(SIGFPE, regs, code, regs->nip); return; } + +void SPEFloatingPointRoundException(struct pt_regs *regs) +{ + extern int speround_handler(struct pt_regs *regs); + int err; + + preempt_disable(); + if (regs->msr & MSR_SPE) + giveup_spe(current); + preempt_enable(); + + regs->nip -= 4; + err = speround_handler(regs); + if (err == 0) { + regs->nip += 4; /* skip emulated instruction */ + emulate_single_step(regs); + return; + } + + if (err == -EFAULT) { + /* got an error reading the instruction */ + _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip); + } else if (err == -EINVAL) { + /* didn't recognize the instruction */ + printk(KERN_ERR "unrecognized spe instruction " + "in %s at %lx\n", current->comm, regs->nip); + } else { + _exception(SIGFPE, regs, 0, regs->nip); + return; + } +} #endif /* diff --git a/arch/powerpc/math-emu/Makefile b/arch/powerpc/math-emu/Makefile index 03aa98d..f9e506a 100644 --- a/arch/powerpc/math-emu/Makefile +++ b/arch/powerpc/math-emu/Makefile @@ -11,6 +11,8 @@ obj-$(CONFIG_MATH_EMULATION) += fabs.o fadd.o fadds.o fcmpo.o fcmpu.o \ mcrfs.o mffs.o mtfsb0.o mtfsb1.o \ mtfsf.o mtfsfi.o stfiwx.o stfs.o +obj-$(CONFIG_SPE) += math_efp.o + CFLAGS_fabs.o = -fno-builtin-fabs CFLAGS_math.o = -fno-builtin-fabs diff --git a/arch/powerpc/math-emu/math_efp.c b/arch/powerpc/math-emu/math_efp.c new file mode 100644 index 0000000..0c50301 --- /dev/null +++ b/arch/powerpc/math-emu/math_efp.c @@ -0,0 +1,720 @@ +/* + * arch/powerpc/math-emu/math_efp.c + * + * Copyright (C) 2006-2008 Freescale Semiconductor, Inc. All rights reserved. + * + * Author: Ebony Zhu, <ebony.zhu@freescale.com> + * Yu Liu, <yu.liu@freescale.com> + * + * Derived from arch/alpha/math-emu/math.c + * arch/powerpc/math-emu/math.c + * + * Description: + * This file is the exception handler to make E500 SPE instructions + * fully comply with IEEE-754 floating point standard. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version + * 2 of the License, or (at your option) any later version. + */ + +#include <linux/types.h> + +#include <asm/uaccess.h> +#include <asm/reg.h> + +#define FP_EX_BOOKE_E500_SPE +#include <asm/sfp-machine.h> + +#include <math-emu/soft-fp.h> +#include <math-emu/single.h> +#include <math-emu/double.h> + +#define EFAPU 0x4 + +#define VCT 0x4 +#define SPFP 0x6 +#define DPFP 0x7 + +#define EFSADD 0x2c0 +#define EFSSUB 0x2c1 +#define EFSABS 0x2c4 +#define EFSNABS 0x2c5 +#define EFSNEG 0x2c6 +#define EFSMUL 0x2c8 +#define EFSDIV 0x2c9 +#define EFSCMPGT 0x2cc +#define EFSCMPLT 0x2cd +#define EFSCMPEQ 0x2ce +#define EFSCFD 0x2cf +#define EFSCFSI 0x2d1 +#define EFSCTUI 0x2d4 +#define EFSCTSI 0x2d5 +#define EFSCTUF 0x2d6 +#define EFSCTSF 0x2d7 +#define EFSCTUIZ 0x2d8 +#define EFSCTSIZ 0x2da + +#define EVFSADD 0x280 +#define EVFSSUB 0x281 +#define EVFSABS 0x284 +#define EVFSNABS 0x285 +#define EVFSNEG 0x286 +#define EVFSMUL 0x288 +#define EVFSDIV 0x289 +#define EVFSCMPGT 0x28c +#define EVFSCMPLT 0x28d +#define EVFSCMPEQ 0x28e +#define EVFSCTUI 0x294 +#define EVFSCTSI 0x295 +#define EVFSCTUF 0x296 +#define EVFSCTSF 0x297 +#define EVFSCTUIZ 0x298 +#define EVFSCTSIZ 0x29a + +#define EFDADD 0x2e0 +#define EFDSUB 0x2e1 +#define EFDABS 0x2e4 +#define EFDNABS 0x2e5 +#define EFDNEG 0x2e6 +#define EFDMUL 0x2e8 +#define EFDDIV 0x2e9 +#define EFDCTUIDZ 0x2ea +#define EFDCTSIDZ 0x2eb +#define EFDCMPGT 0x2ec +#define EFDCMPLT 0x2ed +#define EFDCMPEQ 0x2ee +#define EFDCFS 0x2ef +#define EFDCTUI 0x2f4 +#define EFDCTSI 0x2f5 +#define EFDCTUF 0x2f6 +#define EFDCTSF 0x2f7 +#define EFDCTUIZ 0x2f8 +#define EFDCTSIZ 0x2fa + +#define AB 2 +#define XA 3 +#define XB 4 +#define XCR 5 +#define NOTYPE 0 + +#define SIGN_BIT_S (1UL << 31) +#define SIGN_BIT_D (1ULL << 63) +#define FP_EX_MASK (FP_EX_INEXACT | FP_EX_INVALID | FP_EX_DIVZERO | \ + FP_EX_UNDERFLOW | FP_EX_OVERFLOW) + +union dw_union { + u64 dp[1]; + u32 wp[2]; +}; + +static unsigned long insn_type(unsigned long speinsn) +{ + unsigned long ret = NOTYPE; + + switch (speinsn & 0x7ff) { + case EFSABS: ret = XA; break; + case EFSADD: ret = AB; break; + case EFSCFD: ret = XB; break; + case EFSCMPEQ: ret = XCR; break; + case EFSCMPGT: ret = XCR; break; + case EFSCMPLT: ret = XCR; break; + case EFSCTSF: ret = XB; break; + case EFSCTSI: ret = XB; break; + case EFSCTSIZ: ret = XB; break; + case EFSCTUF: ret = XB; break; + case EFSCTUI: ret = XB; break; + case EFSCTUIZ: ret = XB; break; + case EFSDIV: ret = AB; break; + case EFSMUL: ret = AB; break; + case EFSNABS: ret = XA; break; + case EFSNEG: ret = XA; break; + case EFSSUB: ret = AB; break; + case EFSCFSI: ret = XB; break; + + case EVFSABS: ret = XA; break; + case EVFSADD: ret = AB; break; + case EVFSCMPEQ: ret = XCR; break; + case EVFSCMPGT: ret = XCR; break; + case EVFSCMPLT: ret = XCR; break; + case EVFSCTSF: ret = XB; break; + case EVFSCTSI: ret = XB; break; + case EVFSCTSIZ: ret = XB; break; + case EVFSCTUF: ret = XB; break; + case EVFSCTUI: ret = XB; break; + case EVFSCTUIZ: ret = XB; break; + case EVFSDIV: ret = AB; break; + case EVFSMUL: ret = AB; break; + case EVFSNABS: ret = XA; break; + case EVFSNEG: ret = XA; break; + case EVFSSUB: ret = AB; break; + + case EFDABS: ret = XA; break; + case EFDADD: ret = AB; break; + case EFDCFS: ret = XB; break; + case EFDCMPEQ: ret = XCR; break; + case EFDCMPGT: ret = XCR; break; + case EFDCMPLT: ret = XCR; break; + case EFDCTSF: ret = XB; break; + case EFDCTSI: ret = XB; break; + case EFDCTSIDZ: ret = XB; break; + case EFDCTSIZ: ret = XB; break; + case EFDCTUF: ret = XB; break; + case EFDCTUI: ret = XB; break; + case EFDCTUIDZ: ret = XB; break; + case EFDCTUIZ: ret = XB; break; + case EFDDIV: ret = AB; break; + case EFDMUL: ret = AB; break; + case EFDNABS: ret = XA; break; + case EFDNEG: ret = XA; break; + case EFDSUB: ret = AB; break; + + default: + printk(KERN_ERR "\nOoops! SPE instruction no type found."); + printk(KERN_ERR "\ninst code: %08lx\n", speinsn); + } + + return ret; +} + +int do_spe_mathemu(struct pt_regs *regs) +{ + FP_DECL_EX; + int IR, cmp; + + unsigned long type, func, fc, fa, fb, src, speinsn; + union dw_union vc, va, vb; + + if (get_user(speinsn, (unsigned int __user *) regs->nip)) + return -EFAULT; + if ((speinsn >> 26) != EFAPU) + return -EINVAL; /* not an spe instruction */ + + type = insn_type(speinsn); + if (type == NOTYPE) + return -ENOSYS; + + func = speinsn & 0x7ff; + fc = (speinsn >> 21) & 0x1f; + fa = (speinsn >> 16) & 0x1f; + fb = (speinsn >> 11) & 0x1f; + src = (speinsn >> 5) & 0x7; + + vc.wp[0] = current->thread.evr[fc]; + vc.wp[1] = regs->gpr[fc]; + va.wp[0] = current->thread.evr[fa]; + va.wp[1] = regs->gpr[fa]; + vb.wp[0] = current->thread.evr[fb]; + vb.wp[1] = regs->gpr[fb]; + + __FPU_FPSCR = mfspr(SPRN_SPEFSCR); + +#ifdef DEBUG + printk("speinsn:%08lx spefscr:%08lx\n", speinsn, __FPU_FPSCR); + printk("vc: %08x %08x\n", vc.wp[0], vc.wp[1]); + printk("va: %08x %08x\n", va.wp[0], va.wp[1]); + printk("vb: %08x %08x\n", vb.wp[0], vb.wp[1]); +#endif + + switch (src) { + case SPFP: { + FP_DECL_S(SA); FP_DECL_S(SB); FP_DECL_S(SR); + + switch (type) { + case AB: + case XCR: + FP_UNPACK_SP(SA, va.wp + 1); + case XB: + FP_UNPACK_SP(SB, vb.wp + 1); + break; + case XA: + FP_UNPACK_SP(SA, va.wp + 1); + break; + } + +#ifdef DEBUG + printk("SA: %ld %08lx %ld (%ld)\n", SA_s, SA_f, SA_e, SA_c); + printk("SB: %ld %08lx %ld (%ld)\n", SB_s, SB_f, SB_e, SB_c); +#endif + + switch (func) { + case EFSABS: + vc.wp[1] = va.wp[1] & ~SIGN_BIT_S; + goto update_regs; + + case EFSNABS: + vc.wp[1] = va.wp[1] | SIGN_BIT_S; + goto update_regs; + + case EFSNEG: + vc.wp[1] = va.wp[1] ^ SIGN_BIT_S; + goto update_regs; + + case EFSADD: + FP_ADD_S(SR, SA, SB); + goto pack_s; + + case EFSSUB: + FP_SUB_S(SR, SA, SB); + goto pack_s; + + case EFSMUL: + FP_MUL_S(SR, SA, SB); + goto pack_s; + + case EFSDIV: + FP_DIV_S(SR, SA, SB); + goto pack_s; + + case EFSCMPEQ: + cmp = 0; + goto cmp_s; + + case EFSCMPGT: + cmp = 1; + goto cmp_s; + + case EFSCMPLT: + cmp = -1; + goto cmp_s; + + case EFSCTSF: + case EFSCTUF: + if (!((vb.wp[1] >> 23) == 0xff && ((vb.wp[1] & 0x7fffff) > 0))) { + /* NaN */ + if (((vb.wp[1] >> 23) & 0xff) == 0) { + /* denorm */ + vc.wp[1] = 0x0; + } else if ((vb.wp[1] >> 31) == 0) { + /* positive normal */ + vc.wp[1] = (func == EFSCTSF) ? + 0x7fffffff : 0xffffffff; + } else { /* negative normal */ + vc.wp[1] = (func == EFSCTSF) ? + 0x80000000 : 0x0; + } + } else { /* rB is NaN */ + vc.wp[1] = 0x0; + } + goto update_regs; + + case EFSCFD: { + FP_DECL_D(DB); + FP_CLEAR_EXCEPTIONS; + FP_UNPACK_DP(DB, vb.dp); +#ifdef DEBUG + printk("DB: %ld %08lx %08lx %ld (%ld)\n", + DB_s, DB_f1, DB_f0, DB_e, DB_c); +#endif + FP_CONV(S, D, 1, 2, SR, DB); + goto pack_s; + } + + case EFSCTSI: + case EFSCTSIZ: + case EFSCTUI: + case EFSCTUIZ: + if (func & 0x4) { + _FP_ROUND(1, SB); + } else { + _FP_ROUND_ZERO(1, SB); + } + FP_TO_INT_S(vc.wp[1], SB, 32, ((func & 0x3) != 0)); + goto update_regs; + + default: + goto illegal; + } + break; + +pack_s: +#ifdef DEBUG + printk("SR: %ld %08lx %ld (%ld)\n", SR_s, SR_f, SR_e, SR_c); +#endif + FP_PACK_SP(vc.wp + 1, SR); + goto update_regs; + +cmp_s: + FP_CMP_S(IR, SA, SB, 3); + if (IR == 3 && (FP_ISSIGNAN_S(SA) || FP_ISSIGNAN_S(SB))) + FP_SET_EXCEPTION(FP_EX_INVALID); + if (IR == cmp) { + IR = 0x4; + } else { + IR = 0; + } + goto update_ccr; + } + + case DPFP: { + FP_DECL_D(DA); FP_DECL_D(DB); FP_DECL_D(DR); + + switch (type) { + case AB: + case XCR: + FP_UNPACK_DP(DA, va.dp); + case XB: + FP_UNPACK_DP(DB, vb.dp); + break; + case XA: + FP_UNPACK_DP(DA, va.dp); + break; + } + +#ifdef DEBUG + printk("DA: %ld %08lx %08lx %ld (%ld)\n", + DA_s, DA_f1, DA_f0, DA_e, DA_c); + printk("DB: %ld %08lx %08lx %ld (%ld)\n", + DB_s, DB_f1, DB_f0, DB_e, DB_c); +#endif + + switch (func) { + case EFDABS: + vc.dp[0] = va.dp[0] & ~SIGN_BIT_D; + goto update_regs; + + case EFDNABS: + vc.dp[0] = va.dp[0] | SIGN_BIT_D; + goto update_regs; + + case EFDNEG: + vc.dp[0] = va.dp[0] ^ SIGN_BIT_D; + goto update_regs; + + case EFDADD: + FP_ADD_D(DR, DA, DB); + goto pack_d; + + case EFDSUB: + FP_SUB_D(DR, DA, DB); + goto pack_d; + + case EFDMUL: + FP_MUL_D(DR, DA, DB); + goto pack_d; + + case EFDDIV: + FP_DIV_D(DR, DA, DB); + goto pack_d; + + case EFDCMPEQ: + cmp = 0; + goto cmp_d; + + case EFDCMPGT: + cmp = 1; + goto cmp_d; + + case EFDCMPLT: + cmp = -1; + goto cmp_d; + + case EFDCTSF: + case EFDCTUF: + if (!((vb.wp[0] >> 20) == 0x7ff && + ((vb.wp[0] & 0xfffff) > 0 || (vb.wp[1] > 0)))) { + /* not a NaN */ + if (((vb.wp[0] >> 20) & 0x7ff) == 0) { + /* denorm */ + vc.wp[1] = 0x0; + } else if ((vb.wp[0] >> 31) == 0) { + /* positive normal */ + vc.wp[1] = (func == EFDCTSF) ? + 0x7fffffff : 0xffffffff; + } else { /* negative normal */ + vc.wp[1] = (func == EFDCTSF) ? + 0x80000000 : 0x0; + } + } else { /* NaN */ + vc.wp[1] = 0x0; + } + goto update_regs; + + case EFDCFS: { + FP_DECL_S(SB); + FP_CLEAR_EXCEPTIONS; + FP_UNPACK_SP(SB, vb.wp + 1); +#ifdef DEBUG + printk("SB: %ld %08lx %ld (%ld)\n", + SB_s, SB_f, SB_e, SB_c); +#endif + FP_CONV(D, S, 2, 1, DR, SB); + goto pack_d; + } + + case EFDCTUIDZ: + case EFDCTSIDZ: + _FP_ROUND_ZERO(2, DB); + FP_TO_INT_D(vc.dp[0], DB, 64, ((func & 0x1) == 0)); + goto update_regs; + + case EFDCTUI: + case EFDCTSI: + case EFDCTUIZ: + case EFDCTSIZ: + if (func & 0x4) { + _FP_ROUND(2, DB); + } else { + _FP_ROUND_ZERO(2, DB); + } + FP_TO_INT_D(vc.wp[1], DB, 32, ((func & 0x3) != 0)); + goto update_regs; + + default: + goto illegal; + } + break; + +pack_d: +#ifdef DEBUG + printk("DR: %ld %08lx %08lx %ld (%ld)\n", + DR_s, DR_f1, DR_f0, DR_e, DR_c); +#endif + FP_PACK_DP(vc.dp, DR); + goto update_regs; + +cmp_d: + FP_CMP_D(IR, DA, DB, 3); + if (IR == 3 && (FP_ISSIGNAN_D(DA) || FP_ISSIGNAN_D(DB))) + FP_SET_EXCEPTION(FP_EX_INVALID); + if (IR == cmp) { + IR = 0x4; + } else { + IR = 0; + } + goto update_ccr; + + } + + case VCT: { + FP_DECL_S(SA0); FP_DECL_S(SB0); FP_DECL_S(SR0); + FP_DECL_S(SA1); FP_DECL_S(SB1); FP_DECL_S(SR1); + int IR0, IR1; + + switch (type) { + case AB: + case XCR: + FP_UNPACK_SP(SA0, va.wp); + FP_UNPACK_SP(SA1, va.wp + 1); + case XB: + FP_UNPACK_SP(SB0, vb.wp); + FP_UNPACK_SP(SB1, vb.wp + 1); + break; + case XA: + FP_UNPACK_SP(SA0, va.wp); + FP_UNPACK_SP(SA1, va.wp + 1); + break; + } + +#ifdef DEBUG + printk("SA0: %ld %08lx %ld (%ld)\n", SA0_s, SA0_f, SA0_e, SA0_c); + printk("SA1: %ld %08lx %ld (%ld)\n", SA1_s, SA1_f, SA1_e, SA1_c); + printk("SB0: %ld %08lx %ld (%ld)\n", SB0_s, SB0_f, SB0_e, SB0_c); + printk("SB1: %ld %08lx %ld (%ld)\n", SB1_s, SB1_f, SB1_e, SB1_c); +#endif + + switch (func) { + case EVFSABS: + vc.wp[0] = va.wp[0] & ~SIGN_BIT_S; + vc.wp[1] = va.wp[1] & ~SIGN_BIT_S; + goto update_regs; + + case EVFSNABS: + vc.wp[0] = va.wp[0] | SIGN_BIT_S; + vc.wp[1] = va.wp[1] | SIGN_BIT_S; + goto update_regs; + + case EVFSNEG: + vc.wp[0] = va.wp[0] ^ SIGN_BIT_S; + vc.wp[1] = va.wp[1] ^ SIGN_BIT_S; + goto update_regs; + + case EVFSADD: + FP_ADD_S(SR0, SA0, SB0); + FP_ADD_S(SR1, SA1, SB1); + goto pack_vs; + + case EVFSSUB: + FP_SUB_S(SR0, SA0, SB0); + FP_SUB_S(SR1, SA1, SB1); + goto pack_vs; + + case EVFSMUL: + FP_MUL_S(SR0, SA0, SB0); + FP_MUL_S(SR1, SA1, SB1); + goto pack_vs; + + case EVFSDIV: + FP_DIV_S(SR0, SA0, SB0); + FP_DIV_S(SR1, SA1, SB1); + goto pack_vs; + + case EVFSCMPEQ: + cmp = 0; + goto cmp_vs; + + case EVFSCMPGT: + cmp = 1; + goto cmp_vs; + + case EVFSCMPLT: + cmp = -1; + goto cmp_vs; + + case EVFSCTSF: + __asm__ __volatile__ ("mtspr 512, %4\n" + "efsctsf %0, %2\n" + "efsctsf %1, %3\n" + : "=r" (vc.wp[0]), "=r" (vc.wp[1]) + : "r" (vb.wp[0]), "r" (vb.wp[1]), "r" (0)); + goto update_regs; + + case EVFSCTUF: + __asm__ __volatile__ ("mtspr 512, %4\n" + "efsctuf %0, %2\n" + "efsctuf %1, %3\n" + : "=r" (vc.wp[0]), "=r" (vc.wp[1]) + : "r" (vb.wp[0]), "r" (vb.wp[1]), "r" (0)); + goto update_regs; + + case EVFSCTUI: + case EVFSCTSI: + case EVFSCTUIZ: + case EVFSCTSIZ: + if (func & 0x4) { + _FP_ROUND(1, SB0); + _FP_ROUND(1, SB1); + } else { + _FP_ROUND_ZERO(1, SB0); + _FP_ROUND_ZERO(1, SB1); + } + FP_TO_INT_S(vc.wp[0], SB0, 32, ((func & 0x3) != 0)); + FP_TO_INT_S(vc.wp[1], SB1, 32, ((func & 0x3) != 0)); + goto update_regs; + + default: + goto illegal; + } + break; + +pack_vs: +#ifdef DEBUG + printk("SR0: %ld %08lx %ld (%ld)\n", SR0_s, SR0_f, SR0_e, SR0_c); + printk("SR1: %ld %08lx %ld (%ld)\n", SR1_s, SR1_f, SR1_e, SR1_c); +#endif + FP_PACK_SP(vc.wp, SR0); + FP_PACK_SP(vc.wp + 1, SR1); + goto update_regs; + +cmp_vs: + { + int ch, cl; + + FP_CMP_S(IR0, SA0, SB0, 3); + FP_CMP_S(IR1, SA1, SB1, 3); + if (IR0 == 3 && (FP_ISSIGNAN_S(SA0) || FP_ISSIGNAN_S(SB0))) + FP_SET_EXCEPTION(FP_EX_INVALID); + if (IR1 == 3 && (FP_ISSIGNAN_S(SA1) || FP_ISSIGNAN_S(SB1))) + FP_SET_EXCEPTION(FP_EX_INVALID); + ch = (IR0 == cmp) ? 1 : 0; + cl = (IR1 == cmp) ? 1 : 0; + IR = (ch << 3) | (cl << 2) | ((ch | cl) << 1) | + ((ch & cl) << 0); + goto update_ccr; + } + } + default: + return -EINVAL; + } + +update_ccr: + regs->ccr &= ~(15 << ((7 - ((speinsn >> 23) & 0x7)) << 2)); + regs->ccr |= (IR << ((7 - ((speinsn >> 23) & 0x7)) << 2)); + +update_regs: + __FPU_FPSCR &= ~FP_EX_MASK; + __FPU_FPSCR |= (FP_CUR_EXCEPTIONS & FP_EX_MASK); + mtspr(SPRN_SPEFSCR, __FPU_FPSCR); + + current->thread.evr[fc] = vc.wp[0]; + regs->gpr[fc] = vc.wp[1]; + +#ifdef DEBUG + printk("ccr = %08lx\n", regs->ccr); + printk("cur exceptions = %08x spefscr = %08lx\n", + FP_CUR_EXCEPTIONS, __FPU_FPSCR); + printk("vc: %08x %08x\n", vc.wp[0], vc.wp[1]); + printk("va: %08x %08x\n", va.wp[0], va.wp[1]); + printk("vb: %08x %08x\n", vb.wp[0], vb.wp[1]); +#endif + + return 0; + +illegal: + printk(KERN_ERR "\nOoops! IEEE-754 compliance handler encountered un-supported instruction.\ninst code: %08lx\n", speinsn); + return -ENOSYS; +} + +int speround_handler(struct pt_regs *regs) +{ + union dw_union fgpr; + int s_lo, s_hi; + unsigned long speinsn, type, fc; + + if (get_user(speinsn, (unsigned int __user *) regs->nip)) + return -EFAULT; + if ((speinsn >> 26) != 4) + return -EINVAL; /* not an spe instruction */ + + type = insn_type(speinsn & 0x7ff); + if (type == XCR) return -ENOSYS; + + fc = (speinsn >> 21) & 0x1f; + s_lo = regs->gpr[fc] & SIGN_BIT_S; + s_hi = current->thread.evr[fc] & SIGN_BIT_S; + fgpr.wp[0] = current->thread.evr[fc]; + fgpr.wp[1] = regs->gpr[fc]; + + __FPU_FPSCR = mfspr(SPRN_SPEFSCR); + + switch ((speinsn >> 5) & 0x7) { + /* Since SPE instructions on E500 core can handle round to nearest + * and round toward zero with IEEE-754 complied, we just need + * to handle round toward +Inf and round toward -Inf by software. + */ + case SPFP: + if ((FP_ROUNDMODE) == FP_RND_PINF) { + if (!s_lo) fgpr.wp[1]++; /* Z > 0, choose Z1 */ + } else { /* round to -Inf */ + if (s_lo) fgpr.wp[1]++; /* Z < 0, choose Z2 */ + } + break; + + case DPFP: + if (FP_ROUNDMODE == FP_RND_PINF) { + if (!s_hi) fgpr.dp[0]++; /* Z > 0, choose Z1 */ + } else { /* round to -Inf */ + if (s_hi) fgpr.dp[0]++; /* Z < 0, choose Z2 */ + } + break; + + case VCT: + if (FP_ROUNDMODE == FP_RND_PINF) { + if (!s_lo) fgpr.wp[1]++; /* Z_low > 0, choose Z1 */ + if (!s_hi) fgpr.wp[0]++; /* Z_high word > 0, choose Z1 */ + } else { /* round to -Inf */ + if (s_lo) fgpr.wp[1]++; /* Z_low < 0, choose Z2 */ + if (s_hi) fgpr.wp[0]++; /* Z_high < 0, choose Z2 */ + } + break; + + default: + return -EINVAL; + } + + current->thread.evr[fc] = fgpr.wp[0]; + regs->gpr[fc] = fgpr.wp[1]; + + return 0; +} -- 1.5.4 ^ permalink raw reply related [flat|nested] 9+ messages in thread
* Re: [PATCH 4/4] Add SPE/EFP math emulation for E500v1/v2 processors. 2008-10-28 3:50 ` [PATCH 4/4] Add SPE/EFP math emulation for E500v1/v2 processors Liu Yu @ 2008-11-06 4:05 ` Kumar Gala 0 siblings, 0 replies; 9+ messages in thread From: Kumar Gala @ 2008-11-06 4:05 UTC (permalink / raw) To: Liu Yu; +Cc: linuxppc-dev On Oct 27, 2008, at 10:50 PM, Liu Yu wrote: > This patch add the handlers of SPE/EFP exceptions. > The code is used to emulate float point arithmetic, > when MSR(SPE) is enabled and receive EFP data interrupt or EFP round > interrupt. > > This patch has no conflict with or dependence on FP math-emu. > > The code has been tested by TestFloat. > > Now the code doesn't support SPE/EFP instructions emulation > (it won't be called when receive program interrupt), > but it could be easily added. > > Signed-off-by: Liu Yu <yu.liu@freescale.com> > --- > arch/powerpc/include/asm/processor.h | 6 + > arch/powerpc/include/asm/sfp-machine.h | 36 ++- > arch/powerpc/kernel/head_fsl_booke.S | 7 +- > arch/powerpc/kernel/traps.c | 62 +++- > arch/powerpc/math-emu/Makefile | 2 + > arch/powerpc/math-emu/math_efp.c | 720 +++++++++++++++++++++++ > +++++++++ > 6 files changed, 813 insertions(+), 20 deletions(-) > create mode 100644 arch/powerpc/math-emu/math_efp.c applied to next. - k ^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH 3/4] math-emu: Remove redundant 'ret' 2008-10-28 3:50 ` [PATCH 3/4] math-emu: Remove redundant 'ret' Liu Yu 2008-10-28 3:50 ` [PATCH 4/4] Add SPE/EFP math emulation for E500v1/v2 processors Liu Yu @ 2008-11-06 4:05 ` Kumar Gala 1 sibling, 0 replies; 9+ messages in thread From: Kumar Gala @ 2008-11-06 4:05 UTC (permalink / raw) To: Liu Yu; +Cc: linuxppc-dev On Oct 27, 2008, at 10:50 PM, Liu Yu wrote: > FP_DECL_EX is already used, so ret is redundant. > And FP_SET_EXCEPTION will add status into return value. > > Signed-off-by: Liu Yu <yu.liu@freescale.com> > --- > arch/powerpc/math-emu/fadd.c | 1 - > arch/powerpc/math-emu/fcmpo.c | 5 ++--- > arch/powerpc/math-emu/fdiv.c | 9 ++++----- > arch/powerpc/math-emu/fdivs.c | 9 ++++----- > arch/powerpc/math-emu/fmadd.c | 5 ++--- > arch/powerpc/math-emu/fmadds.c | 5 ++--- > arch/powerpc/math-emu/fmsub.c | 5 ++--- > arch/powerpc/math-emu/fmsubs.c | 5 ++--- > arch/powerpc/math-emu/fmul.c | 3 +-- > arch/powerpc/math-emu/fmuls.c | 3 +-- > arch/powerpc/math-emu/fnmadd.c | 5 ++--- > arch/powerpc/math-emu/fnmadds.c | 5 ++--- > arch/powerpc/math-emu/fnmsub.c | 5 ++--- > arch/powerpc/math-emu/fnmsubs.c | 5 ++--- > arch/powerpc/math-emu/fsqrt.c | 5 ++--- > arch/powerpc/math-emu/fsqrts.c | 5 ++--- > arch/powerpc/math-emu/fsub.c | 3 +-- > arch/powerpc/math-emu/fsubs.c | 3 +-- > 18 files changed, 34 insertions(+), 52 deletions(-) applied to next. - k ^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN 2008-10-28 3:50 ` [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN Liu Yu 2008-10-28 3:50 ` [PATCH 3/4] math-emu: Remove redundant 'ret' Liu Yu @ 2008-11-06 4:05 ` Kumar Gala 1 sibling, 0 replies; 9+ messages in thread From: Kumar Gala @ 2008-11-06 4:05 UTC (permalink / raw) To: Liu Yu; +Cc: linuxppc-dev On Oct 27, 2008, at 10:50 PM, Liu Yu wrote: > Move to using the same macro definition for _FP_CHOOSENAN as s390, > sh, sparc32/64. The original author didn't understand this and > matched what sparc64 was doing and they have updated to this > definition. > > Signed-off-by: Liu Yu <yu.liu@freescale.com> > --- > arch/powerpc/include/asm/sfp-machine.h | 26 ++++++++++++++++ > +--------- > 1 files changed, 17 insertions(+), 9 deletions(-) applied to next. - k ^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH 1/4] math-emu: Fix single float point division bug 2008-10-28 3:50 ` [PATCH 1/4] math-emu: Fix single float point division bug Liu Yu 2008-10-28 3:50 ` [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN Liu Yu @ 2008-11-06 4:03 ` Kumar Gala 1 sibling, 0 replies; 9+ messages in thread From: Kumar Gala @ 2008-11-06 4:03 UTC (permalink / raw) To: Liu Yu; +Cc: linuxppc-dev On Oct 27, 2008, at 10:50 PM, Liu Yu wrote: > PowerPC float point division emulation is derived from gcc. > I reported this problem on gcc maillist and got this reply: > http://gcc.gnu.org/ml/gcc/2008-03/msg00543.html > > Since UDIV_NEEDS_NORMALIZATION is not used by kernel, we should use > _FP_DIV_MEAT_1_udiv_norm to make sure the single float point > is normalized before udiv_qrnnd. > > Signed-off-by: Liu Yu <yu.liu@freescale.com> > --- > arch/powerpc/include/asm/sfp-machine.h | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) applied to next. - k ^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2008-11-06 4:06 UTC | newest] Thread overview: 9+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2008-10-28 3:50 [Patch 0/4] math-emu Liu Yu 2008-10-28 3:50 ` [PATCH 1/4] math-emu: Fix single float point division bug Liu Yu 2008-10-28 3:50 ` [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN Liu Yu 2008-10-28 3:50 ` [PATCH 3/4] math-emu: Remove redundant 'ret' Liu Yu 2008-10-28 3:50 ` [PATCH 4/4] Add SPE/EFP math emulation for E500v1/v2 processors Liu Yu 2008-11-06 4:05 ` Kumar Gala 2008-11-06 4:05 ` [PATCH 3/4] math-emu: Remove redundant 'ret' Kumar Gala 2008-11-06 4:05 ` [PATCH 2/4] math-emu: Adopt new version of _FP_CHOOSENAN Kumar Gala 2008-11-06 4:03 ` [PATCH 1/4] math-emu: Fix single float point division bug Kumar Gala
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).