diff options
Diffstat (limited to 'src')
-rw-r--r-- | src/mesa/tnl/t_vertex_sse.c | 2 | ||||
-rw-r--r-- | src/mesa/x86/rtasm/x86sse.c | 246 | ||||
-rw-r--r-- | src/mesa/x86/rtasm/x86sse.h | 63 |
3 files changed, 205 insertions, 106 deletions
diff --git a/src/mesa/tnl/t_vertex_sse.c b/src/mesa/tnl/t_vertex_sse.c index 9515d9f81f5..922737009af 100644 --- a/src/mesa/tnl/t_vertex_sse.c +++ b/src/mesa/tnl/t_vertex_sse.c @@ -648,7 +648,7 @@ void _tnl_generate_sse_emit( GLcontext *ctx ) p.ctx = ctx; p.inputs_safe = 0; /* for now */ - p.outputs_safe = 1; /* for now */ + p.outputs_safe = 0; /* for now */ p.have_sse2 = cpu_has_xmm2; p.identity = x86_make_reg(file_XMM, 6); p.chan0 = x86_make_reg(file_XMM, 7); diff --git a/src/mesa/x86/rtasm/x86sse.c b/src/mesa/x86/rtasm/x86sse.c index 612cd51a6ee..39c0e9b851a 100644 --- a/src/mesa/x86/rtasm/x86sse.c +++ b/src/mesa/x86/rtasm/x86sse.c @@ -1,4 +1,4 @@ -#if defined(USE_X86_ASM) || defined(SLANG_X86) +#if defined(__i386__) || defined(__386__) #include "imports.h" #include "x86sse.h" @@ -6,54 +6,78 @@ #define DISASSEM 0 #define X86_TWOB 0x0f -/* Emit bytes to the instruction stream: - */ -static void emit_1b( struct x86_function *p, GLbyte b0 ) +static unsigned char *cptr( void (*label)() ) { - *(GLbyte *)(p->csr++) = b0; + return (unsigned char *)(unsigned long)label; } -static void emit_1i( struct x86_function *p, GLint i0 ) + +static void do_realloc( struct x86_function *p ) { - *(GLint *)(p->csr) = i0; - p->csr += 4; + if (p->size == 0) { + p->size = 1024; + p->store = _mesa_exec_malloc(p->size); + p->csr = p->store; + } + else { + unsigned used = p->csr - p->store; + unsigned char *tmp = p->store; + p->size *= 2; + p->store = _mesa_exec_malloc(p->size); + memcpy(p->store, tmp, used); + p->csr = p->store + used; + _mesa_exec_free(tmp); + } } -static void disassem( struct x86_function *p, const char *fn ) +/* Emit bytes to the instruction stream: + */ +static unsigned char *reserve( struct x86_function *p, int bytes ) { -#if DISASSEM && 0 - if (fn && fn != p->fn) { - _mesa_printf("0x%x: %s\n", p->csr, fn); - p->fn = fn; + if (p->csr + bytes - p->store > p->size) + do_realloc(p); + + { + unsigned char *csr = p->csr; + p->csr += bytes; + return csr; } -#endif } -static void emit_1ub_fn( struct x86_function *p, GLubyte b0, const char *fn ) + + +static void emit_1b( struct x86_function *p, char b0 ) { - disassem(p, fn); - *(p->csr++) = b0; + char *csr = (char *)reserve(p, 1); + *csr = b0; } -static void emit_2ub_fn( struct x86_function *p, GLubyte b0, GLubyte b1, const char *fn ) +static void emit_1i( struct x86_function *p, int i0 ) { - disassem(p, fn); - *(p->csr++) = b0; - *(p->csr++) = b1; + int *icsr = (int *)reserve(p, sizeof(i0)); + *icsr = i0; } -static void emit_3ub_fn( struct x86_function *p, GLubyte b0, GLubyte b1, GLubyte b2, const char *fn ) +static void emit_1ub( struct x86_function *p, unsigned char b0 ) { - disassem(p, fn); - *(p->csr++) = b0; - *(p->csr++) = b1; - *(p->csr++) = b2; + unsigned char *csr = reserve(p, 1); + *csr++ = b0; } -#define emit_1ub(p, b0) emit_1ub_fn(p, b0, __FUNCTION__) -#define emit_2ub(p, b0, b1) emit_2ub_fn(p, b0, b1, __FUNCTION__) -#define emit_3ub(p, b0, b1, b2) emit_3ub_fn(p, b0, b1, b2, __FUNCTION__) +static void emit_2ub( struct x86_function *p, unsigned char b0, unsigned char b1 ) +{ + unsigned char *csr = reserve(p, 2); + *csr++ = b0; + *csr++ = b1; +} +static void emit_3ub( struct x86_function *p, unsigned char b0, unsigned char b1, unsigned char b2 ) +{ + unsigned char *csr = reserve(p, 3); + *csr++ = b0; + *csr++ = b1; + *csr++ = b2; +} /* Build a modRM byte + possible displacement. No treatment of SIB @@ -63,7 +87,7 @@ static void emit_modrm( struct x86_function *p, struct x86_reg reg, struct x86_reg regmem ) { - GLubyte val = 0; + unsigned char val = 0; assert(reg.mod == mod_REG); @@ -71,13 +95,13 @@ static void emit_modrm( struct x86_function *p, val |= reg.idx << 3; /* reg field */ val |= regmem.idx; /* r/m field */ - emit_1ub_fn(p, val, 0); + emit_1ub(p, val); /* Oh-oh we've stumbled into the SIB thing. */ if (regmem.file == file_REG32 && regmem.idx == reg_SP) { - emit_1ub_fn(p, 0x24, 0); /* simplistic! */ + emit_1ub(p, 0x24); /* simplistic! */ } switch (regmem.mod) { @@ -98,7 +122,7 @@ static void emit_modrm( struct x86_function *p, static void emit_modrm_noreg( struct x86_function *p, - GLuint op, + unsigned op, struct x86_reg regmem ) { struct x86_reg dummy = x86_make_reg(file_REG32, op); @@ -111,21 +135,21 @@ static void emit_modrm_noreg( struct x86_function *p, * the arguments presented. */ static void emit_op_modrm( struct x86_function *p, - GLubyte op_dst_is_reg, - GLubyte op_dst_is_mem, + unsigned char op_dst_is_reg, + unsigned char op_dst_is_mem, struct x86_reg dst, struct x86_reg src ) { switch (dst.mod) { case mod_REG: - emit_1ub_fn(p, op_dst_is_reg, 0); + emit_1ub(p, op_dst_is_reg); emit_modrm(p, dst, src); break; case mod_INDIRECT: case mod_DISP32: case mod_DISP8: assert(src.mod == mod_REG); - emit_1ub_fn(p, op_dst_is_mem, 0); + emit_1ub(p, op_dst_is_mem); emit_modrm(p, src, dst); break; default: @@ -156,7 +180,7 @@ struct x86_reg x86_make_reg( enum x86_reg_file file, } struct x86_reg x86_make_disp( struct x86_reg reg, - GLint disp ) + int disp ) { assert(reg.file == file_REG32); @@ -185,7 +209,7 @@ struct x86_reg x86_get_base_reg( struct x86_reg reg ) return x86_make_reg( reg.file, reg.idx ); } -GLubyte *x86_get_label( struct x86_function *p ) +unsigned char *x86_get_label( struct x86_function *p ) { return p->csr; } @@ -199,13 +223,13 @@ GLubyte *x86_get_label( struct x86_function *p ) void x86_jcc( struct x86_function *p, enum x86_cc cc, - GLubyte *label ) + unsigned char *label ) { - GLint offset = label - (x86_get_label(p) + 2); + int offset = label - (x86_get_label(p) + 2); if (offset <= 127 && offset >= -128) { emit_1ub(p, 0x70 + cc); - emit_1b(p, (GLbyte) offset); + emit_1b(p, (char) offset); } else { offset = label - (x86_get_label(p) + 6); @@ -216,7 +240,7 @@ void x86_jcc( struct x86_function *p, /* Always use a 32bit offset for forward jumps: */ -GLubyte *x86_jcc_forward( struct x86_function *p, +unsigned char *x86_jcc_forward( struct x86_function *p, enum x86_cc cc ) { emit_2ub(p, 0x0f, 0x80 + cc); @@ -224,14 +248,14 @@ GLubyte *x86_jcc_forward( struct x86_function *p, return x86_get_label(p); } -GLubyte *x86_jmp_forward( struct x86_function *p) +unsigned char *x86_jmp_forward( struct x86_function *p) { emit_1ub(p, 0xe9); emit_1i(p, 0); return x86_get_label(p); } -GLubyte *x86_call_forward( struct x86_function *p) +unsigned char *x86_call_forward( struct x86_function *p) { emit_1ub(p, 0xe8); emit_1i(p, 0); @@ -241,28 +265,41 @@ GLubyte *x86_call_forward( struct x86_function *p) /* Fixup offset from forward jump: */ void x86_fixup_fwd_jump( struct x86_function *p, - GLubyte *fixup ) + unsigned char *fixup ) { *(int *)(fixup - 4) = x86_get_label(p) - fixup; } -void x86_jmp( struct x86_function *p, GLubyte *label) +void x86_jmp( struct x86_function *p, unsigned char *label) { emit_1ub(p, 0xe9); emit_1i(p, label - x86_get_label(p) - 4); } -void x86_call( struct x86_function *p, GLubyte *label) +#if 0 +/* This doesn't work once we start reallocating & copying the + * generated code on buffer fills, because the call is relative to the + * current pc. + */ +void x86_call( struct x86_function *p, void (*label)()) { emit_1ub(p, 0xe8); - emit_1i(p, label - x86_get_label(p) - 4); + emit_1i(p, cptr(label) - x86_get_label(p) - 4); +} +#else +void x86_call( struct x86_function *p, struct x86_reg reg) +{ + emit_1ub(p, 0xff); + emit_modrm(p, reg, reg); } +#endif + /* michal: * Temporary. As I need immediate operands, and dont want to mess with the codegen, * I load the immediate into general purpose register and use it. */ -void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, GLint imm ) +void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm ) { assert(dst.mod == mod_REG); emit_1ub(p, 0xb8 + dst.idx); @@ -502,6 +539,14 @@ void sse_addss( struct x86_function *p, emit_modrm( p, dst, src ); } +void sse_andnps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x55); + emit_modrm( p, dst, src ); +} + void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ) @@ -510,6 +555,13 @@ void sse_andps( struct x86_function *p, emit_modrm( p, dst, src ); } +void sse_rsqrtps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x52); + emit_modrm( p, dst, src ); +} void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, @@ -538,6 +590,21 @@ void sse_movlhps( struct x86_function *p, emit_modrm( p, dst, src ); } +void sse_orps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x56); + emit_modrm( p, dst, src ); +} + +void sse_xorps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x57); + emit_modrm( p, dst, src ); +} void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, @@ -559,7 +626,7 @@ void sse_cvtps2pi( struct x86_function *p, void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, - GLubyte shuf) + unsigned char shuf) { emit_2ub(p, X86_TWOB, 0xC6); emit_modrm(p, dest, arg0); @@ -569,13 +636,21 @@ void sse_shufps( struct x86_function *p, void sse_cmpps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, - GLubyte cc) + unsigned char cc) { emit_2ub(p, X86_TWOB, 0xC2); emit_modrm(p, dest, arg0); emit_1ub(p, cc); } +void sse_pmovmskb( struct x86_function *p, + struct x86_reg dest, + struct x86_reg src) +{ + emit_3ub(p, 0x66, X86_TWOB, 0xD7); + emit_modrm(p, dest, src); +} + /*********************************************************************** * SSE2 instructions */ @@ -586,13 +661,21 @@ void sse_cmpps( struct x86_function *p, void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, - GLubyte shuf) + unsigned char shuf) { emit_3ub(p, 0x66, X86_TWOB, 0x70); emit_modrm(p, dest, arg0); emit_1ub(p, shuf); } +void sse2_cvttps2dq( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_3ub( p, 0xF3, X86_TWOB, 0x5B ); + emit_modrm( p, dst, src ); +} + void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src ) @@ -625,6 +708,14 @@ void sse2_packuswb( struct x86_function *p, emit_modrm( p, dst, src ); } +void sse2_rcpps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x53); + emit_modrm( p, dst, src ); +} + void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ) @@ -712,11 +803,11 @@ void x87_fclex( struct x86_function *p ) static void x87_arith_op( struct x86_function *p, struct x86_reg dst, struct x86_reg arg, - GLubyte dst0ub0, - GLubyte dst0ub1, - GLubyte arg0ub0, - GLubyte arg0ub1, - GLubyte argmem_noreg) + unsigned char dst0ub0, + unsigned char dst0ub1, + unsigned char arg0ub0, + unsigned char arg0ub1, + unsigned char argmem_noreg) { assert(dst.file == file_x87); @@ -729,7 +820,7 @@ static void x87_arith_op( struct x86_function *p, struct x86_reg dst, struct x86 assert(0); } else if (dst.idx == 0) { - assert(arg.file = file_REG32); + assert(arg.file == file_REG32); emit_1ub(p, 0xd8); emit_modrm_noreg(p, argmem_noreg, arg); } @@ -1056,44 +1147,41 @@ void mmx_movq( struct x86_function *p, * account any push/pop activity: */ struct x86_reg x86_fn_arg( struct x86_function *p, - GLuint arg ) + unsigned arg ) { return x86_make_disp(x86_make_reg(file_REG32, reg_SP), p->stack_offset + arg * 4); /* ??? */ } -/** - * Initialize an x86_function object, allocating space for up to - * 'code_size' bytes of code. - */ -GLboolean x86_init_func( struct x86_function *p, GLuint code_size ) +void x86_init_func( struct x86_function *p ) { - assert(!p->store); + p->size = 0; + p->store = NULL; + p->csr = p->store; +} + +void x86_init_func_size( struct x86_function *p, unsigned code_size ) +{ + p->size = code_size; p->store = _mesa_exec_malloc(code_size); - if (p->store) { - p->csr = p->store; - return GL_TRUE; - } - else { - p->csr = NULL; - return GL_FALSE; - } + p->csr = p->store; } void x86_release_func( struct x86_function *p ) { - if (p->store) - _mesa_exec_free(p->store); - p->store = p->csr = NULL; + _mesa_exec_free(p->store); + p->store = NULL; + p->csr = NULL; + p->size = 0; } void (*x86_get_func( struct x86_function *p ))(void) { - if (DISASSEM) + if (DISASSEM && p->store) _mesa_printf("disassemble %p %p\n", p->store, p->csr); - return (void (*)(void))p->store; + return (void (*)(void)) (unsigned long) p->store; } #else diff --git a/src/mesa/x86/rtasm/x86sse.h b/src/mesa/x86/rtasm/x86sse.h index 42b09937bca..c2aa416492e 100644 --- a/src/mesa/x86/rtasm/x86sse.h +++ b/src/mesa/x86/rtasm/x86sse.h @@ -2,26 +2,25 @@ #ifndef _X86SSE_H_ #define _X86SSE_H_ -#if defined(USE_X86_ASM) || defined(SLANG_X86) - -#include "glheader.h" +#if defined(__i386__) || defined(__386__) /* It is up to the caller to ensure that instructions issued are * suitable for the host cpu. There are no checks made in this module * for mmx/sse/sse2 support on the cpu. */ struct x86_reg { - GLuint file:3; - GLuint idx:3; - GLuint mod:2; /* mod_REG if this is just a register */ - GLint disp:24; /* only +/- 23bits of offset - should be enough... */ + unsigned file:3; + unsigned idx:3; + unsigned mod:2; /* mod_REG if this is just a register */ + int disp:24; /* only +/- 23bits of offset - should be enough... */ }; struct x86_function { - GLubyte *store; - GLubyte *csr; - GLuint stack_offset; - GLint need_emms; + unsigned size; + unsigned char *store; + unsigned char *csr; + unsigned stack_offset; + int need_emms; const char *fn; }; @@ -80,7 +79,8 @@ enum sse_cc { */ -GLboolean x86_init_func( struct x86_function *p, GLuint code_size ); +void x86_init_func( struct x86_function *p ); +void x86_init_func_size( struct x86_function *p, unsigned code_size ); void x86_release_func( struct x86_function *p ); void (*x86_get_func( struct x86_function *p ))( void ); @@ -92,7 +92,7 @@ struct x86_reg x86_make_reg( enum x86_reg_file file, enum x86_reg_name idx ); struct x86_reg x86_make_disp( struct x86_reg reg, - GLint disp ); + int disp ); struct x86_reg x86_deref( struct x86_reg reg ); @@ -101,31 +101,32 @@ struct x86_reg x86_get_base_reg( struct x86_reg reg ); /* Labels, jumps and fixup: */ -GLubyte *x86_get_label( struct x86_function *p ); +unsigned char *x86_get_label( struct x86_function *p ); void x86_jcc( struct x86_function *p, enum x86_cc cc, - GLubyte *label ); + unsigned char *label ); -GLubyte *x86_jcc_forward( struct x86_function *p, +unsigned char *x86_jcc_forward( struct x86_function *p, enum x86_cc cc ); -GLubyte *x86_jmp_forward( struct x86_function *p); +unsigned char *x86_jmp_forward( struct x86_function *p); -GLubyte *x86_call_forward( struct x86_function *p); +unsigned char *x86_call_forward( struct x86_function *p); void x86_fixup_fwd_jump( struct x86_function *p, - GLubyte *fixup ); + unsigned char *fixup ); -void x86_jmp( struct x86_function *p, GLubyte *label ); +void x86_jmp( struct x86_function *p, unsigned char *label ); -void x86_call( struct x86_function *p, GLubyte *label ); +/* void x86_call( struct x86_function *p, void (*label)() ); */ +void x86_call( struct x86_function *p, struct x86_reg reg); /* michal: * Temporary. As I need immediate operands, and dont want to mess with the codegen, * I load the immediate into general purpose register and use it. */ -void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, GLint imm ); +void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm ); /* Macro for sse_shufps() and sse2_pshufd(): @@ -141,19 +142,24 @@ void mmx_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg sr void mmx_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse2_cvttps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse2_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse2_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse2_packsswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse2_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); -void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, GLubyte shuf ); +void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, + unsigned char shuf ); +void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_divss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_andnps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); -void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src, GLubyte cc ); +void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src, + unsigned char cc ); void sse_maxps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_maxss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_minps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); @@ -166,9 +172,14 @@ void sse_movss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ) void sse_movups( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_mulss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_orps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_xorps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_subps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_rsqrtps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); -void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, GLubyte shuf ); +void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, + unsigned char shuf ); +void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src ); void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); @@ -239,7 +250,7 @@ void x87_fucom( struct x86_function *p, struct x86_reg arg ); * account any push/pop activity. Note - doesn't track explict * manipulation of ESP by other instructions. */ -struct x86_reg x86_fn_arg( struct x86_function *p, GLuint arg ); +struct x86_reg x86_fn_arg( struct x86_function *p, unsigned arg ); #endif #endif |