From 9b3bf392e1af72d29afa0804260cac4d8ffe24e1 Mon Sep 17 00:00:00 2001 From: Andre Maasikas Date: Thu, 29 Jul 2010 15:18:19 +0300 Subject: r600: since 8744c36e added asserts - use another random register for shader with no output --- src/mesa/drivers/dri/r600/r700_assembler.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/mesa/drivers/dri/r600/r700_assembler.c') diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index 99a33df4fc..8f6cc1d875 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -6473,7 +6473,7 @@ GLboolean Process_Fragment_Exports(r700_AssemblerBase *pR700AsmCode, * results are undefined anyway */ if(export_count == 0) { - Process_Export(pR700AsmCode, SQ_EXPORT_PIXEL, 0, 1, 0, GL_FALSE); + Process_Export(pR700AsmCode, SQ_EXPORT_PIXEL, 0, 1, pR700AsmCode->starting_export_register_number, GL_FALSE); } if(pR700AsmCode->cf_last_export_ptr != NULL) -- cgit v1.2.3 From d6a5f94ea4d03b05c434fcad125d1f9c50c638e8 Mon Sep 17 00:00:00 2001 From: Andre Maasikas Date: Mon, 2 Aug 2010 15:11:22 +0300 Subject: r600: fix sin,cos functions on r600 r600 doesnt need the same normalization as r700 - instead it requires range to be truncated to -pi..pi I left the range trunc also effective on r700 althouch according the docs it has sufficent range (-512*PI, +512*PI). The instructions seem to be used not too often to cause perf loss because of this Based on patches and testing by Conn Clark and Alain Perrot --- src/mesa/drivers/dri/r600/r700_assembler.c | 142 +++++++++++++++++++++++++++-- 1 file changed, 133 insertions(+), 9 deletions(-) (limited to 'src/mesa/drivers/dri/r600/r700_assembler.c') diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index 8f6cc1d875..b555ea683c 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -2872,25 +2872,92 @@ GLboolean assemble_CMP(r700_AssemblerBase *pAsm) GLboolean assemble_TRIG(r700_AssemblerBase *pAsm, BITS opcode) { + /* + * r600 - trunc to -PI..PI range + * r700 - normalize by dividing by 2PI + * see fdo bug 27901 + */ + int tmp; checkop1(pAsm); tmp = gethelpr(pAsm); - pAsm->D.dst.opcode = SQ_OP2_INST_MUL; + pAsm->D.dst.opcode = SQ_OP3_INST_MULADD; + pAsm->D.dst.op3 = 1; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); pAsm->D.dst.rtype = DST_REG_TEMPORARY; pAsm->D.dst.reg = tmp; - pAsm->D.dst.writex = 1; assemble_src(pAsm, 0, -1); pAsm->S[1].src.rtype = SRC_REC_LITERAL; setswizzle_PVSSRC(&(pAsm->S[1].src), SQ_SEL_X); + + pAsm->S[2].src.rtype = SRC_REC_LITERAL; + setswizzle_PVSSRC(&(pAsm->S[2].src), SQ_SEL_Y); + pAsm->D2.dst2.literal_slots = 1; pAsm->C[0].f = 1/(3.1415926535 * 2); - pAsm->C[1].f = 0.0F; - next_ins(pAsm); + pAsm->C[1].f = 0.5f; + + if ( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + pAsm->D.dst.opcode = SQ_OP2_INST_FRACT; + + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + pAsm->D.dst.writex = 1; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp; + setswizzle_PVSSRC(&(pAsm->S[0].src), SQ_SEL_X); + + if(( GL_FALSE == next_ins(pAsm) )) + { + return GL_FALSE; + } + pAsm->D.dst.opcode = SQ_OP3_INST_MULADD; + pAsm->D.dst.op3 = 1; + + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp; + setswizzle_PVSSRC(&(pAsm->S[0].src), SQ_SEL_X); + + pAsm->S[1].src.rtype = SRC_REC_LITERAL; + setswizzle_PVSSRC(&(pAsm->S[1].src), SQ_SEL_X); + + pAsm->S[2].src.rtype = SRC_REC_LITERAL; + setswizzle_PVSSRC(&(pAsm->S[2].src), SQ_SEL_Y); + + pAsm->D2.dst2.literal_slots = 1; + + if (pAsm->bR6xx) + { + pAsm->C[0].f = 3.1415926535897f * 2.0f; + pAsm->C[1].f = -3.1415926535897f; + } + else + { + pAsm->C[0].f = 1.0f; + pAsm->C[1].f = -0.5f; + } + + if(( GL_FALSE == next_ins(pAsm) )) + { + return GL_FALSE; + } pAsm->D.dst.opcode = opcode; pAsm->D.dst.math = 1; @@ -4030,22 +4097,79 @@ GLboolean assemble_SCS(r700_AssemblerBase *pAsm) checkop1(pAsm); tmp = gethelpr(pAsm); - /* tmp.x = src /2*PI */ - pAsm->D.dst.opcode = SQ_OP2_INST_MUL; + + pAsm->D.dst.opcode = SQ_OP3_INST_MULADD; + pAsm->D.dst.op3 = 1; + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); pAsm->D.dst.rtype = DST_REG_TEMPORARY; pAsm->D.dst.reg = tmp; - pAsm->D.dst.writex = 1; assemble_src(pAsm, 0, -1); pAsm->S[1].src.rtype = SRC_REC_LITERAL; setswizzle_PVSSRC(&(pAsm->S[1].src), SQ_SEL_X); + + pAsm->S[2].src.rtype = SRC_REC_LITERAL; + setswizzle_PVSSRC(&(pAsm->S[2].src), SQ_SEL_Y); + pAsm->D2.dst2.literal_slots = 1; pAsm->C[0].f = 1/(3.1415926535 * 2); - pAsm->C[1].f = 0.0F; + pAsm->C[1].f = 0.5F; - next_ins(pAsm); + if ( GL_FALSE == next_ins(pAsm) ) + { + return GL_FALSE; + } + + pAsm->D.dst.opcode = SQ_OP2_INST_FRACT; + + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + pAsm->D.dst.writex = 1; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp; + setswizzle_PVSSRC(&(pAsm->S[0].src), SQ_SEL_X); + + if(( GL_FALSE == next_ins(pAsm) )) + { + return GL_FALSE; + } + pAsm->D.dst.opcode = SQ_OP3_INST_MULADD; + pAsm->D.dst.op3 = 1; + + setaddrmode_PVSDST(&(pAsm->D.dst), ADDR_ABSOLUTE); + pAsm->D.dst.rtype = DST_REG_TEMPORARY; + pAsm->D.dst.reg = tmp; + + setaddrmode_PVSSRC(&(pAsm->S[0].src), ADDR_ABSOLUTE); + pAsm->S[0].src.rtype = SRC_REG_TEMPORARY; + pAsm->S[0].src.reg = tmp; + setswizzle_PVSSRC(&(pAsm->S[0].src), SQ_SEL_X); + + pAsm->S[1].src.rtype = SRC_REC_LITERAL; + setswizzle_PVSSRC(&(pAsm->S[1].src), SQ_SEL_X); + + pAsm->S[2].src.rtype = SRC_REC_LITERAL; + setswizzle_PVSSRC(&(pAsm->S[2].src), SQ_SEL_Y); + + pAsm->D2.dst2.literal_slots = 1; + + if(pAsm->bR6xx) { + pAsm->C[0].f = 3.1415926535897f * 2.0f; + pAsm->C[1].f = -3.1415926535897f; + } else { + pAsm->C[0].f = 1.0f; + pAsm->C[1].f = -0.5f; + } + + if(( GL_FALSE == next_ins(pAsm) )) + { + return GL_FALSE; + } // COS dst.x, a.x pAsm->D.dst.opcode = SQ_OP2_INST_COS; -- cgit v1.2.3 From 8ad5b76d52f1c009f48ea90556633e497b40ba87 Mon Sep 17 00:00:00 2001 From: Andre Maasikas Date: Wed, 4 Aug 2010 12:49:39 +0300 Subject: r600: relax stride/alignment requirements for vertices seems hw can do unaligned accesses and unaligned strides removes extra conversion when using vbo's however I needed to switch 3 component byte format to 4 component formats for tests to pass. Somewhat sililar to GL_SHORT fix done earlier removes assert and gains +2 piglit especially draw-vertices --- src/mesa/drivers/dri/r600/r700_assembler.c | 5 ++++- src/mesa/drivers/dri/r600/r700_render.c | 23 +++++++---------------- src/mesa/drivers/dri/r600/r700_vertprog.c | 6 +++--- 3 files changed, 14 insertions(+), 20 deletions(-) (limited to 'src/mesa/drivers/dri/r600/r700_assembler.c') diff --git a/src/mesa/drivers/dri/r600/r700_assembler.c b/src/mesa/drivers/dri/r600/r700_assembler.c index b555ea683c..9c954cbf70 100644 --- a/src/mesa/drivers/dri/r600/r700_assembler.c +++ b/src/mesa/drivers/dri/r600/r700_assembler.c @@ -275,7 +275,10 @@ GLuint GetSurfaceFormat(GLenum eType, GLuint nChannels, GLuint * pClient_size) case 2: format = FMT_8_8; break; case 3: - format = FMT_8_8_8; break; + /* for some (small/unaligned) strides using 4 comps works + * better, probably same as GL_SHORT below + * test piglit/draw-vertices */ + format = FMT_8_8_8_8; break; case 4: format = FMT_8_8_8_8; break; default: diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c index 316a0943f4..ba55f38e05 100644 --- a/src/mesa/drivers/dri/r600/r700_render.c +++ b/src/mesa/drivers/dri/r600/r700_render.c @@ -625,11 +625,11 @@ static void r700SetupStreams(GLcontext *ctx, const struct gl_client_array *input stride = (input[i]->StrideB == 0) ? getTypeSize(input[i]->Type) * input[i]->Size : input[i]->StrideB; - if (input[i]->Type == GL_DOUBLE || input[i]->Type == GL_UNSIGNED_INT || input[i]->Type == GL_INT || + if (input[i]->Type == GL_DOUBLE || input[i]->Type == GL_UNSIGNED_INT || input[i]->Type == GL_INT #if MESA_BIG_ENDIAN - getTypeSize(input[i]->Type) != 4 || + || getTypeSize(input[i]->Type) != 4 #endif - stride < 4) + ) { r700ConvertAttrib(ctx, count, input[i], &context->stream_desc[index]); } @@ -637,19 +637,10 @@ static void r700SetupStreams(GLcontext *ctx, const struct gl_client_array *input { if (input[i]->BufferObj->Name) { - if (stride % 4 != 0) - { - assert(((intptr_t) input[i]->Ptr) % input[i]->StrideB == 0); - r700AlignDataToDword(ctx, input[i], count, &context->stream_desc[index]); - context->stream_desc[index].is_named_bo = GL_FALSE; - } - else - { - context->stream_desc[index].stride = input[i]->StrideB; - context->stream_desc[index].bo_offset = (intptr_t) input[i]->Ptr; - context->stream_desc[index].bo = get_radeon_buffer_object(input[i]->BufferObj)->bo; - context->stream_desc[index].is_named_bo = GL_TRUE; - } + context->stream_desc[index].stride = input[i]->StrideB; + context->stream_desc[index].bo_offset = (intptr_t) input[i]->Ptr; + context->stream_desc[index].bo = get_radeon_buffer_object(input[i]->BufferObj)->bo; + context->stream_desc[index].is_named_bo = GL_TRUE; } else { diff --git a/src/mesa/drivers/dri/r600/r700_vertprog.c b/src/mesa/drivers/dri/r600/r700_vertprog.c index 137f3007ce..6a2a09eaf1 100644 --- a/src/mesa/drivers/dri/r600/r700_vertprog.c +++ b/src/mesa/drivers/dri/r600/r700_vertprog.c @@ -461,11 +461,11 @@ static void r700TranslateAttrib(GLcontext *ctx, GLuint unLoc, int count, const s stride = (input->StrideB == 0) ? getTypeSize(input->Type) * input->Size : input->StrideB; - if (input->Type == GL_DOUBLE || input->Type == GL_UNSIGNED_INT || input->Type == GL_INT || + if (input->Type == GL_DOUBLE || input->Type == GL_UNSIGNED_INT || input->Type == GL_INT #if MESA_BIG_ENDIAN - getTypeSize(input->Type) != 4 || + || getTypeSize(input->Type) != 4 #endif - stride < 4) + ) { pStreamDesc->type = GL_FLOAT; -- cgit v1.2.3