[PATCH vkd3d v2] vkd3d-shader/hlsl: Explicitly specify the swizzle type for sm4 source registers.
Some register types do not use a consistent swizzle type, so the sm4_swizzle_type() function is removed.
The swizzle type now must be specified using the swizzle_type field.
Signed-off-by: Francisco Casas fcasas@codeweavers.com --- libs/vkd3d-shader/hlsl_sm4.c | 69 ++++++++++++++++-------------------- 1 file changed, 31 insertions(+), 38 deletions(-)
diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c index e597425a..2ec8d7d9 100644 --- a/libs/vkd3d-shader/hlsl_sm4.c +++ b/libs/vkd3d-shader/hlsl_sm4.c @@ -748,6 +748,7 @@ struct sm4_instruction struct { struct sm4_register reg; + enum vkd3d_sm4_swizzle_type swizzle_type; unsigned int swizzle; } srcs[2]; unsigned int src_count; @@ -756,27 +757,9 @@ struct sm4_instruction unsigned int idx_count; };
-static unsigned int sm4_swizzle_type(enum vkd3d_sm4_register_type type) -{ - switch (type) - { - case VKD3D_SM4_RT_IMMCONST: - return VKD3D_SM4_SWIZZLE_NONE; - - case VKD3D_SM4_RT_CONSTBUFFER: - case VKD3D_SM4_RT_INPUT: - case VKD3D_SM4_RT_RESOURCE: - case VKD3D_SM4_RT_TEMP: - return VKD3D_SM4_SWIZZLE_VEC4; - - default: - FIXME("Unhandled register type %#x.\n", type); - return VKD3D_SM4_SWIZZLE_VEC4; - } -} - static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *reg, - unsigned int *writemask, const struct hlsl_deref *deref, const struct hlsl_type *data_type) + unsigned int *writemask, enum vkd3d_sm4_swizzle_type *swizzle_type, + const struct hlsl_deref *deref, const struct hlsl_type *data_type) { const struct hlsl_ir_var *var = deref->var;
@@ -797,6 +780,7 @@ static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *r assert(data_type->type <= HLSL_CLASS_VECTOR); reg->type = VKD3D_SM4_RT_CONSTBUFFER; reg->dim = VKD3D_SM4_DIMENSION_VEC4; + if(swizzle_type) *swizzle_type = VKD3D_SM4_SWIZZLE_VEC4; reg->idx[0] = var->buffer->reg.id; reg->idx[1] = offset / 4; reg->idx_count = 2; @@ -825,6 +809,7 @@ static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *r assert(hlsl_reg.allocated); reg->type = VKD3D_SM4_RT_INPUT; reg->dim = VKD3D_SM4_DIMENSION_VEC4; + if(swizzle_type) *swizzle_type = VKD3D_SM4_SWIZZLE_VEC4; reg->idx[0] = hlsl_reg.id; reg->idx_count = 1; *writemask = hlsl_reg.writemask; @@ -867,17 +852,20 @@ static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *r assert(hlsl_reg.allocated); reg->type = VKD3D_SM4_RT_TEMP; reg->dim = VKD3D_SM4_DIMENSION_VEC4; + if(swizzle_type) *swizzle_type = VKD3D_SM4_SWIZZLE_VEC4; reg->idx[0] = hlsl_reg.id; reg->idx_count = 1; *writemask = hlsl_reg.writemask; } }
-static void sm4_register_from_node(struct sm4_register *reg, unsigned int *writemask, const struct hlsl_ir_node *instr) +static void sm4_register_from_node(struct sm4_register *reg, unsigned int *writemask, + enum vkd3d_sm4_swizzle_type *swizzle_type, const struct hlsl_ir_node *instr) { assert(instr->reg.allocated); reg->type = VKD3D_SM4_RT_TEMP; reg->dim = VKD3D_SM4_DIMENSION_VEC4; + if(swizzle_type) *swizzle_type = VKD3D_SM4_SWIZZLE_VEC4; reg->idx[0] = instr->reg.id; reg->idx_count = 1; *writemask = instr->reg.writemask; @@ -929,7 +917,7 @@ static void write_sm4_instruction(struct vkd3d_bytecode_buffer *buffer, const st for (i = 0; i < instr->src_count; ++i) { token = sm4_encode_register(&instr->srcs[i].reg); - token |= sm4_swizzle_type(instr->srcs[i].reg.type) << VKD3D_SM4_SWIZZLE_TYPE_SHIFT; + token |= (uint32_t)instr->srcs[i].swizzle_type << VKD3D_SM4_SWIZZLE_TYPE_SHIFT; token |= instr->srcs[i].swizzle << VKD3D_SM4_SWIZZLE_SHIFT; if (instr->srcs[i].reg.mod) token |= VKD3D_SM4_EXTENDED_OPERAND; @@ -968,6 +956,7 @@ static void write_sm4_dcl_constant_buffer(struct vkd3d_bytecode_buffer *buffer, .srcs[0].reg.type = VKD3D_SM4_RT_CONSTBUFFER, .srcs[0].reg.idx = {cbuffer->reg.id, (cbuffer->used_size + 3) / 4}, .srcs[0].reg.idx_count = 2, + .srcs[0].swizzle_type = VKD3D_SM4_SWIZZLE_VEC4, .srcs[0].swizzle = HLSL_SWIZZLE(X, Y, Z, W), .src_count = 1, }; @@ -1129,10 +1118,10 @@ static void write_sm4_unary_op(struct vkd3d_bytecode_buffer *buffer, enum vkd3d_ memset(&instr, 0, sizeof(instr)); instr.opcode = opcode;
- sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, dst); + sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, NULL, dst); instr.dst_count = 1;
- sm4_register_from_node(&instr.srcs[0].reg, &writemask, src); + sm4_register_from_node(&instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, src); instr.srcs[0].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask); instr.srcs[0].reg.mod = src_mod; instr.src_count = 1; @@ -1149,12 +1138,12 @@ static void write_sm4_binary_op(struct vkd3d_bytecode_buffer *buffer, enum vkd3d memset(&instr, 0, sizeof(instr)); instr.opcode = opcode;
- sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, dst); + sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, NULL, dst); instr.dst_count = 1;
- sm4_register_from_node(&instr.srcs[0].reg, &writemask, src1); + sm4_register_from_node(&instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, src1); instr.srcs[0].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask); - sm4_register_from_node(&instr.srcs[1].reg, &writemask, src2); + sm4_register_from_node(&instr.srcs[1].reg, &writemask, &instr.srcs[1].swizzle_type, src2); instr.srcs[1].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask); instr.src_count = 2;
@@ -1171,11 +1160,13 @@ static void write_sm4_constant(struct hlsl_ctx *ctx, memset(&instr, 0, sizeof(instr)); instr.opcode = VKD3D_SM4_OP_MOV;
- sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, &constant->node); + sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, NULL, &constant->node); instr.dst_count = 1;
instr.srcs[0].reg.dim = (dimx > 1) ? VKD3D_SM4_DIMENSION_VEC4 : VKD3D_SM4_DIMENSION_SCALAR; instr.srcs[0].reg.type = VKD3D_SM4_RT_IMMCONST; + instr.srcs[0].swizzle_type = VKD3D_SM4_SWIZZLE_NONE; + for (i = 0; i < dimx; ++i) instr.srcs[0].reg.immconst_uint[i] = constant->value[i].u; instr.src_count = 1, @@ -1193,10 +1184,10 @@ static void write_sm4_ld(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buf memset(&instr, 0, sizeof(instr)); instr.opcode = VKD3D_SM4_OP_LD;
- sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, dst); + sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, NULL, dst); instr.dst_count = 1;
- sm4_register_from_node(&instr.srcs[0].reg, &writemask, coords); + sm4_register_from_node(&instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, coords); instr.srcs[0].swizzle = hlsl_swizzle_from_writemask(writemask);
/* Mipmap level is in the last component in the IR, but needs to be in the W @@ -1219,7 +1210,8 @@ static void write_sm4_ld(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buf assert(0); }
- sm4_register_from_deref(ctx, &instr.srcs[1].reg, &writemask, resource, resource_type); + sm4_register_from_deref(ctx, &instr.srcs[1].reg, &writemask, &instr.srcs[1].swizzle_type, + resource, resource_type); instr.srcs[1].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask);
instr.src_count = 2; @@ -1440,7 +1432,7 @@ static void write_sm4_if(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buf
assert(iff->condition.node->data_type->dimx == 1);
- sm4_register_from_node(&instr.srcs[0].reg, &writemask, iff->condition.node); + sm4_register_from_node(&instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, iff->condition.node); instr.srcs[0].swizzle = hlsl_swizzle_from_writemask(writemask); write_sm4_instruction(buffer, &instr);
@@ -1465,10 +1457,11 @@ static void write_sm4_load(struct hlsl_ctx *ctx, memset(&instr, 0, sizeof(instr)); instr.opcode = VKD3D_SM4_OP_MOV;
- sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, &load->node); + sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, NULL, &load->node); instr.dst_count = 1;
- sm4_register_from_deref(ctx, &instr.srcs[0].reg, &writemask, &load->src, load->node.data_type); + sm4_register_from_deref(ctx, &instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, + &load->src, load->node.data_type); instr.srcs[0].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask); instr.src_count = 1;
@@ -1531,11 +1524,11 @@ static void write_sm4_store(struct hlsl_ctx *ctx, memset(&instr, 0, sizeof(instr)); instr.opcode = VKD3D_SM4_OP_MOV;
- sm4_register_from_deref(ctx, &instr.dsts[0].reg, &writemask, &store->lhs, rhs->data_type); + sm4_register_from_deref(ctx, &instr.dsts[0].reg, &writemask, NULL, &store->lhs, rhs->data_type); instr.dsts[0].writemask = hlsl_combine_writemasks(writemask, store->writemask); instr.dst_count = 1;
- sm4_register_from_node(&instr.srcs[0].reg, &writemask, rhs); + sm4_register_from_node(&instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, rhs); instr.srcs[0].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask); instr.src_count = 1;
@@ -1551,10 +1544,10 @@ static void write_sm4_swizzle(struct hlsl_ctx *ctx, memset(&instr, 0, sizeof(instr)); instr.opcode = VKD3D_SM4_OP_MOV;
- sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, &swizzle->node); + sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, NULL, &swizzle->node); instr.dst_count = 1;
- sm4_register_from_node(&instr.srcs[0].reg, &writemask, swizzle->val.node); + sm4_register_from_node(&instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, swizzle->val.node); instr.srcs[0].swizzle = hlsl_map_swizzle(hlsl_combine_swizzles(hlsl_swizzle_from_writemask(writemask), swizzle->swizzle, swizzle->node.data_type->dimx), instr.dsts[0].writemask); instr.src_count = 1;
On 11/11/21 13:14, Francisco Casas wrote:
Some register types do not use a consistent swizzle type, so the sm4_swizzle_type() function is removed.
The swizzle type now must be specified using the swizzle_type field.
Signed-off-by: Francisco Casas fcasas@codeweavers.com
Sorry for the late review.
This omits the swizzle types for input semantics. Granted, we only have one so far, and due to other constraints we aren't actually capable of emitting it yet, but for the sake of correctness of architecture we should at least still put the path in.
(FWIW, the correct swizzle type for VKD3D_SM4_RT_PRIMID seems to be NONE).
While you're at it, could you please follow the code style used in the rest of vkd3d? Namely, space after the "if" keyword, and put the body on a new line.
-
Francisco Casas -
Zebediah Figura (she/her)